Multi-branch convolutional neural networks with integrated cross-entropy for fault diagnosis in diesel engines
Haipeng Zhao, Zhiwei Mao, Jinjie Zhang, Xudong Zhang, Nanyang Zhao, Zhinong Jiang
Abstract
Abstract Fault diagnosis based on deep learning has become a hot research topic because of the successful application of deep learning in other fields. Due to variable operating conditions and a harsh operating environment, it is extremely difficult to effectively diagnose some typical faults of diesel engines. When operating conditions and environmental factors change, the performance of deep learning models also become extremely unstable. In order to solve these problems, this paper proposes a novel deep learning model, called multi-branch convolutional neural networks (MBCNNs) with an integrated cross-entropy. MBCNN can be embedded in the proposed model and simultaneously equipped with four auxiliary classifiers. The proposed model is trained on two different datasets separately, which consist of six diesel engine faults. The trained network model was compared with other methods to prove the superiority of this network model. Meanwhile, by adding Gaussian white noise, the performance of the MBCNN in different noise environments is investigated. The final results show that the MBCNN with integrated cross-entropy can effectively diagnose different diesel engine faults under variable operating conditions.