Improved Deep Convolution Neural Network with Applications in Fault Diagnosis of Naval Gun Mechanical Components

Aiming at the problems of complex structure and low accuracy of fault diagnosis of mechanical components, this paper proposes a fault diagnosis method based on improved deep convolution neural network, and applies it to the diagnosis of naval gun mechanical components to realize "End-to-End" fault diagnosis. Firstly, the vibration signal is preprocessed by Fast Fourier Transform to get the time-frequency spectrum samples, which are divided into training set and test set. Then, the improved activation function, dropout and other deep learning training techniques are introduced to prevent the model from over fitting, and SVM is introduced to output the fault recognition results. Finally, the improved deep convolution neural network model is established to realize the fault diagnosis of mechanical components. The model proposed in this paper was verified through the bearing data set of Case Western Reserve University, and the results showed that the average diagnostic accuracy rate can reach 97.10% under variable working conditions, indicating that the model has good feasibility and advantages; Moreover, under the training of the laboratory's actual measurement data set of naval gun mechanical components, it has shown a high recognition accuracy, and the diagnosis accuracy rate can reach 98.3%, which can provide a reference for future fault diagnosis work of naval gun mechanical components.