research paper on algorithms

Suggestions or feedback?

MIT News | Massachusetts Institute of Technology

• Machine learning
• Social justice
• Black holes
• Classes and programs

Departments

• Aeronautics and Astronautics
• Brain and Cognitive Sciences
• Architecture
• Political Science
• Mechanical Engineering

Centers, Labs, & Programs

• Abdul Latif Jameel Poverty Action Lab (J-PAL)
• Picower Institute for Learning and Memory
• Lincoln Laboratory
• School of Architecture + Planning
• School of Engineering
• School of Humanities, Arts, and Social Sciences
• Sloan School of Management
• School of Science
• MIT Schwarzman College of Computing

How quickly do algorithms improve?

Press contact :.

Previous image Next image

Algorithms are sort of like a parent to a computer. They tell the computer how to make sense of information so they can, in turn, make something useful out of it.

The more efficient the algorithm, the less work the computer has to do. For all of the technological progress in computing hardware, and the much debated lifespan of Moore’s Law, computer performance is only one side of the picture.

Behind the scenes a second trend is happening: Algorithms are being improved, so in turn less computing power is needed. While algorithmic efficiency may have less of a spotlight, you’d definitely notice if your trusty search engine suddenly became one-tenth as fast, or if moving through big datasets felt like wading through sludge.

This led scientists from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) to ask: How quickly do algorithms improve?  

Existing data on this question were largely anecdotal, consisting of case studies of particular algorithms that were assumed to be representative of the broader scope. Faced with this dearth of evidence, the team set off to crunch data from 57 textbooks and more than 1,110 research papers, to trace the history of when algorithms got better. Some of the research papers directly reported how good new algorithms were, and others needed to be reconstructed by the authors using “pseudocode,” shorthand versions of the algorithm that describe the basic details.

In total, the team looked at 113 “algorithm families,” sets of algorithms solving the same problem that had been highlighted as most important by computer science textbooks. For each of the 113, the team reconstructed its history, tracking each time a new algorithm was proposed for the problem and making special note of those that were more efficient. Ranging in performance and separated by decades, starting from the 1940s to now, the team found an average of eight algorithms per family, of which a couple improved its efficiency. To share this assembled database of knowledge, the team also created Algorithm-Wiki.org.

The scientists charted how quickly these families had improved, focusing on the most-analyzed feature of the algorithms — how fast they could guarantee to solve the problem (in computer speak: “worst-case time complexity”). What emerged was enormous variability, but also important insights on how transformative algorithmic improvement has been for computer science.

For large computing problems, 43 percent of algorithm families had year-on-year improvements that were equal to or larger than the much-touted gains from Moore’s Law. In 14 percent of problems, the improvement to performance from algorithms vastly outpaced those that have come from improved hardware. The gains from algorithm improvement were particularly large for big-data problems, so the importance of those advancements has grown in recent decades.

The single biggest change that the authors observed came when an algorithm family transitioned from exponential to polynomial complexity. The amount of effort it takes to solve an exponential problem is like a person trying to guess a combination on a lock. If you only have a single 10-digit dial, the task is easy. With four dials like a bicycle lock, it’s hard enough that no one steals your bike, but still conceivable that you could try every combination. With 50, it’s almost impossible — it would take too many steps. Problems that have exponential complexity are like that for computers: As they get bigger they quickly outpace the ability of the computer to handle them. Finding a polynomial algorithm often solves that, making it possible to tackle problems in a way that no amount of hardware improvement can.

As rumblings of Moore’s Law coming to an end rapidly permeate global conversations, the researchers say that computing users will increasingly need to turn to areas like algorithms for performance improvements. The team says the findings confirm that historically, the gains from algorithms have been enormous, so the potential is there. But if gains come from algorithms instead of hardware, they’ll look different. Hardware improvement from Moore’s Law happens smoothly over time, and for algorithms the gains come in steps that are usually large but infrequent. 

“This is the first paper to show how fast algorithms are improving across a broad range of examples,” says Neil Thompson, an MIT research scientist at CSAIL and the Sloan School of Management and senior author on the new paper . “Through our analysis, we were able to say how many more tasks could be done using the same amount of computing power after an algorithm improved. As problems increase to billions or trillions of data points, algorithmic improvement becomes substantially more important than hardware improvement. In an era where the environmental footprint of computing is increasingly worrisome, this is a way to improve businesses and other organizations without the downside.”

Thompson wrote the paper alongside MIT visiting student Yash Sherry. The paper is published in the Proceedings of the IEEE . The work was funded by the Tides foundation and the MIT Initiative on the Digital Economy.

Share this news article on:

Related links.

• Neil Thompson
• Yash Sherry
• MIT Initiative on the Digital Economy
• Computer Science and Artificial Intelligence Laboratory (CSAIL)
• Department of Electrical Engineering and Computer Science
• MIT Sloan School of Management

Related Topics

• Technology and society
• Computer science and technology
• Electrical Engineering & Computer Science (eecs)

Related Articles

A universal system for decoding any type of data sent across a network

Software to accelerate R&D

If transistors can’t get smaller, then coders have to get smarter

Previous item Next item

More MIT News

The power of App Inventor: Democratizing possibilities for mobile applications

Read full story →

Using MRI, engineers have found a way to detect light deep in the brain

A better way to control shape-shifting soft robots

From steel engineering to ovarian tumor research

Professor Emeritus David Lanning, nuclear engineer and key contributor to the MIT Reactor, dies at 96

Discovering community and cultural connections

• More news on MIT News homepage →

Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA, USA

• Map (opens in new window)
• Events (opens in new window)
• People (opens in new window)
• Careers (opens in new window)
• Accessibility
• Social Media Hub
• MIT on Facebook
• MIT on YouTube
• MIT on Instagram

Q-Learning Algorithms: A Comprehensive Classification and Applications

Ieee account.

• Change Username/Password
• Update Address

Purchase Details

• Payment Options
• Order History
• View Purchased Documents

Profile Information

• Communications Preferences
• Profession and Education
• Technical Interests
• US & Canada: +1 800 678 4333
• Worldwide: +1 732 981 0060
• Contact & Support
• About IEEE Xplore
• Accessibility
• Terms of Use
• Nondiscrimination Policy
• Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

Help | Advanced Search

Computer Science > Data Structures and Algorithms

Title: teaching algorithm design: a literature review.

Abstract: Algorithm design is a vital skill developed in most undergraduate Computer Science (CS) programs, but few research studies focus on pedagogy related to algorithms coursework. To understand the work that has been done in the area, we present a systematic survey and literature review of CS Education studies. We search for research that is both related to algorithm design and evaluated on undergraduate-level students. Across all papers in the ACM Digital Library prior to August 2023, we only find 94 such papers. We first classify these papers by topic, evaluation metric, evaluation methods, and intervention target. Through our classification, we find a broad sparsity of papers which indicates that many open questions remain about teaching algorithm design, with each algorithm topic only being discussed in between 0 and 10 papers. We also note the need for papers using rigorous research methods, as only 38 out of 88 papers presenting quantitative data use statistical tests, and only 15 out of 45 papers presenting qualitative data use a coding scheme. Only 17 papers report controlled trials. We then synthesize the results of the existing literature to give insights into what the corpus reveals about how we should teach algorithms. Much of the literature explores implementing well-established practices, such as active learning or automated assessment, in the algorithms classroom. However, there are algorithms-specific results as well: a number of papers find that students may under-utilize certain algorithmic design techniques, and studies describe a variety of ways to select algorithms problems that increase student engagement and learning. The results we present, along with the publicly available set of papers collected, provide a detailed representation of the current corpus of CS Education work related to algorithm design and can orient further research in the area.

Submission history

Access paper:.

• HTML (experimental)
• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

Algorithms & optimization

We perform fundamental research in algorithms, markets, optimization, and graph analysis, and use it to deliver solutions to challenges across Google’s business.

About the team

Our team comprises multiple overlapping research groups working on graph mining, large-scale optimization, and market algorithms. We collaborate closely with teams across Google, benefiting Ads, Search, YouTube, Play, Infrastructure, Geo, Social, Image Search, Cloud and more. Along with these collaborations, we perform research related to algorithmic foundations of machine learning, distributed optimization, economics, data mining, and data-driven optimization. Our researchers are involved in both long-term research efforts as well as immediate applications of our technology.

Examples of recent research interests include online ad allocation problems, distributed algorithms for large-scale graph mining, mechanism design for advertising exchanges, and robust and dynamic pricing for ad auctions.

Team focus summaries

Large-scale optimization.

Our mission is to develop large-scale, distributed, and data-driven optimization techniques and use them to improve the efficiency and robustness of infrastructure and machine learning systems at Google. We achieve such goals as increasing throughput and decreasing latency in distributed systems, or improving feature selection and parameter tuning in machine learning. To do this, we apply techniques from areas such as combinatorial optimization, online algorithms, and control theory. Our research is used in critical infrastructure that supports products such as Search and Cloud.

Understanding places

Our mission is to discover all the world’s places and to understand people’s interactions with those places. We accomplish this by using ML to develop deep understanding of user trajectories and actions in the physical world, and we apply that understanding to solve the recurrent hard problems in geolocation data analysis. This research has enabled many of the novel features that appear in Google geo applications such as Maps.

Structured information extraction

Our mission is to extract salient information from templated documents and web pages and then use that information to assist users. We focus our efforts on extracting data such as flight information from email, event data form the web, and product information from the web. This enables many features in products such as Google Now, Search, and Shopping.

Search and information retrieval

Our mission is to conduct research to enable new or more effective search capabilities. This includes developing deeper understanding of correlations between documents and queries; modeling user attention and product satisfaction; developing Q&A models, particularly for the “next billion Internet users”; and, developing effective personal search models even when Google engineers cannot inspect private user input data.

Medical knowledge and learning

Our mission is offer a premier source of high-quality medical information along your entire online health journey. We provide relevant, targeted medical information to users by applying advanced ML on Google Search data. Examples of technologies created by this team include Symptom Search, Allergy Prediction, and other epidemiological applications.

Featured publications

Highlighted work.

