Towards Formula SAE Driver/Vehicle Optimization

Slides:



Advertisements
Similar presentations
Developed by Reneta Barneva, SUNY Fredonia
Advertisements

Filament Winding Machine. Project Name – Filament Winding Machine Project Number – P09226 Project Family – Competitive FSAE Autosports Systems Project.
Proactive Prediction Models for Web Application Resource Provisioning in the Cloud _______________________________ Samuel A. Ajila & Bankole A. Akindele.
Snapshot Presentation for:  NIATT seeks to be a leader in “Green Vehicle” technologies  Hybrid Technologies can improve performance as well as efficiency.
Civil and Environmental Engineering Carnegie Mellon University Sensors & Knowledge Discovery (a.k.a. Data Mining) H. Scott Matthews April 14, 2003.
Data Acquisition Real-Time System Integration DARSI Matt Hulse Preston Schipper Marc Kessler Chris Lites Andy Lin.
Data Acquisition Real-Time System Integration Preston Schipper Matt Hulse Adrienne Baile DARSI II.
Forces, Energy and Motion Robert Jesberg FETC Conference Orlando, FL January 28 – 31,
What FSAE is?? It is an organisation that carries out student automotive racing competitions around the world. FSAE racing competitions take place every.
John Lisiecki Dynamic Events Chief Steward Ed Arthur Assistant Chief 3/21/2012 © 2012 SAE International 2012.
A Lightweight Platform for Integration of Resource Limited Devices into Pervasive Grids Stavros Isaiadis and Vladimir Getov University of Westminster
Annealing Paths for the Evaluation of Topic Models James Foulds Padhraic Smyth Department of Computer Science University of California, Irvine* *James.
Se Over the past decade, there has been an increased interest in providing new environments for teaching children about computer programming. This has.
Dr. Michael D. Featherstone Summer 2013 Introduction to e-Commerce Web Analytics.
Testing Considerations Because of potential life threatening consequences resulting from device malfunction, it is critical that the device be fully tested.
Neural Network Implementation of Poker AI
CMDBs: Above and Beyond…
Engineering Design: Bridge Building Michael Starr 1, Tamara Jonson 2 1 College of Engineering, University of Cincinnati, Cincinnati OH; 2 Withrow University.
Project Final Presentation Joe Featherall.
Impact Research 1 Enabling Decision Making Through Business Intelligence: Preview of Report.
Information Eastman. Business Process Skills Order to Cash, Forecasting & Budgeting, etc. Process Modeling Project Management Technical Skills.
Data Science Interview Questions 1.What do you mean by word Data Science? Data Science is the extraction of knowledge from large.
Data Summit 2016 H104: Building Hadoop Applications Abhik Roy Database Technologies - Experian LinkedIn Profile:
Prediction of Box Office Gross Revenue
Seminar Announcement December 24, Saturday, 15:00-17:00, Room: A302, WNLO Title: Quality-of-Experience (QoE) and Power Efficiency Tradeoff for Fog Computing.
Big data classification using neural network
Acknowledgement: Khem Gyawali
Gist of Achieving Human Parity in Conversational Speech Recognition
SNS COLLEGE OF TECHNOLOGY
OPERATING SYSTEMS CS 3502 Fall 2017
African-American Stimuli
as presented on that date, with special formatting removed
SAMPLE Glimpse Into the Future Using Predictive HR Analytics
Siemens Enables Digitalization: Data Analytics & Artificial Intelligence Dr. Mike Roshchin, CT RDA BAM.
Cisco Data Virtualization
Assessing Students' Understanding of the Scientific Process Amy Marion, Department of Biology, New Mexico State University Abstract The primary goal of.
Literature Surveys Source : : Keshav P. Dahal (Bradford University)
Utilizing AI & GPUs to Build Cloud-based Real-Time Video Event Detection Solutions Zvika Ashani CTO.
Big-Data Fundamentals
Street Cleanliness Assessment System for Smart City using Mobile and Cloud Bharat Bhushan, Kavin Pradeep Sriram Kumar, Mithra Desinguraj, Sonal Gupta Project.
Development of built-in diagnostics in the RADE framework (EN2746)
Campus Locator – Definition Phase (May04-04)
Worthing College Sports Science Liam Lee 2015
Validation through Instrumentation
CIS 336 STR Education Your Life-- cis336papers.com.
Elinar Ai Miner powered by IBM watson.
ASAP and Deep ASAP: End-to-End Audio Sentiment Analysis Pipelines
Reverse Engineering: A Roadmap
Using Tensorflow to Detect Objects in an Image
The Design Process What Is Design? What Is a Design Process?
Kickoff Presentation Date of Presentation Presenter Name
Speech Capture, Transcription and Analysis App
Northern New Mexico College Department of Engineering
Integrating Deep Learning with Cyber Forensics
REMOTE POWER MONITORING OF MARINE SITES
UNIT 5 EMBEDDED SYSTEM DEVELOPMENT
UNIT 5 EMBEDDED SYSTEM DEVELOPMENT
NHD NATIONAL HISTORY DAY.
UNH Precision Racing Data Acquisition, Telemetry and Analysis
DESIGN OF EXPERIMENTS by R. C. Baker
Big Data Analytics 2019 Applicative Exercises : Exercise 2
Design of Experiments CHM 585 Chapter 15.
Towards Formula SAE Driver/Vehicle Optimization
From Use Cases to Implementation
LHC beam mode classification
Process Wind Tunnel for Improving Business Processes
Credit Card Fraudulent Transaction Detection
Adrian E. Gonzalez , David Parra Department of Computer Science
Austin Karingada, Jacob Handy, Adviser : Dr
Machine Learning for Cyber
Presentation transcript:

