Performance Analysis and Enhancement of Deep Convolutional Neural Network

Convolutional neural network has been widely investigated for machinery condition monitoring, but its performance is highly affected by the learning of input signal representation and model structure. To address these issues, this paper presents a comprehensive deep convolutional neural network (DCNN) based condition monitoring framework to improve model performance. First, various signal representation techniques are investigated for better feature learning of the DCNN model by transforming the time series signal into different domains, such as the frequency domain, the time-frequency domain, and the reconstructed phase space. Next, the DCNN model is customized by taking into account the dimension of model, the depth of layers, and the convolutional kernel functions. The model parameters are then optimized by a mini-batch stochastic gradient descendent algorithm. Experimental studies on a gearbox test rig are utilized to evaluate the effectiveness of presented DCNN models, and the results show that the one-dimensional DCNN model with a frequency domain input outperforms the others in terms of fault classification accuracy and computational efficiency. Finally, the guidelines for choosing appropriate signal representation techniques and DCNN model structures are comprehensively discussed for machinery condition monitoring.