A hybrid ARIMA–WNN approach to model vehicle operating behavior and detect unhealthy states

Abstract

As modern vehicles system becomes increasingly complex, there is an urgent need to develop a framework to monitor the behavior and detect the unhealthy states to appropriately arrange the maintenance in order to extend the vehicle life cycle. Sensors installed in vehicles are able to record a huge amount of multiple channel time series data. This paper develops a prediction model and an unhealthy state detection strategy for monitoring the behavior of the operating vehicle by analyzing multiple channel time series data channels. Due to the complexity of the underlying process of multiple channel time series data, a hybrid ARIMA–WNN model is proposed by combining the Auto-Regressive Integrated Moving Average (ARIMA) and Wavelet Neural Network (WNN) models. This model exploits the ARIMA capacity for predicting the unseen patterns without labeled historical data and WNN’s flexibility in analyzing different variations of time series data. Furthermore, a threshold-based anomaly detection strategy is developed to timely recognize the unhealthy states of the vehicle. To test the performance of the proposed hybrid ARIMA–WNN, a large scale multiple-channel time series data including three months of second-wise records of 101 channels of an operating vehicle is used. Results reveal the higher accuracy of the proposed ARIMA–WNN compared with the ARIMA and WNN models in modeling the behavior of different data channels. By estimating the proper distribution of the threshold, the proposed anomaly detection method is also efficient for identifying the unhealthy states of the vehicle.