Some of our locations

Some of our people

Gagan Aggarwal

• Algorithms and Theory
• Data Mining and Modeling
• Economics and Electronic Commerce

David Applegate

Aaron Archer

• Distributed Systems and Parallel Computing

Ashwinkumar Badanidiyuru Varadaraja

• Machine Intelligence

Mohammadhossein Bateni

Michael Bendersky

• Information Retrieval and the Web

Kshipra Bhawalkar

Edith Cohen

• Machine Learning

Alessandro Epasto

Alejandra Estanislao

Andrei Z. Broder

Jon Feldman

Nadav Golbandi

Jeongwoo Ko

• Natural Language Processing

Marc Najork

Nitish Korula

Kostas Kollias

Silvio Lattanzi

Mohammad Mahdian

Alex Fabrikant

Rich Washington

Andrew Tomkins

• Human-Computer Interaction and Visualization

Vidhya Navalpakkam

• Machine Perception

Bhargav Kanagal

Aranyak Mehta

Guillaume Chatelet

• Hardware and Architecture
• Software Engineering
• Software Systems

Sandeep Tata

• Data Management

Balasubramanian Sivan

Vahab S. Mirrokni

Xuanhui Wang

Renato Paes Leme

Bryan Perozzi

Morteza Zadimoghaddam

Fabien Viger

Tamas Sarlos

James B. Wendt

We're always looking for more talented, passionate people.

From Predictive Algorithms to Automatic Generation of Anomalies

Machine learning algorithms can find predictive signals that researchers fail to notice; yet they are notoriously hard-to-interpret. How can we extract theoretical insights from these black boxes? History provides a clue. Facing a similar problem – how to extract theoretical insights from their intuitions – researchers often turned to “anomalies:” constructed examples that highlight flaws in an existing theory and spur the development of new ones. Canonical examples include the Allais paradox and the Kahneman-Tversky choice experiments for expected utility theory. We suggest anomalies can extract theoretical insights from black box predictive algorithms. We develop procedures to automatically generate anomalies for an existing theory when given a predictive algorithm. We cast anomaly generation as an adversarial game between a theory and a falsifier, the solutions to which are anomalies: instances where the black box algorithm predicts - were we to collect data - we would likely observe violations of the theory. As an illustration, we generate anomalies for expected utility theory using a large, publicly available dataset on real lottery choices. Based on an estimated neural network that predicts lottery choices, our procedures recover known anomalies and discover new ones for expected utility theory. In incentivized experiments, subjects violate expected utility theory on these algorithmically generated anomalies; moreover, the violation rates are similar to observed rates for the Allais paradox and Common ratio effect.

We thank Peter G. Chang for exceptional research assistance. We also thank audiences at Bristol, Georgetown, Harvard, MIT, Stanford, UCLA, UCL, Warwick, Yale, and the NBER Summer Institute Digital Economics and AI session, Nikhil Agarwal, Rohan Alur, Nicolas Barberis, Raf Batista, Alex Imas, Roshni Sahoo, Suproteem Sarkar, Josh Schwartzstein, Andrei Shleifer, Janani Sekar, Cassidy Shubatt, Richard Thaler, Keyon Vafa, and especially our discussant Colin F. Camerer for helpful comments. We are grateful to the Center for Applied Artificial Intelligence at the Booth School of Business for generous funding. All errors are our own. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.

MARC RIS BibTeΧ

Download Citation Data

Working Groups

Conferences, more from nber.

In addition to working papers , the NBER disseminates affiliates’ latest findings through a range of free periodicals — the NBER Reporter , the NBER Digest , the Bulletin on Retirement and Disability , the Bulletin on Health , and the Bulletin on Entrepreneurship  — as well as online conference reports , video lectures , and interviews .

Why are algorithms called algorithms? A brief history of the Persian polymath you’ve likely never heard of

Digital Health Research Fellow, The University of Melbourne

Disclosure statement

Debbie Passey does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

University of Melbourne provides funding as a founding partner of The Conversation AU.

View all partners

Algorithms have become integral to our lives. From social media apps to Netflix, algorithms learn your preferences and prioritise the content you are shown. Google Maps and artificial intelligence are nothing without algorithms.

So, we’ve all heard of them, but where does the word “algorithm” even come from?

Over 1,000 years before the internet and smartphone apps, Persian scientist and polymath Muhammad ibn Mūsā al-Khwārizmī invented the concept of algorithms.

In fact, the word itself comes from the Latinised version of his name, “algorithmi”. And, as you might suspect, it’s also related to algebra.

Largely lost to time

Al-Khwārizmī lived from 780 to 850 CE, during the Islamic Golden Age . He is considered the “ father of algebra ”, and for some, the “ grandfather of computer science ”.

Yet, few details are known about his life. Many of his original works in Arabic have been lost to time.

It is believed al-Khwārizmī was born in the Khwarazm region south of the Aral Sea in present-day Uzbekistan. He lived during the Abbasid Caliphate, which was a time of remarkable scientific progress in the Islamic Empire.

Al-Khwārizmī made important contributions to mathematics, geography, astronomy and trigonometry. To help provide a more accurate world map, he corrected Alexandrian polymath Ptolemy’s classic cartography book, Geographia.

He produced calculations for tracking the movement of the Sun, Moon and planets. He also wrote about trigonometric functions and produced the first table of tangents.

Al-Khwārizmī was a scholar in the House of Wisdom ( Bayt al-Hikmah ) in Baghdad. At this intellectual hub , scholars were translating knowledge from around the world into Arabic, synthesising it to make meaningful progress in a range of disciplines. This included mathematics, a field deeply connected to Islam .

The ‘father of algebra’

Al-Khwārizmī was a polymath and a religious man. His scientific writings started with dedications to Allah and the Prophet Muhammad. And one of the major projects Islamic mathematicians undertook at the House of Wisdom was to develop algebra.

Around 830 CE, Caliph al-Ma’mun encouraged al-Khwārizmī to write a treatise on algebra , Al-Jabr (or The Compendious Book on Calculation by Completion and Balancing). This became his most important work.

At this point, “algebra” had been around for hundreds of years, but al-Khwārizmī was the first to write a definitive book on it. His work was meant to be a practical teaching tool. Its Latin translation was the basis for algebra textbooks in European universities until the 16th century.

In the first part, he introduced the concepts and rules of algebra, and methods for calculating the volumes and areas of shapes. In the second part he provided real-life problems and worked out solutions, such as inheritance cases, the partition of land and calculations for trade.

Al-Khwārizmī didn’t use modern-day mathematical notation with numbers and symbols. Instead, he wrote in simple prose and employed geometric diagrams:

Four roots are equal to twenty, then one root is equal to five, and the square to be formed of it is twenty-five.

In modern-day notation we’d write that like so:

4x = 20, x = 5, x 2 = 25

Grandfather of computer science

Al-Khwārizmī’s mathematical writings introduced the Hindu-Arabic numerals to Western mathematicians. These are the ten symbols we all use today: 1, 2, 3, 4, 5, 6, 7, 8, 9, 0.

The Hindu-Arabic numerals are important to the history of computing because they use the number zero and a base-ten decimal system. Importantly, this is the numeral system that underpins modern computing technology.

Al-Khwārizmī’s art of calculating mathematical problems laid the foundation for the concept of algorithms . He provided the first detailed explanations for using decimal notation to perform the four basic operations (addition, subtraction, multiplication, division) and computing fractions.

This was a more efficient computation method than using the abacus. To solve a mathematical equation, al-Khwārizmī systematically moved through a sequence of steps to find the answer. This is the underlying concept of an algorithm.

Algorism , a Medieval Latin term named after al-Khwārizmī, refers to the rules for performing arithmetic using the Hindu-Arabic numeral system. Translated to Latin, al-Khwārizmī’s book on Hindu numerals was titled Algorithmi de Numero Indorum.

In the early 20th century, the word algorithm came into its current definition and usage: “a procedure for solving a mathematical problem in a finite number of steps; a step-by-step procedure for solving a problem”.

Muhammad ibn Mūsā al-Khwārizmī played a central role in the development of mathematics and computer science as we know them today.

The next time you use any digital technology – from your social media feed to your online bank account to your Spotify app – remember that none of it would be possible without the pioneering work of an ancient Persian polymath.

Correction: This article was amended to correct a quote from al-Khwārizmī’s work.

• Mathematics
• History of computing

Events and Communications Coordinator

Assistant Editor - 1 year cadetship

Executive Dean, Faculty of Health

Lecturer/Senior Lecturer, Earth System Science (School of Science)

Sydney Horizon Educators (Identified)

Ohio State nav bar

The Ohio State University

• BuckeyeLink
• Find People
• Search Ohio State

New machine learning algorithm promises advances in computing

Digital twin models may enhance future autonomous systems.

Systems controlled by next-generation computing algorithms could give rise to better and more efficient machine learning products, a new study suggests. 

Using machine learning tools to create a digital twin, or a virtual copy, of an electronic circuit that exhibits chaotic behavior, researchers found that they were successful at predicting how it would behave and using that information to control it.

Many everyday devices, like thermostats and cruise control, utilize linear controllers – which use simple rules to direct a system to a desired value. Thermostats, for example, employ such rules to determine how much to heat or cool a space based on the difference between the current and desired temperatures.

As a result, advanced devices like self-driving cars and aircraft often rely on machine learning-based controllers, which use intricate networks to learn the optimal control algorithm needed to best operate. However, these algorithms have significant drawbacks, the most demanding of which is that they can be extremely challenging and computationally expensive to implement. 

Now, having access to an efficient digital twin is likely to have a sweeping impact on how scientists develop future autonomous technologies, said Robert Kent, lead author of the study and a graduate student in physics at The Ohio State University. 

“The problem with most machine learning-based controllers is that they use a lot of energy or power and they take a long time to evaluate,” said Kent. “Developing traditional controllers for them has also been difficult because chaotic systems are extremely sensitive to small changes.”

These issues, he said, are critical in situations where milliseconds can make a difference between life and death, such as when self-driving vehicles must decide to brake to prevent an accident.

The study was published recently in Nature Communications.

Compact enough to fit on an inexpensive computer chip capable of balancing on your fingertip and able to run without an internet connection, the team’s digital twin was built to optimize a controller’s efficiency and performance, which researchers found resulted in a reduction of power consumption. It achieves this quite easily, mainly because it was trained using a type of machine learning approach called reservoir computing. 

“The great thing about the machine learning architecture we used is that it’s very good at learning the behavior of systems that evolve in time,” Kent said. “It’s inspired by how connections spark in the human brain.”

Although similarly sized computer chips have been used in devices like smart fridges, according to the study, this novel computing ability makes the new model especially well-equipped to handle dynamic systems such as self-driving vehicles as well as heart monitors, which must be able to quickly adapt to a patient’s heartbeat.   

“Big machine learning models have to consume lots of power to crunch data and come out with the right parameters, whereas our model and training is so extremely simple that you could have systems learning on the fly,” he said. 

To test this theory, researchers directed their model to complete complex control tasks and compared its results to those from previous control techniques. The study revealed that their approach achieved a higher accuracy at the tasks than its linear counterpart and is significantly less computationally complex than a previous machine learning-based controller. 

“The increase in accuracy was pretty significant in some cases,” said Kent. Though the outcome showed that their algorithm does require more energy than a linear controller to operate, this tradeoff means that when it is powered up, the team’s model lasts longer and is considerably more efficient than current machine learning-based controllers on the market. 

