A Bayesian-optimized Hybrid Neural Network Based on CNN and BiLSTM for Predictive Maintenance of Diesel Generator

Maintenance routines on power supply vehicles used to follow reactive maintenance which can be regarded as post-failure repair and might create large costs. Recently, prognostic and health management (PHM) has emerged as a potential way to develop an ideal maintenance strategy and reduce industrial losses. The data-driven methods for diagnosis and prognosis have been deeply concerned. This paper presents a fault diagnosis and prognosis method for diesel generator in the power supply vehicles based on convolutional neural network (CNN) and bidirectional long short-term memory network (BiLSTM). The CNN is used to extract the features of time-series data by converting sequence data into two-dimensional images without artificial feature extraction. A softmax regression layer added to the hybrid neural network of CNN and BiLSTM makes it available for fault diagnosis. An experiment verification using the dataset acquired from a certain model of power supply vehicle is implemented. In this work, the Bayesian optimization algorithm is used to search the optimal hyperparameters of the models. Results show the excellent ability of prediction, and demonstrate the superiority of the proposed method in the field of fault diagnosis and prognosis compared with other models.