Predictive Maintenance (PdM) of automobiles requires the storage and analysis of large amounts of sensor data. This requirement can be challenging in deploying PdM algorithms onboard the vehicles due to limited storage and computational power on the hardware of the vehicle. Hence, this study seeks to obtain low dimensional descriptive features from high dimensional data using Representation Learning. The low dimensional representation can then be used for predicting vehicle faults, in particular a component related to the powertrain. A Parallel Stacked Autoencoder based architecture is presented with the aim of producing better representations when compared to individual Autoen-coders with focus on vehicle data. Also, Embeddings are employed on categorical Variables to aid the performance of the artificial neural networks (ANN) models. This architecture is shown to achieve excellent performance, and in close standards to the previous state-of-the-art research. Significant improvement in powertrain failure prediction is obtained along with a reduction in the size of input data using our novel deep learning ANN architecture.

© Springer Nature Switzerland AG 2020