“People will find good use out of it just based on how efficient it is,” Kent said. “You can implement it on pretty much any platform and it’s very simple to understand.” The algorithm was recently made available to scientists. 

Outside of inspiring potential advances in engineering, there’s also an equally important economic and environmental incentive for creating more power-friendly algorithms, said Kent. 

As society becomes more dependent on computers and AI for nearly all aspects of daily life, demand for data centers is soaring, leading many experts to worry over digital systems’ enormous power appetite and what future industries will need to do to keep up with it. 

And because building these data centers as well as large-scale computing experiments can generate a large carbon footprint , scientists are looking for ways to curb carbon emissions from this technology. 

To advance their results, future work will likely be steered toward training the model to explore other applications like quantum information processing, Kent said. In the meantime, he expects that these new elements will reach far into the scientific community. 

“Not enough people know about these types of algorithms in the industry and engineering, and one of the big goals of this project is to get more people to learn about them,” said Kent. “This work is a great first step toward reaching that potential.”

This study was supported by the U.S. Air Force’s Office of Scientific Research. Other Ohio State co-authors include Wendson A.S. Barbosa and Daniel J. Gauthier. 

More Ohio State News

Ohio state academic year culminates with student fashion show.

Students at The Ohio State University who are preparing to enter a variety of career fields showcased their creativity with a recent fashion show at the Ohio Union.

Putting sourdough under the microscope

Since sourdough starters are created from wild yeast and bacteria in the flour, it creates a favorable environment for many types of microbes to flourish. There can be more than 20 different species of yeast and 50 different species of bacteria in a sourdough starter. The most robust become the dominant species.

Ohio State President’s Buckeye Accelerator to launch student ventures

The third annual President’s Buckeye Accelerator has awarded $50,000 each to six teams of Ohio State student entrepreneurs to further their startup business ventures. 

Contact: Admissions | Webmaster | Page maintained by University Communications

Request an alternate format of this page | Web Services Status | Nondiscrimination notice

New machine learning algorithm promises advances in computing

Digital twin models may enhance future autonomous systems.

Systems controlled by next-generation computing algorithms could give rise to better and more efficient machine learning products, a new study suggests.

Using machine learning tools to create a digital twin, or a virtual copy, of an electronic circuit that exhibits chaotic behavior, researchers found that they were successful at predicting how it would behave and using that information to control it.

Many everyday devices, like thermostats and cruise control, utilize linear controllers -- which use simple rules to direct a system to a desired value. Thermostats, for example, employ such rules to determine how much to heat or cool a space based on the difference between the current and desired temperatures.

Yet because of how straightforward these algorithms are, they struggle to control systems that display complex behavior, like chaos.

As a result, advanced devices like self-driving cars and aircraft often rely on machine learning-based controllers, which use intricate networks to learn the optimal control algorithm needed to best operate. However, these algorithms have significant drawbacks, the most demanding of which is that they can be extremely challenging and computationally expensive to implement.

Now, having access to an efficient digital twin is likely to have a sweeping impact on how scientists develop future autonomous technologies, said Robert Kent, lead author of the study and a graduate student in physics at The Ohio State University.

"The problem with most machine learning-based controllers is that they use a lot of energy or power and they take a long time to evaluate," said Kent. "Developing traditional controllers for them has also been difficult because chaotic systems are extremely sensitive to small changes."

These issues, he said, are critical in situations where milliseconds can make a difference between life and death, such as when self-driving vehicles must decide to brake to prevent an accident.

The study was published recently in Nature Communications.

Compact enough to fit on an inexpensive computer chip capable of balancing on your fingertip and able to run without an internet connection, the team's digital twin was built to optimize a controller's efficiency and performance, which researchers found resulted in a reduction of power consumption. It achieves this quite easily, mainly because it was trained using a type of machine learning approach called reservoir computing.

"The great thing about the machine learning architecture we used is that it's very good at learning the behavior of systems that evolve in time," Kent said. "It's inspired by how connections spark in the human brain."

Although similarly sized computer chips have been used in devices like smart fridges, according to the study, this novel computing ability makes the new model especially well-equipped to handle dynamic systems such as self-driving vehicles as well as heart monitors, which must be able to quickly adapt to a patient's heartbeat.

"Big machine learning models have to consume lots of power to crunch data and come out with the right parameters, whereas our model and training is so extremely simple that you could have systems learning on the fly," he said.

To test this theory, researchers directed their model to complete complex control tasks and compared its results to those from previous control techniques. The study revealed that their approach achieved a higher accuracy at the tasks than its linear counterpart and is significantly less computationally complex than a previous machine learning-based controller.

"The increase in accuracy was pretty significant in some cases," said Kent. Though the outcome showed that their algorithm does require more energy than a linear controller to operate, this tradeoff means that when it is powered up, the team's model lasts longer and is considerably more efficient than current machine learning-based controllers on the market.

"People will find good use out of it just based on how efficient it is," Kent said. "You can implement it on pretty much any platform and it's very simple to understand." The algorithm was recently made available to scientists.

Outside of inspiring potential advances in engineering, there's also an equally important economic and environmental incentive for creating more power-friendly algorithms, said Kent.

As society becomes more dependent on computers and AI for nearly all aspects of daily life, demand for data centers is soaring, leading many experts to worry over digital systems' enormous power appetite and what future industries will need to do to keep up with it.

And because building these data centers as well as large-scale computing experiments can generate a large carbon footprint, scientists are looking for ways to curb carbon emissions from this technology.

To advance their results, future work will likely be steered toward training the model to explore other applications like quantum information processing, Kent said. In the meantime, he expects that these new elements will reach far into the scientific community.

"Not enough people know about these types of algorithms in the industry and engineering, and one of the big goals of this project is to get more people to learn about them," said Kent. "This work is a great first step toward reaching that potential."

This study was supported by the U.S. Air Force's Office of Scientific Research. Other Ohio State co-authors include Wendson A.S. Barbosa and Daniel J. Gauthier.

• Electronics
• Telecommunications
• Energy Technology
• Computers and Internet
• Neural Interfaces
• Information Technology
• Computer Science
• Artificial intelligence
• Alan Turing
• Computing power everywhere
• Grid computing
• Data mining

Story Source:

Materials provided by Ohio State University . Original written by Tatyana Woodall. Note: Content may be edited for style and length.

Journal Reference :

• Robert M. Kent, Wendson A. S. Barbosa, Daniel J. Gauthier. Controlling chaos using edge computing hardware . Nature Communications , 2024; 15 (1) DOI: 10.1038/s41467-024-48133-3

Cite This Page :

Explore More

• Controlling Shape-Shifting Soft Robots
• Brain Flexibility for a Complex World
• ONe Nova to Rule Them All
• AI Systems Are Skilled at Manipulating Humans
• Planet Glows With Molten Lava
• A Fragment of Human Brain, Mapped
• Symbiosis Solves Long-Standing Marine Mystery
• Surprising Common Ideas in Environmental ...
• 2D All-Organic Perovskites: 2D Electronics
• Generative AI That Imitates Human Motion

Trending Topics

Strange & offbeat.

Article  

• Volume 15, issue 1
• MS, 15, 315–330, 2024
• Related articles

Research on the optimal speed of vehicles passing speed bumps on the highway based on an immune algorithm

Zhiyong yang, ruixiang zhang.

With the advancement of vehicle technology, there is a growing demand for vehicle comfort in addition to the focus on safety and functionality. On certain accident-prone sections of highways, such as entrance and exit ramps, tunnels, and downhill stretches, continuous speed bumps are typically installed to remind vehicles to reduce their speed. However, while enhancing safety, these measures also introduce a degree of discomfort for passengers and drivers alike. Vehicle speed and the type of road speed bump are key factors influencing vehicle comfort. In order to improve the ride comfort, this paper investigates the problem of adaptive speed control for vehicles passing over different types of continuous speed bumps and proposes a method for solving the optimal speed. In this research, a 4-degree-of-freedom vehicle suspension model and a road excitation model are employed to simulate vehicle vibrations. Simulation optimisation is performed using MATLAB in conjunction with an immune algorithm to obtain the optimal vehicle speeds for traversing three types of continuous speed bumps – sinusoidal, rectangular, and trapezoidal – while adhering to specified constraints. The simulation results demonstrate that this optimisation algorithm effectively enhances the ride comfort of vehicles when navigating speed bumps. The algorithm, when applied, reduces vehicle vertical displacement, acceleration, suspension deflection, and tyre load to varying degrees when crossing speed bumps. It also reduces tyre ground clearance to some extent, achieving a balance between comfort and safety. Furthermore, the study identifies the range of comfortable vehicle speeds for traversing these three types of speed bumps, providing valuable insights for selecting the appropriate speed bump design on roads with varying speed limits.

• Article (PDF, 6002 KB)
• Article (6002 KB)
• Full-text XML

Yang, Z., Zhang, R., Guo, Z., Guo, J., and Zhou, Y.: Research on the optimal speed of vehicles passing speed bumps on the highway based on an immune algorithm, Mech. Sci., 15, 315–330, https://doi.org/10.5194/ms-15-315-2024, 2024.

Intelligent driving technology has progressively matured with the rapid development of a new generation of information technology and has developed into a practical application. The driving comfort and safety of the vehicle are crucial as a product for real-world applications. However, the majority of current research on the comfort of vehicles has been carried out with vehicles travelling on flat surfaces, such as in route planning (Motallebi et al., 2020) and suspension optimisation (Gao and Qi, 2021). There are few studies on the comfort and safety of vehicles when passing over uneven surfaces, especially when it comes to passing speed bumps. When drivers encounter speed bumps while driving, they rapidly identify the type of speed bump and adjust their vehicle's speed based on their subjective experience to minimise discomfort during passage (Barreno et al., 2022). In the case of semi-automated driving, forward-facing cameras are used to detect speed bumps on the road ahead, and the driving assistance system assists the driver in adjusting the vehicle's speed to reduce vehicle vibrations (Zein and Darwiche, 2020). However, vehicles lack the capability for subjective evaluation, making it challenging to adjust to an appropriate speed based on comfort requirements when encountering speed bumps. This presents a significant hurdle in achieving the optimal balance between driving comfort and safety.

