Intelligent fault diagnosis of marine diesel engine based on deep belief network

  Objectives  In order to improve the accuracy of intelligent fault diagnosis of marine diesel engine, the deep learning is introduced, and a method based on deep belief network(DBN)for intelligent fault diagnosis of marine diesel engine is proposed.

  Methods  The multilayer restricted Boltzmann machine (RBM) was used to stack DBN, and the parameters of the model were solved by contrast divergence method. This method adopted a new training mode including unsupervised pre-training and supervised fine-tuning, which could learn and extract deep hidden features from the fault sample data automatically, and obtain better initialization weights.

  Results  After the analysis of the sample data collected from the experiment of fault simulation for marine diesel engine based on AVL BOOST, the results show that the recognition rate of DBN to training sample set and test sample set is 98.26% and 98.61% respectively, so DBN has higher fault identification accuracy and higher generalization performance than BP neural network(BPNN)and support vector machine(SVM), and can avoid the shortcomings of the shallow neural network due to randomly initialized weights, such as local minima and low precision.

  Conclusions  Compared with BPNN and SVM, DBN is more suitable for intelligent fault diagnosis of marine diesel engine.