A comparative study of series hybrid approaches to model and predict the vehicle operating states. (December 2021)

• Record Type: Journal Article
• Authors: Alizadeh, Morteza; Ma, Junfeng
• Is Part Of: Computers & industrial engineering. Volume 162(2021)
• Publication Date: 2021-12
• Subjects: Vehicle Operating Subsystems -- Behavior Modeling and Prediction -- Auto-Regressive Integrated Moving Average -- Wavelet Neural Network -- Multilayer Perceptrons Neural Network; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107770
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: Compare series hybrid approaches to model and predict vehicle operating behavior. Threshold based anomaly detection method to early detect unhealthy states of vehicle. Compare performance models using a large scale time series vehicle behavior data. Abstract: With the growing complexity of modern vehicle system, the capability of modeling the behavior of different subsystems and predicting their forthcoming patterns become vital. It helps to extend vehicle's life cycle and control their maintenance costs. Leveraging statistical and deep learning techniques, the massive maintenance data can be used to model the behaviors of different subsystems of vehicle, predict the future trend, and consequently assist in making appropriate maintenance decisions. In this study, Auto-Regressive Integrated Moving Average (ARIMA), Multilayer Perceptrons Neural Network (MLPNN), and Wavelet Neural Network (WNN) were used to develop several series hybrid models (i.e. ARIMA-MLPNN, ARIMA-WNN, MLPNN-ARIMA, and WNN-ARIMA) to model and predict the operational behaviors of vehicle's subsystems. Moreover, a threshold-based anomaly detection method was developed for early detection of abnormalities. A real case study including three months records of 97 subsystems of an operating vehicle was used to validate the efficiency of these hybrid models. Results showed that the WNN-ARIMA model obtained the most accurate results compared with other hybrid models. A threshold-based anomaly detection approach was developed based on the residual errors of the WNN-ARIMA model. This approach precisely captures the abnormal states of various subsystems of the vehicle which can help to make more accurate decisions regarding the maintenance of the vehicle.