Fault Tolerance in Electric Vehicles Using Deep Learning for Intelligent Transportation Systems

Intelligent transportation systems (ITS) such as hybrid electric vehicles make use of sensing technologies to improve mobility, safety and efficiency. Automated manual transmission (AMT) is a mechatronic device consisting of Internet of Things (IoT)-enabled sensors and actuators responsible for automatic gear shifting in hybrid electric vehicles. Any failure in these sensors or actuators can affect the normal operation of vehicles. Therefore, this study aims to discuss the characteristics of AMT when it breaks down and its impact on the whole hybrid electric vehicle system. Firstly, this paper briefly introduces the relevant overview of hybrid electric vehicles and AMT. Next, analytical redundancy analysis is used to find out the possible faults of each part of AMT and the causes of each fault. The normal and faulty signals are then passed to a reduced depth kernel extreme learning machine (RDK-ELM) algorithm, which combines the deep learning network structure with the kernel-based selection of support vectors from the training samples. The RDK-ELM is a fault diagnosis algorithm that classifies the normal and faulty signals, which represent whether the sensor is faulty or not. Simulation results show that the algorithm has high classification accuracy of 97.12% and it requires less time for training the model.

Automated summary

Intelligent transportation systems (ITS), including hybrid electric vehicles, use sensing technologies to enhance mobility, safety, and efficiency. Automated manual transmission (AMT) is a mechatronic device in hybrid electric vehicles that automatically shifts gears using IoT-enabled sensors and actuators. This study investigates the characteristics of AMT failures and their impact on the overall hybrid electric vehicle system. The researchers apply an analytical redundancy analysis to identify possible faults in each part of the AMT, and then use a reduced depth kernel extreme learning machine (RDK-ELM) algorithm to classify normal and faulty signals for sensor fault diagnosis. Simulation results demonstrate the algorithm's high accuracy of 97.12% and its efficiency in training the model.