Random Convolutional Neural Network Structure: An intelligent health monitoring scheme for diesel engines

Abstract Automatic and accurate identification on the health condition of the diesel engine is a challenging task in the modern industry. In this paper, an innovative deep learning network structure called Random Convolutional Neural Network (RCNN) is designed for the intelligent health monitoring of diesel engines, taking full advantages of deep learning and ensemble learning. Firstly, this novel network framework is constructed with several individual convolutional neural networks (CNN), which can automatically extract the discriminative features of vibration signals by convolutional calculation and pooling operation. Secondly, an improved optimizer called Adabound and the Dropout technique are adopted in the framework of RCNN. The Adabound optimizer uses adaptive learning rates to accelerate the training of network and avoid plunging into local optimum. Finally, a combinational rule is used to fuse the diagnostic results from several individual CNN. The experimental vibration signals acquired from the diesel engine prove that the efficiency and superiority of the proposed RCNN.