The discomfort experienced during rides primarily arises from vertical vibrations induced by uneven road surfaces, and vehicle speed also affects the intensity of the vibrations. Therefore, the enhancement of ride comfort predominantly focuses on the suppression of vertical vibrations due to the presence of the driver. The suspension system, as a crucial component ensuring comfort, can absorb vibrations generated by road disturbances and has become an active area of research. To date, numerous studies have been conducted on suspension systems. Researchers such as Yan et al. (2019) have applied H ∞ suspension control to quarter-vehicle active-suspension systems to dampen vibrations, thereby enhancing vehicle handling stability and driver safety; however, considering road incentives to be disturbances has limited the guidance of road information. Mahmoodabadi et al. (2020) proposed a method based on optimal fuzzy adaptive robust proportional–integral–derivative (PID) controllers, aiming to minimise the relative displacement between the vehicle body acceleration and tyre spring mass, thereby improving active-suspension performance and ride comfort. Nguyen and Nguyen (2022a) proposed a state-multivariable-based sliding-mode PID-integrated suspension control method combined with a quarter-dynamics model, significantly reducing spring mass displacement and acceleration and thus enhancing vehicle stability and comfort. Further research introduced the optimised sliding-mode control (OSMC) algorithm to control the operation of active-suspension systems, substantially improving vehicle oscillations on uneven road surfaces while effectively preventing wheel separation, further enhancing ride comfort (Nguyen and Nguyen, 2022b). Additionally, an AFSPIDF active-suspension control algorithm, blending PID, SMC, and various fuzzy algorithms, successfully suppressed vehicle vibrations, significantly reducing vertical body displacement and acceleration and ultimately enhancing vehicle stability and comfort (Nguyen and Nguyen, 2023). However, due to the complexity of vehicle systems, the above researchers mainly utilised a quarter-vehicle model in studying ride comfort and safety. Nonetheless, the quarter-vehicle model can only investigate vehicle motion in the vertical direction, overlooking many crucial vehicle vibration characteristics and thus making it challenging to fully capture vehicle system information (Yang et al., 2015). Moreover, the use of more complex models would increase the difficulty of mathematical modelling and dynamic analysis. Therefore, employing the 4-degree-of-freedom (4-DOF) nonlinear vehicle suspension model is a moderately complex yet closer-to-reality approach, enabling more accurate acquisition of vehicle feedback information and providing precise adjustment solutions for vehicles navigating different speed bumps.

For advanced intelligent vehicles, their sensors and network systems have the capability to acquire information about the road ahead. Consequently, based on this context, researchers have proposed novel approaches. For instance, drawing upon road information obtained from networks and forward sensors, Wu et al. (2020) introduced a comfort optimisation strategy capable of coordinating speed variations and suspension vibrations. This approach utilised a hybrid horizon variance (HV) model predictive control (MPC) method, resulting in enhanced comfort for passive suspension vehicles operating at a fixed speed. Building upon this foundation, Liu et al. (2023) developed an integrated approach based on road information, combining preview active-suspension control and longitudinal velocity planning. This integrated approach employed a road-information-based semi-explicit model predictive controller (SE-MPC) for active-suspension control, further elevating the ride comfort of autonomous vehicles. Huang et al. (2023) and others approached the problem from the perspective of adaptive nonlinear control, introducing a novel suspension control method. Leveraging X-shaped biomimetics inspired by the skeletal structures of animals or insects and utilising road information to solve multi-objective optimisation problems, this method significantly enhanced vehicle comfort.

It is evident that forward road information is paramount for speed adjustment, particularly when considering ride comfort. However, current research has not adequately addressed whether speed planning can achieve optimal comfort when navigating uneven road surfaces, especially over speed bumps. In various road segments, such as tunnel entrances, downhill slopes, and pedestrian areas, speed bumps are commonly deployed to control vehicle speed and to enhance the safety of both vehicles and pedestrians. These speed bumps come in different shapes, including rectangular, semi-sinusoidal, and trapezoidal designs. However, current research concerning vehicle traversal of speed bumps predominantly focuses on trapezoidal speed bumps, with limited exploration of other shapes (Walavalkar et al., 2021; Miracle et al., 2021). In fact, different shapes of speed bumps have different optimal passing speeds, and the study of trapezoidal speed bumps alone is relatively homogeneous and lacks universality. In addition, there are different speed limits at different road locations, and the study of what kind of speed bumps should be laid on different speed-limited sections is relatively limited. Hence, conducting research on the optimal vehicle speeds for driving vehicles to navigate different speed bump shapes holds significant importance. Such research endeavours contribute to enhancing the comfort and safety of vehicles, facilitating their adaptation to the diverse road conditions and speed limit requirements.

When adjusting vehicle speed, it is imperative not only to ascertain the optimal speed for comfort but also to comprehensively consider the influence of other factors on comfort. While the immune algorithm (IA) (Gong et al., 2009) is a multi-objective optimisation method constructed by mimicking the biological immune mechanism and integrating the incentive of gene evolution, it has the advantages of self-adaptation, stochasticity, and population diversity, as well as superior global search capabilities, parallelism, and robustness, which overcomes the phenomenon of prematurity that exists in general optimisation. Significantly, the IA has found successful application in the realm of automotive mechatronic systems. For example, Chen (2020) used the IA to optimise the front- and rear-suspension parameters of an off-road vehicle, which enhanced the suspension performance of the off-road vehicle, thus improving the ride comfort and stability of the whole vehicle. Similarly, Shieh et al. (2014) combined the IA with adaptive fuzzy control and finally developed an integrated adaptive fuzzy controller which was integrated into the vehicle suspension system to achieve a balance between comfort and operability. However, it is noteworthy that these studies predominantly treated vehicle speed as a quantitative parameter, focusing on the enhancement of comfort through suspension system parameter improvements. Regrettably, the pivotal role of vehicle speed as a determinant of driving and riding comfort has often been overlooked.

Based on the aforementioned statements, this paper approaches the topic from a different perspective. Utilising the information obtainable with regard to road speed bumps through networks and sensors, the vehicle speed is considered to be an unknown condition. The research focuses on a 4-degree-of-freedom nonlinear vehicle suspension model. It employs the immune algorithm for optimising the speed of vehicles when traversing various speed bumps. This optimisation aims to achieve the ideal speed for navigating speed bumps, thereby reducing vibrations in vehicles. This approach facilitates adaptive speed adjustment in vehicles, ultimately enhancing both ride comfort and safety. Simultaneously, determining the optimal speed for traversing speed bumps provides valuable guidance for the installation of speed bumps in different speed limit zones.

The structure of this paper is as follows: the first part analyses the current state of research on vehicle suspension systems and comfort; the second part introduces the 4-degree-of-freedom nonlinear vehicle suspension model, as well as the trapezoidal, half-sine, and rectangular-wave speed bump models, and also describes the relevant parameters of the models; the third part outlines a multi-objective optimisation algorithm based on the IA; and the fourth part details the experiments on applications in different speed bump scenarios, and the results are analysed.

2.1  The 4-degree-of-freedom nonlinear vehicle suspension model

Figure 1 depicts the simplified diagram of the 4-DOF nonlinear vehicle suspension model applied in this paper (Yang et al., 2016). The nonlinear suspension model consists mainly of the vehicle body, unsprung masses of the front and rear suspension, springs, front and rear suspensions, tyres, and dampers. It enables the study not only of the vertical and pitch motion of the vehicle body but also of the vertical motion of the front and rear wheels (Yang et al., 2014). Table 1 shows the definitions of the symbols used in the model shown in Fig. 1.

Figure 1 Nonlinear vehicle suspension model of 4-DOF.

Table 1 Symbolic interpretation of the model.

Download Print Version | Download XLSX

From the d'Alembert principle, the system equation of motion can be expressed as in Eq. (1):

In some simplified nonlinear dynamics studies of vehicle suspension, the suspension spring is regarded to be linear; that is, the deformation amount of the spring has a linear relationship with the spring force range; f s = k Δ x holds, where k is the stiffness coefficient, and Δ x is the deformation amount of the spring. However, in actual situations, the suspension spring will only be approximately linear when there is a small deformation, and nonlinear motion characteristics will appear when the deformation is large. The suspension spring will exhibit nonlinear deformation under road excitation. To better fit the actual situation, the nonlinear spring characteristics are expressed as follows (Yang et al., 2016):

where f s represents the dynamic spring force, sgn(⋅) denotes the signum function, Δ s signifies the deformation of the spring, k s represents the stiffness coefficient of the spring, and n denotes the nonlinearity coefficient of the spring. When n ≠1 , the spring exhibits nonlinear characteristics; otherwise, it demonstrates linear characteristics. Consequently, the nonlinear characteristics of each spring in the vehicle can be expressed as follows:

We express the damping force of the nonlinear suspension system as follows (Yang et al., 2016):

where f sc represents the damping force, Δ x ˙ s signifies the relative velocity of the damper, and c s represents the damping coefficient, which exhibits different characteristics when the damper is stretched and compressed. Therefore, the nonlinear damping forces of various vehicle suspension systems are expressed as follows:

where c f1 and c r1 denote the front- and rear-tyre damping coefficients, and c f2 and c r2 denote the front- and rear-tyre damping coefficients. Furthermore,

Let the state variable x 1 = x b , x 2 = x ˙ b , x 3 = θ , x 4 = θ ˙ , x 5 = x f , x 6 = x ˙ f , x 7 = x r , and x 8 = x ˙ r . The equation of state of the 4-DOF nonlinear suspension system is expressed as in Eq. (8):

When the suspension system is in relative static equilibrium, it can be calculated using the following equation:

where Δ sfi i = 1 , 2 and Δ sri i = 1 , 2 denote the static deformations, which can be obtained from Eq. (4); the masses of the spring load mass m b at the front and rear axes, m bf and m br , respectively, are expressed as follows:

2.2  Periodic speed bump excitation model

Accurately obtaining road surface information is crucial for analysing and evaluating vehicles. The uneven road surface between speed bumps also serves as a source of vehicle vibrations. When the vehicle's speed changes, the road surface excitation experienced by the vehicle also varies. In other words, the road surface excitation encountered by the vehicle when passing over speed bumps is influenced jointly by the speed bumps and the uneven road surface. Hence, this study takes into consideration the unevenness excitation that the uneven road surface between speed bumps imposes on vehicles and simulates the road surface's unevenness excitation using a sine wave model. Let x h (t) be the excitation function of the speed bumps, let A be the average amplitude of the uneven road surface, and let f r represent the excitation frequency of the uneven road surface. Then the excitations x fd and x rd of the front and rear wheels of the vehicle are represented as follows:

with Δ t being the time difference between the front and rear wheels of the vehicle through the road point, approximately equal to l f + l r / v .

2.2.1  Periodic trapezoidal speed bumps

Figure 2 shows the excitation model for the periodic trapezoidal speed bumps (Yang et al., 2022); d is the separation between the speed bumps, and w and h represent the width and height of the speed bumps, respectively. The vehicle passes over the speed bumps with speed v , and then the excitation period of the periodic trapezoidal speed bumps is T = w + d / v .

Figure 2 Periodic trapezoidal speed bump excitation model.

