Intelligent Maintenance System For Large Equipment Based On Signal Feature Fusion

Machinery and equipment in the industrial production of the most basic and one of the most important links,with the continuous development of material technology,industrial information technology,a lot of special equipment gradually to the direction of heavy equipment,large equipment development.Large equipment is large and complex,and the consequences of a major accident can be unbearable.At present,planned maintenance is the main means of equipment maintenance,which is relatively conservative.Most of the replaced equipment still has a certain amount of remaining use time,and the utilization rate of manpower and material resources is low.Fault diagnosis and residual life prediction based on equipment degradation can provide theoretical and data support for intelligent maintenance,greatly improve resource utilization and reduce equipment operating costs.On the basis of the research group’s existing intelligent monitoring and decline prediction of large equipment,this paper analyzed the characteristics of various signals of the equipment,and carried out fault diagnosis and residual life prediction.The main work contents are as follows:(1)According to the characteristics of different types of signals,the signal processing and feature extraction are analyzed,and the noise reduction method based on KM-SVD is proposed to reduce the noise of vibration signals.The different features of the device signal in the time domain and frequency domain are extracted to form the original feature vector set.(2)Fault diagnosis is carried out through two time-varying signals of the equipment-sound and vibration signals.For acoustic signals,a Cluster-Hidden Markov Model based on feature fusion is proposed for fault diagnosis.The improved empirical wavelet transform(EWT)is used to diagnose vibration signals based on support vector machine(SVM)and its good decomposition effect is verified.Both of the two fault diagnosis methods have been proved to have good fault diagnosis ability by experiment.(3)Correlation,monotonicity and robustness are combined as evaluation indexes,and thresholds are set to screen features in the original feature vector set,so as to obtain effective features that can represent equipment degradation.Then,the effective features were fused by the method of principal component analysis,and the health factor HI was constructed to represent the health degree of the equipment.The constructed HI could effectively represent the degradation process of the equipment.The extracted HI was input into LSTM to predict the remaining life.The experiment was compared with the prediction methods in the literature,and it was verified that the proposed method had better performance.(4)Based on the above method of fault diagnosis and residual life prediction,an intelligent maintenance system for large equipment based on the fusion of signal features is designed,including modules of condition monitoring,fault diagnosis and warning,condition assessment and prediction,maintenance strategy,etc.This paper integrates feature extraction and fusion,fault diagnosis and early warning,residual life evaluation and prediction,and provides a data-driven intelligent maintenance method.This method can provide effective theoretical and data support for the optimization of maintenance strategy,and is of great significance for improving the utilization rate of maintenance resources,realizing intelligent maintenance,and ensuring the reliable and stable operation of large equipment.