UNTRIPPED AND TRIPPED ROLLOVERS WITH A NEURAL NETWORK

, Gridsada Phanomchoeng; , Kailerk Treetipsounthorn; , Sunhapos Chantranuwathana; , Lunchakorn Wuttisittikulkij

doi:2023.24.3.811

International Journal of Automotive Technology > Volume 24(3); 2023 > Article

Body and Safety, Connected Automated Vehicles and ITS, Electrical and Electronics

International Journal of Automotive Technology 2023;24(3): 811-828.
doi: https://doi.org/10.1007/s12239-023-0067-9

UNTRIPPED AND TRIPPED ROLLOVERS WITH A NEURAL NETWORK

Gridsada Phanomchoeng ^1,2, Kailerk Treetipsounthorn ³, Sunhapos Chantranuwathana ^1,2, Lunchakorn Wuttisittikulkij ⁴

¹Department of Mechanical Engineering, Chulalongkorn University
²Smart Mobility Research Unit, Faculty of Engineering, Chulalongkorn University
³Department of Computer Engineering, Chulalongkorn University
⁴Department of Electrical Engineering, Chulalongkorn University

Corresponding Author. Gridsada Phanomchoeng , Email. gridsada.p@chula.ac.th

ABSTRACT

To improve rollover prevention and rollover warning systems, indicators for detecting rollover risks are extremely important. Vehicle rollover accidents occur in one of two ways: tripped and untripped rollovers. For detecting tripped rollovers, the traditional rollover index is ineffective; most precise rollover indicators depend on dynamic models that must identify all the parameters for computations. In this study, we focused on exploring a new index for detecting tripped and untripped rollovers using a neural network (NN). Four types of NNs, i.e., FNN, Tanh, long short-term memory, and gated recurrent unit (GRU), were examined to develop models for estimating rollover indices. The results demonstrated that the GRU and large Tanh network are the most suitable NNs for untripped and tripped rollover prediction, respectively. Moreover, the untripped rollover prediction model having a small GRU network could precisely anticipate the trend of the untripped rollover indicators for up to 0.2 s in advance. Moreover, the created tripped rollover anticipation model with a large Tanh network could precisely forecast the trend of the tripped rollover index up to 0.5 s in advance. Based on these results, rollover prediction in future can be advantageous for rollover prevention and warning systems.

Key Words: Rollover index, Tripped rollover, Untripped rollover, Feedforward neural network, Recurrent neural network, LSTM, gated recurrent unit (GRU), Tanh

TOOLS