The excitation of the wheels in Fig. 2 is represented as in Eq. (5):

Thus, the excitation of the front and rear wheels of the vehicle on the periodic trapezoidal speed bumps is respectively expressed as follows:

2.2.2  Periodic half-sine speed bumps

Figure 3 shows the excitation model for the periodic half-sine speed bumps (Zhang and Zheng, 2022); the speed bump's height is h , its width is w , and the separation between the speed bumps is d . After measurement, the width of the speed bump is roughly equal to the separation; that is, w = d . When the vehicle passes over the speed bump with speed v , the excitation of the front and rear wheels of the vehicle on the periodic half-sine speed bump road surface is defined as follows:

Figure 3 Periodic half-sine speed bump excitation model.

2.2.3  Periodic rectangular-wave speed bumps

Figure 4 shows the excitation model for the periodic rectangular-wave speed bumps (Wu et al., 2014). The height and width of the speed bumps are h and w , respectively, and the separation between the speed bumps is d . The speed over the speed bumps of the vehicle is v , and then the excitation of the periodic rectangular-wave speed bump road surface to the front and rear wheels of the vehicle can be expressed as follows:

where square (⋅) denotes the rectangular-wave function.

Figure 4 Periodic rectangular-wave speed bump excitation model.

2.3  Simulation parameters

The 4-DOF vehicle suspension parameters used in the simulation are shown in Table 2 (Zhu and Ishitobi, 2004). The simulation takes the static equilibrium point x b , x ˙ b , θ , θ ˙ b , x f , x ˙ f , x r , x ˙ r = 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0 in the vertical direction of the vehicle as the initial condition. Due to the nonlinearity of the differential equations, a numerical investigation of the dynamics of the vehicle model was conducted using a fourth-order fixed-step Runge–Kutta algorithm (Yang et al., 2022).

Table 2 Parameters of 4-DOF vehicle suspension.

This chapter aims to discuss the problem of speed adaptive adjustment of vehicles and optimises multiple objectives, such as the speed- and suspension-damping coefficients, simultaneously. Firstly, objective functions are established by combining the optimisation objectives with the vehicle indexes to evaluate the comfort of the optimisation objectives. Then, the optimisation objectives are regarded to be a set of antibodies, and we use the immune algorithm to establish the multi-objective optimisation algorithm for optimisation, which can achieve the purpose of improving the comfort of the vehicle.

3.1  Establish the objective function

The vertical movement of the vehicle body, the dynamic deflection of the front and rear suspension, and the front- and rear-wheel dynamic loads are used as evaluation indexes of the vehicle to assess the overall performance and balance response. In addition, all indexes are combined for a comprehensive assessment of comfort.

The vertical displacement of the body is usually proportional to the vehicle speed when the vehicle passes over speed bumps. Appropriate vertical body vibration displacement can effectively protect the driver and occupants from unevenness excitation of the road surface; thus, this paper takes the vertical body displacement to be the evaluation index of vehicle comfort (Pan and Sun, 2019); combined with Eq. (8), let

The vehicle's suspension deflection directly affects the handling stability of the vehicle. If the dynamic suspension deflection exceeds the design stroke of the vehicle, it will cause damage to the vehicle's suspension components. Therefore, the paper selects the front- and rear-suspension deflection f 2f and f 2r as indicators for assessing the vehicle's smoothness (Sha et al., 2020); these are expressed as follows:

Vibrations generated by the ground excitation on the wheels will cause discomfort to the driver, and the dynamic loads generated by the vibrations will aggravate the wear of the vehicle and even cause damage. Therefore, the dynamic loads f 3f and f 3r of the front and rear wheels are used to evaluate the vehicle's safety when driving on the road (Yu et al., 2019). The stiffness coefficients for the front and rear wheels are expressed in terms of k f1 and k r1 . Moreover, x fd and x rd represent the road excitations to the front and rear wheels; thus, the dynamic loads of the vehicle are expressed, respectively, as follows:

We use the root-mean-square (rms) value to dimensionlessly process the sub-objectives, such as the vertical displacement of the vehicle body, the dynamic deflection of the suspension, and the dynamic load of the wheels, thereby reducing the influence of incidental factors such as data inequality. Based on the influence level of each sub-objective, the linear weighting method is used to sum up the sub-objectives, and the objective function is expressed as follows (Wang et al., 2022):

Due to the varying impact of different influencing factors on the overall comfort of the vehicle, this study takes the relative weights ω 1 , ω 2f , ω 2r , ω 3f , and ω 3r of each sub-objective as an example, setting them at 0.20, 0.15, 0.15, 0.25, and 0.25, respectively, as examples for investigation. This paper simulates the scenario where the vehicle passes over trapezoidal, semi-sinusoidal, and rectangular speed bumps when exiting a highway ramp with a speed limit of 60 km h −1 . The initial speed of the vehicle before optimisation is 40 km h −1 . To ensure the accuracy of the simulation, the duration of vehicle motion in the simulation is set to 120 s (Yang et al., 2022).

3.2  Optimisation process based on the immune algorithm

Establishing the objective function facilitates the evaluation of the optimised comfort, which indicates the degree of optimisation. While this part describes the implementation process of the optimisation objective through the immune algorithm, which is to achieve the practical application of the algorithm by optimising the speed of the vehicle passing over the speed bumps and the relevant parameters, the specific flowchart is shown in Fig. 5.

Figure 5 Basic flow of IA optimisation.

In order to facilitate the acquisition of the vehicle balance response, the relevant vehicle parameters should be initialised to achieve better optimisation of the vehicle's performance. This paper addresses the problem of adaptive control of vehicle speed when passing over speed bumps. It considers a set of antibodies, x = v , C f 2 , C r 2 , where v represents the vehicle speed, and C f2 and C r2 denote the front- and rear-suspension damping coefficients. These antibodies form the initial generation of the antibody population X = x 1 , x 2 , ⋯ , x N using an immune algorithm. Simulations are conducted using the vehicle model to obtain performance feedback, and the objective function (19) is computed accordingly. The affinity of the antibodies is calculated as aff x = 1 / F , where F is the value of the obtained objective function. This affinity assessment is used to evaluate the comfort of the vehicle when passing over speed bumps in conjunction with the reciprocal of the objective function (19) from Sect. 3.1. A higher affinity value implies better comfort when the vehicle traverses speed bumps. While ensuring the algorithm's global search capability, this paper sets the population size N to be 30.

In optimisation problems, it is generally considered to be advantageous to initiate optimisation from the current best data point and to select the nearest data point to the current one for further optimisation. The introduction of heuristic information plays a significant role in aiding the optimisation process. By combining the affinity and concentration of antibodies, the excitation level of each group of antibodies is computed to perform the final assessment of antibody quality. A higher excitation level of antibodies indicates better quality, signifying that the group of antibodies can enhance the comfort of vehicles when traversing speed bumps. The concentration of antibodies is defined as

where N is the population size, and S ( x i , x j ) is the similarity between the antibodies; the incentive of the antibodies is denoted as

The calculation of the excitation degree suppresses the high antibody concentration, which ensures the diversity of the antibody population and ensures that the algorithm can be optimised to obtain the optimal speed and related parameters after optimisation. Optimising the speed of the vehicle has to satisfy a variety of constraints, such as smoothness and safety, in addition to the optimal comfort of the vehicle over the speed bump; however, the optimal solutions obtained so far do not meet the requirements of the various constraints. In order to obtain the optimal vehicle speed and related parameter solutions for the current problem as much as possible, after multiple experiments and verifications, it was found that the affinity of the antibody in the paper had almost converged before the evolution of 200 generations, and good results could be obtained. However, after the number G was set to 200 generations, the affinity of the antibody rarely continued to increase. Therefore, the maximum generation was set to 200.

In each round of the evolutionary process, immune operations play a crucial role in ensuring the continuous improvement of vehicle comfort. Various immune operations from the biological immune response, including immune selection, cloning, mutation, and clone suppression, are employed to enhance the quality of antibody parameters affecting vehicle comfort and to optimise the comfort of vehicle passage over speed bumps. The immune selection operator filters antibodies in the population, activating high-quality antibodies that improve vehicle comfort, with the immune selection ratio set to 50 % of the population (50 % NP). The cloning operator replicates activated parameter antibodies to generate several copies. In order to ensure that changes in antibodies' variations lead to alterations in vehicle comfort, the cloning quantity ( M ) is set at 10. The mutation operator is applied to copies of vehicle speed and suspension damping coefficients, using a real-number-encoding algorithm with a certain probability. This mutation is aimed at modifying the vehicle's comfort while traversing speed bumps, maintaining population diversity, and enhancing local search capabilities, as illustrated below:

where the symbol x i , j , m represents the j th dimension of the m th clone of antibody x i , where δ denotes the defined neighbourhood range, and the mutation rate p m is set to 0.7. The clone suppression operator performs a reselection on the results of mutation, suppressing and eliminating parameter antibodies with low vehicle comfort and ensuring that high-quality antibodies that improve vehicle comfort are retained for the next generation. This ensures that the next generation of antibodies has a rich diversity in terms of vehicle comfort results, maintaining diversity in the new antibody population.

To enhance the global search capability of the optimisation algorithm and to obtain new antibodies with different vehicle comfort levels, a population refresh strategy is employed after each round of evolution. This strategy randomly generates new antibodies and eliminates half of the antibodies with low stimulation levels to ensure that the vehicle comfort of the next generation of antibodies entering the new round of evolution is better than the previous generation. When the specified number of evolution generations is reached in the algorithm, the current best antibody, X best , is output, which includes the optimal vehicle speed and front- and rear-suspension damping coefficients for passing over speed bumps.

This paper details an optimisation algorithm considering vehicle speed v and front- and rear-suspension damping coefficients C f2 and C r2 . Simulating the vehicle passing through the highway exit ramp with a speed limit of 40 km h −1 , combining the reality and hardware conditions, the constraint ranges of each variable are set to be 0 km h −1 < v ≤60  km h −1 , 0 kg s −1   <   C f2 ≤2000  kg s −1 , and 0  kg s −1 < C r2 ≤2000  kg s −1 .

4.1  Application effects of vehicles passing over periodic trapezoidal speed bumps

After applying the multi-objective optimisation algorithm to the trapezoidal speed bumps, the convergence began in the 173rd generation. The affinity of the best antibody was 1.7419, the vehicle speed v was 17.06 km h −1 , the front-suspension damping coefficient C f2 was 1997   kg s −1 , and the rear-suspension damping coefficient C r2 was 1575  kg s −1 .

