Predicting the Equipment Useful Lifetime Based on the Deep Neural Networks

Abstract

Listen

Translate article

AbstractA method for predicting the useful time of equipment based on the processing of diagnostic data using parallel deep recurrent and convolutional neural networks is proposed. In the considered method, the recurrent neural network processes diagnostic data, which are presented in the form of time series. A convolutional neural network processes images obtained on the basis of continuous wavelet transform of diagnostic data. Neural networks implement a multivalued classification of two states of technological equipment—serviceable and faulty. The forecast of the values of the output signals of neural networks, performed on the basis of the recursive least squares method, is used to estimate the useful time of equipment. The results of a model experiment carried out using a program developed in the Matlab 2020b environment that implements the proposed method are presented. To generate training datasets, a transmission system model was used, implemented in the Simulink software package, and supplied as part of the Matlab 2020b environment. The architectures of recurrent and convolutional neural networks used in the experiment are presented. The wavelet transform of training datasets for a convolutional neural network was performed based on the analytical Morse wavelet. The experimental results showed that parallel-connected deep neural networks of various architectures increase the accuracy of estimating the useful time of equipment.KeywordsTechnical diagnosticsData predictingDeep neural networks