Research On Mechanical Fault Diagnosis Technology Based On Deep Learning Theory

Under the industrial background of intelligent manufacturing and big data,how to make use of the big data generated in the process of manufacturing system and solving problems is particularly important.How to use big data to promote the development of intelligent manufacturing is becoming more and more critical.Equipment and maintenance are the core elements of intelligent manufacturing,and the past three industrial revolutions have also revolved around these two core elements for technological upgrading.The normal operation of production system needs reliable health management system of machinery and equipment to guide,but hidden troubles of production system will lead to different types of faults,and different types of faults have different impacts on production system.How to use industrial data to diagnose faults of mechanical parts on the basis of in-depth learning is the focus of this paper.Firstly,how to transform the time domain signal of vibration signal to the frequency domain signal by fast Fourier transform is studied,and the quantifiable and effective health features which are strongly correlated with the vibration signal are extracted.Gauss noise is added to the vibration signal to improve the generalization of the neural network in the fault diagnosis model.Aiming at the low signal-to-noise ratio data collected from industrial environment,the maximum overlapping discrete wavelet transform is proposed to reduce the noise of vibration signals and obtain a fast spectral kurtosis map.This method can reduce the parameters of the neural network model,improve the accuracy of the model diagnosis and accelerate the convergence speed of the neural network at the same time.Through the study of the neural network,a neural network is built based on the framework of deep learning tensorflow.The characteristic index obtained by fast Fourier transform is used as the input of the neural network.The model has achieved good diagnostic results on the bearing data set of Case Western Reserve University.The problem of the recognition rate of the diagnostic model is reduced when the fault diagnosis model obtained under a single working condition is used for fault diagnosis under variable working conditions.In this paper,the migration mode and migration neural network are improved,and the neural network is pruned and compressed.Experiments show that the diagnostic model achieves high diagnostic accuracy and strong adaptability.