Machine Learning-Based Fault Detection for Intelligent Vehicle Control

Tan, Ching Hao (2025) Machine Learning-Based Fault Detection for Intelligent Vehicle Control. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Tan Ching Hao - FULL TEXT.pdf Restricted to Registered users only Download (5MB)

Abstract

This report details the development of the fault detection system for autonomous vehicles utilizing Long Short-Term Memory (LSTM) networks. The system focuses on analyzing time-series data from the Controller Area Network (CAN) bus, including torque, acceleration, and velocity. The LSTM model is chosen for the capability of managing long-term dependencies and sequential patterns within complex vehicle dynamics. The methodology involves data preprocessing, cycle segmentation, and the application of LSTM models to predict normal behavior. Fault types such as spike, drift, bias, hardover, and erratic are simulated and introduced into the velocity data to assess the ability of the system to detect anomalies. The performance of the model is evaluated through regression metrics and comparative plots of predicted versus actual values. Results show strong predictive capabilities, with a Mean Absolute Error (MAE) of 1.2295, Root Mean Square Error (RMSE) of 3.6483, and an R-squared (R2) value of 0.7994. The system also achieves up to 99.52% accuracy, with F1 scores ranging from 83.33% to 96.15% across different fault types. These results demonstrate the effectiveness of LSTM for real-time fault detection, offering improvements in safety and reliability. Future work includes model optimization, expanding sensor coverage, and enhancing multi-fault classification.