Investigating noise reduction in signal analysis in rotary machines fault diagnosing by neural network

Fault diagnosis of mechanical systems is of special importance for better system performance as well as its protection. In this work, a rotary motor laboratory system is used to generate signals and data. The obtained data are placed in the pre-processing process. In this article, to improve the performance of signal analysis, the combined analysis methods using signal features and Kalman filter are proposed. At first, the Kalman filter is used to reduce the signal noise. And in the following, for signal pre-processing, the features of the signal in the time domain and frequency domain are suggested, which have been used as one-dimensional signal pre-processing. In the following, several neural networks such as support vector machine, multilayer perceptron, and convolutional neural networks have been used to analyze the obtained features. To check the results, the data is divided into training data and validation data. Accuracy results for validation data are examined in different methods. The results indicate the better performance of the AlexNet convolutional neural network in the presence of the Kalman filter denoising process. In this case, this network has reached an average of 96.1% accuracy for validation data, which has been improved compared to other classifiers and fault diagnosis without noise reduction.