Hydraulic Fault Classification Using Spectrogram Images

In this paper, a signal processing-based technique for hydraulic fault diagnosis is identified. This technique measures the hydraulic state parameters and extracts the corresponding features. By using algorithms like SVM, Random Forest, and a modified CNN (3 convolutional layers and 2 hidden layers), it is possible to compare the relationship between the measured signals and faults. The respective output accuracies are obtained and compared. The advantages of residual learning and a dense connection are combined in this modified CNN model, which not only makes it simple to extract deep features but also efficient when using shallow features. Then, based on the results we obtained from the modified convolutional neural network, SVM, and random forest, we classify the results, and the average output accuracy is determined as 64.7 for SVM and 70.3 for Random Forest. The average output accuracy we obtained for the modified CNN is 97.5%.