Towards Formula SAE Driver/Vehicle Optimization Salvador Jimenez, Austin Luchsinger Department of Computer Science College of Engineering and Computer Science Abstract   Many factors are considered when designing and racing Formula SAE racing cars. Recently, machine learning has sparked a significant shift in Formula 1 racing with an announcement that they will be transferring their infrastructure over to AWS. Currently, there are no machine learning techniques being used at the collegiate Formula SAE level. In this work, we use a Neural Network to try and learn the relation between throttle, braking, and turning parameters and lap time. Data The data that Cal Poly - Pomona provided was from the 2016 Formula SAE Lincoln event. One of the events at FSAE Lincoln was an endurance event where teams had to race around a track 15 times, with time for one pitstop and driver change. Of the data collected, the primary concern of Bronco Motorsports was the throttle, brake, and lateral acceleration data. Below is a plot of the raw data collected: After some processing, we split the data up into laps, and compared the segments again. Due to some noise and processing challenges (specifically with positional data), there was a small amount of data loss. Results The first Neural Network architecture we set up achieve an accuracy of about 70%. This results of our run are shown below: While this is not quite what we were hoping for, it is better than nothing. We did think though, that we could do better. It was after getting this result that we tried the second network setup shown in the last section. That network, however, produced even worse results. This network couldn’t learn anything about the relation between throttle/brake/lateral acceleration and lap time. Despite our lackluster results, our approach provides evidence for future directions. Future Work We did not achieve the results we were hoping, however, we are currently working on new approaches. Since our data is all timestamped (and position-stamped) we now think that we should try time-series analysis. We had the right idea about learning the parameters of a specific lap segment, but we weren’t taking all of the previous lap segments into consideration. We have begun implementing an LSTM to learn/predict these parameters. We will write the results of our LSTM in our final report by the end of the week. Introduction Every year, all of the vehicles in the Formula SAE racing league are designed and built from scratch. No vehicle is ever used two years in a row. This means that these Formula SAE teams are constantly reevaluating which design aspects are most important when developing their vehicles. With the vast number of factors involved in racecar design, it can be difficult to determine which of these factors is most important. Machine learning may be able to provide greater insight into this challenge. In recent news [1], the Formula One Group announced that it will be using Amazon Web Services Inc (AWS) and data analytics to make in-race predictions about which driver seems most likely to win the race. By analyzing current race information, and comparing it to 60 years worth of race results, expected winner predictions are made. While this is a nice application of machine learning that benifits the fans, it does little to help give the Formula One teams an edge on the competition. At the collegiate level, Formula SAE teams are looking to evolve in the same way. It is common for FSAE teams to install sensors on their vehicles which track various statuses throughout a race. These sensors track various things such as tire pressure, brake temperature, throttle, engine temperature, lateral acceleration, and more. Normally, these readings are used for mid-race diagnostics and anticipating/preventing component failure. Following Formula 1’s lead, some FSAE teams are investigating the use of machine learning on the data they have been collecting. The Problem California State Polytechnic University – Pomona [2] is one of the schools that has begun seeking machine learning solutions for their problems. One of their members contacted us, requesting our assistance in extracting information from their prior race data. Specifically, they want to know how big of an impact the throttle, brake, and lateral acceleration have on lap time. In other words, of all the factors that the driver can control, which is the most critical? 1 Brake (percentage) Lateral Acceleration (percentage) Our Approach Our goal was to learn the relationship between the 3 parameters (throttle, brake, acceleration) and lap time. At first, we wanted to use linear regression, as it is a good predictive model. The problem, however, is that the relationship seems to be non-linear. We then considered polynomial regression, as that is capable of learning non-linear relations, but this requires prior knowledge of how the factors relate to each other. Thus, we landed on a Neural Network. This model seemed to best fit our needs, as Neural Networks are capable of learning non-linear relations with no prior knowledge about the factors. Below, we show the architecture of our Neural Network: The concept behind this setup was simple: given a lap segment, what should the throttle, brake, and lateral acceleration be? As we show in the results section, the network was limited in what it could learn. Following this, we tried another architecture. Given the throttle, brake, lateral acceleration, current position, current time, and next position, can we learn what the next time should be? This architecture is shown below: Acknowledgments We would like to thank David Martinez and the California State Polytechnic University, Pomona for providing motivation and data for this investigation. We would also like to thank Dr. Kim for teaching the Machine Learning class which made our work possible, and for detailed discussions on how to approach this problem. Lastly, we would like to thank Angel Cantu for countless conversations about our project, and for his helpful software implementation recommendations. References Formula One selects AWS as Official Cloud and Machine Learning Provider. 29 June 2018. www.formula1.com Bronco Motorsports – Formula SAE team. California State Polytechnic University, Pomona www.cppfsea.com A Beginner’s Guide to Neural Networks and Deep Learning. https://skymind.ai/wiki/neural-network A Beginner’s Guide to LSTM’s and Recurrent Neural Networks. https://skymind.ai/wiki/lstm Keras: The Python Deep Learning Library https://keras.io