From Fig. 6, it can be observed that the optimal affinity stabilises and gradually increases after the 54th generation, ultimately converging. Between the 54th and 172nd generations, when both v and C f2 remain stable and mostly unchanged, only variations in C r2 contribute to a slight improvement in the optimal antibody's affinity. After the 173rd generation, due to the presence of mutation operators and population refreshing in the optimisation model, mutations in v , C f2 , and C r2 lead to a sudden change in antibody affinity, yielding antibodies with higher affinities that subsequently stabilise. Afterwards, it becomes increasingly challenging to generate new antibodies that would lead to an improvement in affinity. Affinity has reached a state of near-convergence, indicating that the optimal vehicle speed and related parameters have been obtained.

Figure 6 Iterative process for each variable of vehicle passing over the periodic trapezoidal speed bumps. (a)  Iterative process for optimal affinity. (b)  Iterative process for individual optimal v . (c)  Iterative process for individual optimal C f2 . (d)  Iterative process for individual optimal C r2 .

The response curves of the vehicle suspension system before and after optimisation are plotted in Fig. 7.

Figure 7 Comparison of vehicle response curves over periodic trapezoidal speed bumps before and after optimisation. (a) Response curve for objective function values F . (b) Response curve for vertical body displacement x b . (c) Response curve for front-suspension dynamic deflection d df . (d) Response curve for front-wheel dynamic load  d lf .

Before optimisation, the vehicle speed was 40.00 km h −1 , and after optimisation, it was reduced to 17.06 km h −1 . Observing Fig. 7, it is evident that the optimised objective function value is significantly lower than before optimisation. The suspension deflection response has decreased from 0.0119 to 0.0054 m, and the wheel dynamic load response has been reduced from 2510.5 to 1106.9  N, both lower than their respective values before optimisation. Additionally, the maximum vertical displacement of the vehicle body has decreased from 0.0816 to 0.0771  m, and according to the rms criterion, the average value of the vehicle body displacement has decreased from 0.0105 to 0.0092  m. This indicates the effectiveness of the optimisation algorithm. These improvements reflect that the optimisation has enhanced the comfort and safety of the vehicle when passing over speed bumps. The spring mass acceleration curves before and after optimisation are shown in Fig. 8.

Figure 8 Curves of acceleration a before and after optimisation over trapezoidal speed bumps.

Figure 8 shows the acceleration curves before and after optimisation when passing over the trapezoidal speed bump. Prior to optimisation, the maximum acceleration value for the spring mass reached 0.9088 m s −2 , which was reduced to 0.7264 m s −2 after optimisation. The average values were 0.0597 m s −2 before optimisation and 0.0557 m s −2 after optimisation. These findings further confirm the improvement in comfort after optimisation.

4.2  Application effects of vehicles passing over periodic half-sine speed bumps

For the half-sine speed bumps, after 200 generations of optimisation, the affinity of the best antibody converges to 1.4321 when the vehicle speed v takes 15.33 km h −1 , the front-suspension damping coefficient C f2 takes 1971 kg s −1 , and the rear-suspension damping coefficient C r2 takes 1819 kg s −1 .

From Fig. 9, it can be observed that, between the 51st and 123rd generations, the affinity of the best antibodies remains essentially stable, with minimal variation. During this evolutionary phase, changes in speed v are relatively small. However, there was a notable exchange in the values of the front and rear damping coefficients ( C r2 and C r2 ) at the 97th generation due to a mutation, which essentially corresponds to an interchange of front- and rear-wheel damping coefficients, resulting in relatively minor effects on affinity. Between the 130th and 140th generations, vehicle speed gradually starts to converge towards its optimum value. During this period, as the front and rear damping coefficients need to converge continuously around the optimal values, they start to oscillate around this optimum. Consequently, the affinity of the antibodies increases rapidly, leading to improved vehicle comfort. This trend continues until approximately the 170th generation, when the affinity of the antibodies stabilises. At this point, there is little room for further improvement, signifying that the optimal vehicle speed for passing over speed bumps has been obtained.

Figure 9 Iterative process for each variable of the vehicle passing over periodic half-sine speed bumps. (a)  Iterative process for optimal affinity. (b)  Iterative process for individual optimal v . (c)  Iterative process for individual optimal C f 2 . (d)  Iterative process for individual optimal C r 2 .

Figure 10 shows the system response curves of the vehicle suspension system at steady state; these are contrasted to the response curves prior to optimisation.

Figure 10 Comparison of vehicle response curves over periodic half-sine speed bumps before and after optimisation. (a)  Response curve for objective function values F . (b)  Response curve for vertical body displacement x b . (c)  Response curve for front-suspension dynamic deflection d df . (d)  Response curve for front-wheel dynamic load d lf .

Before optimisation, the vehicle speed was 40.00 km h −1 , which was reduced to 15.33 km h −1 after optimisation. Observing Fig. 10, it can be noted that the objective function value after optimisation is significantly lower than before, halving the value from the pre-optimisation state. This indicates an overall improvement in comfort. However, there is a slight increase in vertical body displacement, rising from 0.0097 to 0.0109  m, which has a negligible impact. On the other hand, suspension dynamic deflection decreased from 0.0112 to 0.0086  m, and the wheel dynamic load response was reduced from 2208.8 to 1313.6  N. Further analysis reveals that, before optimisation, there was a risk of wheel lift due to excessive speed, which has been significantly mitigated after optimisation. The spring mass acceleration curve at this stage is shown in the Fig. 11.

Figure 11 Curves of acceleration a before and after optimisation over half-sine speed bumps.

The spring mass acceleration has shown a slight increase due to the increase in vertical body displacement. The maximum acceleration value has increased from 0.8828 to 0.9301 m s −2 , and the average acceleration has increased from 0.0593 to 0.0645 m s −2 . While there is a slight sacrifice in comfort when passing over the semi-sinusoidal speed bump, it ensures the safety and stability of the vehicle.

4.3  Application effects of vehicles passing over periodic rectangular-wave speed bumps

After applying the optimisation algorithm to the rectangular-wave speed bumps, the optimal antibody has an affinity of 1.5087 after convergence, the vehicle speed v is 25.40 km h −1 , the front-suspension damping coefficient C f2 is 1623 kg s −1 , and the rear-suspension damping coefficient C r2 is 1617 kg s −1 .

From Fig. 12, it can be observed that the variations in vehicle speed v occur relatively infrequently, and their impact on the optimal affinity of the antibodies is relatively minor. The optimal affinity remains stable between the 160th and 175th generations. Similarly, the rear suspension damping coefficient C r2 converges around the 175th generation. The trends in changes for the front and rear damping coefficients are relatively similar. Upon comparison, it is evident that the overall optimal affinity increases as the absolute difference between the front and rear damping coefficients decreases. Consequently, when passing over speed bumps of rectangular shape, it is necessary to maintain closely matched front and rear damping coefficients. Ultimately, at around the 190th generation, the optimal affinity nearly converges, indicating that the optimal vehicle speed and related parameters for passing over rectangular speed bumps have been obtained.

Figure 12 Iterative process for each variable of vehicle passing over periodic rectangular-wave speed bumps. (a)  Iterative process for optimal affinity. (b)  Iterative process for individual optimal v . (c)  Iterative process for individual optimal C f 2 . (d)  Iterative process for individual optimal C r 2 .

Figure 13 shows the steady-state response curves of the vehicle suspension system for each parameter system are plotted and contrasted to the response curves prior to optimisation.

Figure 13 Comparison of vehicle response curves over periodic rectangular-wave speed bumps before and after optimisation. (a)  Response curve for objective function values F . (b)  Response curve for vertical body displacement x b . (c)  Response curve for front-suspension dynamic deflection d df . (d)  Response curve for front-wheel dynamic load d lf .

Before optimisation, the vehicle speed was 40.00 km h −1 , and after optimisation, it became 25.40 km h −1 . Upon observation of Fig. 13, it is evident that the objective function value after optimisation is significantly lower than before. The vertical body displacement has been reduced from 0.0129 to 0.0115 m. Additionally, there have been various degrees of improvement in suspension deflection and wheel load response after optimisation, decreasing from 0.0112 to 0.0054  m and from 2209.1 to 1394.9  N, respectively. These values are considerably lower than the responses before optimisation. Furthermore, it is notable that the risk of wheel lift when passing over rectangular speed bumps, which existed before optimisation, has been reduced. This indicates a substantial increase in the comfort and safety of the vehicle when crossing rectangular speed bumps after optimisation. And the acceleration variation curves of spring mass before and after optimisation are shown in Fig. 14.

Figure 14 Acceleration curves before and after optimisation over rectangular-wave speed bumps.

It can be observed that, after optimisation, there has been a certain degree of reduction in the acceleration of the spring mass. The maximum accelerations before and after optimisation are 0.9807 and 0.9330 m s −2 , respectively. According to the rms criterion, the average accelerations before and after optimisation are calculated to be 0.0612 and 0.0543 m s −2 , respectively. This indicates that, through optimisation, there has been an improvement in the comfort and safety of the vehicle when crossing rectangular speed bumps.

4.4  Comparative analysis of the optimisation results for vehicles passing over different periodic speed bumps

After the optimisation of the IA, the comparison results before and after the vehicle pass over different types of periodic speed bumps are shown in Table 3.

Table 3 Analysis of optimisation results for vehicles passing over three types of speed bumps.

After optimisation of the immune optimisation algorithm, the optimum speeds and the relevant parameters for passing trapezoidal, half-sine, and rectangular-wave speed bumps are obtained. The indicators for measuring vehicle safety and comfort have been vastly improved compared to those before optimisation, and the most obvious is the trapezoidal speed bump.

The following information can be gleaned through horizontal comparison:

Under the premise of ensuring the efficiency of passing over speed bumps, the vehicle can maintain better comfort and safety when passing over rectangular-wave speed bumps at a relatively fast speed (25.40 km h −1 ) while keeping the front and rear damping coefficients of the vehicle within a relatively low range and requiring relatively little from the body suspension.

When passing over the half-sine speed bump, it is necessary to maintain a relatively low vehicle speed (15.33 km h −1 ) . However, the relatively large range of the vehicle's front and rear damping coefficients places relatively high demands on the body suspension.

The speeds and suspension damping coefficients over the trapezoidal speed bumps are between (1) and (2).

Passing over the rectangular speed bumps requires low vehicle suspension. These speed bumps can be passed over faster; they suitable for roads where the speed limits are not very high and do not need to be passed over at a very slow speed. Trapezoidal and half-sine speed bumps have relatively high requirements in terms of the body suspension, and the damage to the body suspension is more significant when passing over quickly. Therefore, these speed bumps are suitable for sections where speed limits are demanding and where passing speeds are slow to ensure the safety of the vehicle.

By comparing results before and after the optimisation, it can be observed that the vehicle has a slight increase in vertical displacement due to the influence of the shape of the speed bumps and the change in the relevant vehicle parameters when passing over the half-sine speed bumps. Although a small part of the vehicle's comfort has been sacrificed, it ensures the safety and stability of the vehicle.

