Research On Intelligent Diagnosis And Prediction Of Motor Fault Based On Multi-dimensional Information Fusion And Visual Knowledge

As a complex coupling system of machinery,electricity and magnetism,the motor is the key power source in various industries.Its service quality often determines the service quality of many core equipments and high-end products.However,the motor is developing towards large-scale,high-speed and integration in recent years,which makes its performance behavior complex and the working environment extreme,so it leads to the reduction of service quality and is not conducive to efficient manufacturing and production.For some industrial equipment requiring "near zero fault" operation,it will even bring serious economic losses and major disasters.Therefore,building an information-based,intelligent condition monitoring and fault diagnosis system that matches the rhythm of production demand is the key to reduce the maintenance cost of complex motor system and improve the service quality,which has important theoretical and practical significance.At present,the electromechanical faults of motor system have the characteristics of various types,complex structure and difficult monitoring.The research on conventional signal detection methods and fault diagnosis methods focuses on extracting the detailed features of signal in time domain,frequency domain,and pays insufficient attention to the global information features of signal,which has great limitations.Multidimensional information fusion and multidimensional information visualization are important research trends of big data and artificial intelligence algorithms.Human brain can fuse visual and auditory signals to form multi-dimensional information,so as to realize efficient learning,insight,analysis and reasoning of target objects or tasks,and obtain knowledge.Therefore,this paper takes this as inspiration to realize visual knowledge expression and multi-dimensional information fusion of sensor signals.In this paper,the single dimensional time domain data is extended to the highdimensional information space,and the motor intelligent diagnosis and prediction system based on multi-dimensional information knowledge is established.This research is oriented to engineering practical application problems,focuses on the cutting-edge technology of the discipline,and has important theoretical guidance and engineering practical significance for the reliable operation and safe service of motor system on the basis of innovative theoretical methods and practical engineering application.The key research contents of this paper include:(1)An improved Ada Boost motor fault diagnosis method based on attention mechanism and multi-source one-dimensional information fusion is proposed.In this method,the onedimensional feature information of abundant measurement signals is extracted by using the onedimensional signal processing method,and a multi-dimensional information space is formed after the analysis and fusion of a variety of one-dimensional information to describe and express the different health states of the motor system.Using the complementarity between the multidimensional information can greatly improve the accuracy of fault diagnosis in different motors.This method dynamically evaluates the sensitivity of different types with detection information to different faults.On the basis of extracting the single-dimensional information of a variety of heterogeneous signals,it fuses to form a multi-source information space,establishes the contribution mapping between a variety of single-dimensional heterogeneous signal data,and extends the low-dimensional signal data to the high-dimensional information space through the method of information fusion.(2)A visual knowledge matching motor fault diagnosis method based on invariant point feature is proposed.SDP algorithm is used to express one-dimensional signal as twodimensional image data,and sift image invariant feature extraction algorithm is used to realize high-dimensional spatial data processing and data mining.Combined with the two algorithms,the mapping relationship between actual fault and image intuitive features is established.This method effectively and intuitively expresses the collected single-dimensional signals from the perspective of two-dimensional visual knowledge,which can realize the global expression of motor health state and capture its detailed local features at the same time.This method realizes the extraction and optimal utilization of signal information with high richness,and reduces the demand for the number and types of sensors.(3)An intelligent motor fault diagnosis method based on image visual information and Bag of Words model is proposed.On the basis of previous work,this method combines twodimensional visual knowledge with artificial intelligence algorithm by means of Dense-SIFT and Bag of Words model,makes full use of and fuses image redundant features,obtains the intelligent mapping between motor health status and fault information,and solves the problem that image feature information is difficult to be directly understood and utilized by general machine learning algorithms.This provides a reference for the machine vision system to observe,recognize and predict the operation of motor as keenly as human beings in the future.(4)A motor fault prediction method based on spatial-temporal graph information is proposed.In this method,the motor health state is expressed from another multi-dimensional knowledge form through Markov transition field graph information,and the spatial-temporal correlation feature of graph information is extracted and learned by multivariate spatialtemporal series graph neural network.On the basis of the local spatial state information and global time evolution information of the monitoring signal,the long-term and short-term fault prediction of the motor health state are carried out.This method can transform the collected data into multiple node structure expressions containing multi-dimensional structure and attribute information,and can simulate the spatial and temporal correlation and heterogeneity of its dynamic graph data.It is a new attempt and application expansion of multi-dimensional visual information knowledge.This paper is closely focusing on the major common requirements of China’s industrial manufacturing and people’s livelihood economy for the safety and stability in high-end technical equipment.Facing the complex motor system widely used in modern manufacturing industry,based on the theory of information expression,learning and reasoning with different dimensions of state signal,the whole link “State-Data-Information-Knowledge” is preliminarily established in this paper.Combined with the visual knowledge expression method represented by invariant feature extraction and the corresponding machine learning,this research puts forward effective fault diagnosis and prediction methods,which expands the breadth and depth of motor system in different application fields.It provides theoretical guidance and practical reference for improving service quality and reducing the maintenance cost of complex motor system.