The paper constructs a multi-objective optimisation algorithm based on the IA to optimise the vertical displacement of the body, the dynamic deflection of the suspension, and the dynamic wheel loads. The experiments simulate the 4-DOF vehicle passing over the periodic trapezoidal speed bumps, the periodic half-sine speed bumps, and the periodic rectangular-wave speed bumps to solve for the optimum speed and relevant parameters for passing and draw the following conclusions:

After optimisation by the immune optimisation algorithm, the optimum speed of the vehicle can be obtained when driving over various shapes of continuous speed bumps.

Various factors will affect the safety and comfort of the vehicle when passing over different speed bumps. In addition to adjusting the speed of the vehicle, the coefficients of the front- and rear-suspension dampers need to be adjusted based on an optimisation algorithm to obtain the optimum level of comfort and safety.

After optimisation by the immune optimisation algorithm, the comfort index of the vehicle over the three types of speed bumps was significantly improved, with the most remarkable improvement being in trapezoidal speed bumps, followed by rectangular-wave speed bumps, and finally half-sine wave speed bumps, with improvements of 42.59 %, 33.18 %, and 30.27 %, respectively.

Based on the premise of ensuring the efficiency of passing over speed bumps, rectangular-wave speed bumps are suitable for being passed over at a faster speed and can obtain good comfort and safety; they are suitable for road sections with relatively low speed limit requirements and do not need to be passed over very slowly. However, trapezoidal and half-sine speed bumps require relatively high suspension damping coefficients and need to be passed over at a slower speed to obtain good comfort and safety. They are suitable for road sections where the speed limit is strict and need to be passed over carefully and slowly.

The results of the paper offer a solution to the speed adaptation problem of vehicles passing over different speed bumps, provide a scientific basis for the installation of speed bumps on different-speed-limit roads, provide a reliable reference for the study of the comfort and safety of vehicles on uneven roads, and also provide reliable data for subsequent research on the comfort of speed adaptive regulation.

However, the optimisation algorithm is not sufficient to optimise the vehicle passing over the partial speed bumps; it improves the safety when the vehicle passes over the half-sine speed bumps, but the comfort is slightly sacrificed, and further research is needed. Meanwhile, the results of this study are still at the theoretical stage, and further experimental verification and optimisation on actual vehicles are required to achieve a higher level of research results.

Most the data used in this article can be obtained by request from the corresponding author ([email protected]).

ZY and RZ proposed and developed the overall concept of the paper. ZY, RZ, and YZ conducted the experimental realisation and analysis. RZ, ZG, and JG wrote the whole paper.

The contact author has declared that none of the authors has any competing interests.

Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors.

The authors are grateful to the Fundamental Research Funds for the Science and Technology Research Project of Chongqing Municipal Education Commission, China (grant no. KJQN201903402); the Fundamental Research Funds for the Natural Science Foundation of Chongqing, China (grant no. cstc2021ycjh-bgzxm0088); and the Fundamental Research Funds for Science and Technology Research Project of Chongqing Municipal Education Commission, China (grant no. KJZD-M202303401), for the support.

This research has been supported by the Natural Science Foundation of Chongqing Municipality (grant no. cstc2021ycjh-bgzxm0088), the Program for Innovation Team Building at Institutions of Higher Education in Chongqing Municipality (grant no. CXQT21032), and the Chongqing Municipal Education Commission (grant no. KJZD-M202303401).

This paper was edited by Marek Wojtyra and reviewed by two anonymous referees.

Barreno, F., Santos, M., and Romana, M. G.: A novel adaptive vehicle speed recommender fuzzy system for autonomous vehicles on conventional two-lane roads, Expert Syst., 41, e13046, https://doi.org/10.1111/exsy.13046 , 2022. 

Chen, X.: Research on application of artificial intelligence model in Automobile Machinery Control System, Int. J. Heavy Veh. Syst., 27, 83–96, https://doi.org/10.1504/IJHVS.2020.104409 , 2020. 

Gao, J. and Qi, X.: Study of suspension parameters matching to enhance vehicle ride comfort on Bump Road, Shock. Vib., 2021, 1–20, https://doi.org/10.1155/2021/5806444 , 2021. 

Gong, M., Jiao, L., Liu, F., and Ma, W.: Immune algorithm with orthogonal design based initialization, cloning, and selection for global optimization, Knowl. Inf. Syst., 25, 523–549, https://doi.org/10.1007/s10115-009-0261-8 , 2009. 

Huang, T., Wang, J., and Pan, H.: Adaptive bioinspired preview suspension control with constrained velocity planning for Autonomous Vehicles, IEEE Trans. Intell. Veh., 8, 3925–3935, https://doi.org/10.1109/tiv.2023.3273620 , 2023. 

Liu, Z., Si, Y., and Sun, W.: Ride Comfort Oriented Integrated Design of preview active suspension control and longitudinal velocity planning, Mech. Syst. Signal Process., 208, 110992, https://doi.org/10.2139/ssrn.4545334 , 2023. 

Mahmoodabadi, M. J. and Nejadkourki, N.: Optimal fuzzy adaptive robust PID control for an active suspension system, Aust. J. Mech. Eng., 20, 681–691, https://doi.org/10.1080/14484846.2020.1734154 , 2020. 

Miracle, P. U., Bala, J. A., Aibinu, A. M., and Folorunso, T. A.: Intelligent Speed Bump Identification and speed control system for autonomous vehicles: A conceptual design, in: 2021 1st International Conference on Multidisciplinary Engineering and Applied Science (ICMEAS), Abuja, Nigeria, 15–16 July 2021, https://doi.org/10.1109/ICMEAS52683.2021.9692403 , 1–6, 2021. 

Motallebi, S., Xie, H., Tanin, E., Qi, J., and Ramamohanarao, K.: Route intersection reduction with connected Autonomous Vehicles, Geoinformatica, 25, 99–125, https://doi.org/10.1007/s10707-020-00420-z , 2020. 

Nguyen, D. N. and Nguyen, T. A.: A novel hybrid control algorithm sliding mode–PID for the active suspension system with state multivariable, Complexity, 2022, 1–14, https://doi.org/10.1155/2022/9527384 , 2022a. 

Nguyen, D. N. and Nguyen, T. A.: Evaluate the stability of the vehicle when using the active suspension system with a hydraulic actuator controlled by the OSMC algorithm, Sci. Rep., 12, 19364, https://doi.org/10.1038/s41598-022-24069-w , 2022b. 

Nguyen, D. N. and Nguyen, T. A.: Proposing an original control algorithm for the active suspension system to improve vehicle vibration: Adaptive fuzzy sliding mode proportional-integral-derivative tuned by the Fuzzy (AFSPIDF), Heliyon, 9, 19364, https://doi.org/10.1016/j.heliyon.2023.e14210 , 2023. 

Pan, H. and Sun, W.: Nonlinear output feedback finite–time control for vehicle active suspension systems, IEEE Trans. Industr. Inform., 15, 2073–2082, https://doi.org/10.1109/TII.2018.2866518 , 2019. 

Sha, S., Wang, Z., and Du, H.: Research on performance of vehicle semi–active suspension applied magnetorheological damper based on linear quadratic Gaussian control, Noise Vib. Worldw., 51, 119–126, https://doi.org/10.1177/0957456520923320 , 2020. 

Shieh, M., Chiou, J., and Liu, M. T.: Design of immune–algorithm–based adaptive fuzzy controllers for active suspension systems, Adv. Mech. Eng., 6, 916257, https://doi.org/10.1155/2014/916257 , 2014. 

Walavalkar, S., Tandel, V., Thakur, R. S., Kumar, V. V. P., and Bhuran, S.: Performance comparison of various controllers on semi–active vehicle suspension system, ITM Web Conf., 40, 01001, https://doi.org/10.1051/itmconf/20214001001 , 2021. 

Wang, L., Yang, Z., Chen, X., Zhang, R. and Zhou, Y.: Research on adaptive speed control method of an autonomous vehicle passing a speed bump on the highway based on a genetic algorithm, Mech. Sci., 13, 647–657, https://doi.org/10.5194/ms-13-647-2022 , 2022. 

Wu, J., Zhou, H., Liu, Z., and Gu, M.: Ride Comfort Optimization via speed planning and preview semi-active suspension control for autonomous vehicles on uneven roads, IEEE Trans. Veh. Technol., 69, 8343–8355, https://doi.org/10.1109/tvt.2020.2996681 , 2020. 

Wu, Y., Li, J., Nong, D., and Liu, S.: Nonlinear vibration analysis for automobile stimulated by speed control hump, J. Dyn. Control., 12, 147–152, https://doi.org/10.6052/1672-6553-2013-072 , 2014. 

Yan, S., Sun, W., He, F., and Yao, J.: Adaptive fault detection and isolation for active suspension systems with model uncertainties, IEEE Trans Reliab, 68, 927–937, https://doi.org/10.1109/TR.2018.2868949 , 2019. 

Yang, Z., Liang, S., Sun, Y., and Zhu, Q.: Vibration suppression of four degree–of–freedom nonlinear vehicle suspension model excited by the consecutive speed humps, J. Vib. Control., 22, 1560–1567, https://doi.org/10.1177/1077546314543728 , 2016.  

Yang, Z., Liang, S., Zhu, Q., Zhou, T., and Zeng, C.: Chaotic vibration and comfort analysis of nonlinear full–vehicle model excited by consecutive speed control humps, Math. Probl. Eng., 2014, 1–9, https://doi.org/10.1155/2014/370634 , 2014. 

Yang, Z., Liang, S., and Zhou, T.: Research on Chaotic Vibration Characteristics of Nonlinear Half Vehicle Model under Quasi–Periodic Excitation, J. Zhengzhou. Univ. (Eng Sci), 2015, 5, https://doi.org/10.3969/j.issn.1671-6833.2015.01.026 , 2015. 

Yang, Z., Wang, L., Liu, F., and Li, Z.: Nonlinear dynamic analysis of constant–speed and variable–speed of autonomous vehicle passing uneven road, J. Vibroengineering., 24, 726–744, https://doi.org/10.21595/jve.2022.22250 , 2022. 

Yu, Y., Zhao, L., and Zhou, C.: Influence of rotor–bearing coupling vibration on dynamic behavior of electric vehicle driven by in–wheel motor, IEEE Access, 7, 63540–63549, https://doi.org/10.1109/ACCESS.2019.2916554 , 2019. 

Zein, Y. and Darwiche, M.: Fuzzy logic velocity optimization of autonomous vehicles based on road bump geometry, BAU J. Sci. Technol., 1, 2, https://doi.org/10.54729/2959-331X.1016 , 2020. 

Zhang, W. and Zheng, S.: Dynamic Characteristics Simulation Analysis of Tracked Vehicle Tire under Different Road Conditions, Tire Ind., 42, 7, https://doi.org/10.12135/j.issn.1006-8171.2022.01.0003 , 2022. 

Zhu, Q. and Ishitobi, M.: Chaos and bifurcations in a nonlinear vehicle model, J. Sound Vib., 275, 1136–1146, https://doi.org/10.1016/j.jsv.2003.10.016 , 2004. 

• Introduction
• Simulation models
• Multi-objective optimisation algorithm
• Application of optimisation algorithm
• Conclusions
• Data availability
• Author contributions
• Competing interests
• Acknowledgements
• Financial support
• Review statement

Research on internal quality testing method of dry longan based on terahertz imaging detection technology

• Original Paper
• Published: 11 May 2024

Cite this article

• Jun Hu   ORCID: orcid.org/0000-0003-0027-7993 1 ,
• Hao Wang 1 ,
• Yongqi Zhou 1 ,
• Shimin Yang 1 ,
• Haohao Lv 1 &
• Liang Yang 1  

Longan is a kind of nut with rich nutritional value and homologous function of medicine and food. The quality of longan directly affects its curative effect, and its fullness is the key index to evaluate its quality. However, the internal information of longan cannot be obtained from the outside. Therefore, rapid non-destructive testing of internal quality of dry longan is of great significance. In this paper, rapid non-destructive testing of longan internal fullness based on terahertz transmission imaging technology was carried out. This study takes longan as the research object. Firstly, the terahertz transmission images of longans with different fullness were collected, and the terahertz spectral signals of different regions of interest were extracted for analysis. Then, three qualitative discriminant models, support vector machine (SVM), Random forest (RF) and linear discriminant analysis (LDA), were established to explore the optimal discriminant model and realize the distinction of different regional categories of longan. Finally, the collected longan terahertz transmission image is processed, and the number of white pixels in the connected domain is calculated by using Otsu threshold segmentation and image inversion. The fullness of longan can be achieved by calculating the ratio of core and pulp to the pixel of the shell. The LDA discriminant model had the best prediction effect. It could not only identify the spectral data of background region, shell region, core region, but also reach 98.57% accuracy for the spectral data of pulp region. The maximum error between the measured fullness and the actual fullness of the terahertz image processed by Otsu threshold segmentation is less than 3.11%. Terahertz imaging technique can realize rapid non-destructive detection of longan fullness and recognition of different regions. This study provides an effective scheme for selecting the quality of longan.

Graphical abstract

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price includes VAT (Russian Federation)

Instant access to the full article PDF.

Rent this article via DeepDyve

Institutional subscriptions

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.

X. Zhang, S. Guo, C.T. Ho et al., Phytochemical constituents and biological activities of longan ( Dimocarpus longan Lour.) fruit: a review. Food Sci. Hum. Wellness 9 (2), 95–102 (2020)

Article   Google Scholar  

H. Lim, J. Lee, S. Lee et al., Low-density foreign body detection in food products using single-shot grid-based dark-field X-ray imaging. J. Food Eng. 335 , 111189 (2022)

Article   CAS   Google Scholar  

D. Mery, I. Lillo, H. Loebel et al., Automated fish bone detection using X-ray imaging. J. Food Eng. 105 (3), 485–492 (2011)

I. Orina, M. Manley, S. Kucheryavskiy et al., Application of image texture analysis for evaluation of X-ray images of fungal-infected maize kernels. Food Anal. Methods 11 , 2799–2815 (2018)

X. Cheng, M. Zhang, B. Adhikari et al., Effect of power ultrasound and pulsed vacuum treatments on the dehydration kinetics, distribution, and status of water in osmotically dehydrated strawberry: a combined NMR and DSC study. Food Bioprocess Technol. 7 , 2782–2792 (2014)

S. Baek, J. Lim, J.G. Lee et al., Investigation of the maturity changes of cherry tomato using magnetic resonance imaging. Appl. Sci. 10 (15), 5188 (2020)

M.S. Razavi, A. Asghari, M. Azadbakh et al., Analyzing the pear bruised volume after static loading by magnetic resonance imaging (MRI). Sci. Hortic. 229 , 33–39 (2018)

F. Wang, C. Zhao, H. Yang et al., Non-destructive and in-site estimation of apple quality and maturity by hyperspectral imaging. Comput. Electron. Agric. 195 , 106843 (2022)

P. Zhang, H. Wang, H. Ji et al., Hyperspectral imaging-based early damage degree representation of apple: a method of correlation coefficient. Postharvest Biol. Technol. 199 , 112309 (2023)

J. Sun, K. Tang, X. Wu et al., Nondestructive identification of green tea varieties based on hyperspectral imaging technology. J. Food Process Eng 41 (5), e12800 (2018)

SY M Y. Terahertz Pulsed Imaging in Reflection Geometry[D]. The Chinese University of Hong Kong, 2011.

Z. Zhu, J. Zhang, Y. Song et al., Broadband terahertz signatures and vibrations of dopamine. Analyst 145 (18), 6006–6013 (2020)

Article   CAS   PubMed   Google Scholar  

J. El Haddad, F. de Miollis, J. Bou Sleiman et al., Chemometrics applied to quantitative analysis of ternary mixtures by terahertz spectroscopy. Anal. Chem. 86 (10), 4927–4933 (2014)

Article   PubMed   Google Scholar  

C. Wang, R. Zhou, Y. Huang et al., Terahertz spectroscopic imaging with discriminant analysis for detecting foreign materials among sausages. Food Control 97 , 100–104 (2019)

Y. Shen, Y. Yin, B. Li et al., Detection of impurities in wheat using terahertz spectral imaging and convolutional neural networks. Comput. Electron. Agric. 181 , 105931 (2021)

X. Sun, J. Li, Y. Shen et al., Non-destructive detection of insect foreign bodies in finishing tea product based on terahertz spectrum and image. Front. Nutr. 8 , 757491 (2021)

Article   PubMed   PubMed Central   Google Scholar  

J. Hu, H. Shi, C. Zhan et al., Study on the identification and detection of walnut quality based on terahrtz imaging. Foods 11 (21), 3498 (2022)

B. Li, D. Zhang, Y. Shen, Study on terahertz spectrum analysis and recognition modeling of common agricultural diseases. Spectrochim. Acta Part A Mol. Biomol. Spectrosc. 243 , 118820 (2020)

W. Liu, C. Liu, X. Hu et al., Application of terahertz spectroscopy imaging for discrimination of transgenic rice seeds with chemometrics. Food Chem. 210 , 415–421 (2016)

X. Sun, J. Liu, Measurement of plumpness for intact sunflower seed using Terahertz transmittance imaging. J. Infrared Millimeter Terahertz Waves 41 (3), 307–321 (2020)

D.J. Jwo, W.Y. Chang, I.H. Wu, Windowing techniques, the welch method for improvement of power spectrum estimation. Comput. Mater. Contin 67 , 3983–4003 (2021)

Google Scholar  

S. Pan, H. Zhang, Z. Li et al., Classification of Ginseng with different growth ages based on terahertz spectroscopy and machine learning algorithm. Optik 236 , 166322 (2021)

X. Wei, D. Kong, S. Zhu et al., Rapid identification of soybean varieties by terahertz frequency-domain spectroscopy and grey wolf optimizer-support vector machine. Front. Plant Sci. 13 , 823865 (2022)

H. Zhang, Z. Li, T. Chen et al., Discrimination of traditional herbal medicines based on terahertz spectroscopy. Optik 138 , 95–102 (2017)

Y. Cao, J. Chen, G. Zhang et al., Characterization and discrimination of human colorectal cancer cells using terahertz spectroscopy. Spectrochim. Acta Part A Mol. Biomol. Spectrosc. 256 , 119713 (2021)

S. Yang, C. Li, Y. Mei et al., Determination of the geographical origin of coffee beans using terahertz spectroscopy combined with machine learning methods. Front. Nutr. 8 , 680627 (2021)

J. Liu, H. Xie, B. Zha et al., Detection of genetically modified sugarcane by using terahertz spectroscopy and chemometrics. J. Appl. Spectrosc. 85 , 119–125 (2018)

J. Li, W. Luo, Z. Wang et al., Early detection of decay on apples using hyperspectral reflectance imaging combining both principal component analysis and improved watershed segmentation method. Postharvest Biol. Technol. 149 , 235–246 (2019)

J. Hu, P. Qiao, L. Yang et al., Research on nondestructive detection of pine nut quality based on terahertz imaging. Infrared Phys. Technol. 134 , 104798 (2023)

Wang Y T, Li Q. A segmentation algorithm on terahertz digital holographic image[C]//14th National Conference on Laser Technology and Optoelectronics (LTO 2019). SPIE, 2019, 11170: 310–315.

H. Li, J.X. Wang, Z.N. Xing et al., Influence of improved Kennard/Stone algorithm on the calibration transfer in near-infrared spectroscopy. Spectrosc Spect Anal 31 (2), 362–365 (2011)

Download references

Acknowledgements

National Youth Natural Science Foundation of China (32302261); Jiangxi Ganpo Talented Support Plan -Young science and technology talent Lift Project (2023QT04); Jiangxi Provincial Youth Science Fund Project (20224BAB215042); National Key R&D Program of China (2022YFD2001805).

Author information

Authors and affiliations.

School of Mechatronics & Vehicle Engineering, East China Jiaotong University, Nanchang, 330013, Jiangxi, China

Jun Hu, Hao Wang, Yongqi Zhou, Shimin Yang, Haohao Lv & Liang Yang

You can also search for this author in PubMed   Google Scholar

Contributions

Jun Hu: Investigation, Writing-review and editing, Experimental scheme design, Formal analysis. Hao Wang: Writing-original draft, Formal analysis. Yongqi Zhou: Experiment. Shimin Yang, Haohao Lv: Review and editing. Liang Yang: Formal analysis.

Corresponding author

Correspondence to Jun Hu .

Ethics declarations

Competing interests.

We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of, the manuscript entitled. Jun Hu, Hao Wang, Yongqi Zhou, Shimin Yang, Haohao Lv, Liang Yang declare that they have no conflict of interest.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Hu, J., Wang, H., Zhou, Y. et al. Research on internal quality testing method of dry longan based on terahertz imaging detection technology. Food Measure (2024). https://doi.org/10.1007/s11694-024-02583-x

Download citation

Received : 29 December 2023

Accepted : 18 April 2024

Published : 11 May 2024

DOI : https://doi.org/10.1007/s11694-024-02583-x

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Longan fullness
• Terahertz imaging technology
• Otsu threshold segmentation
• Linear discriminant analysis (LDA)
• Find a journal
• Publish with us
• Track your research