Research On Fusion Method Of Visual Information And Auditory Information In Mechanical Equipment Intelligent Diagnosis

In the existed methods of information fusion for equipment fault diagnosis,the analysis of internal parameters,such as vibration,pressure,temperature,acceleration,are more commonly implemented.However,the utilize of the external state parameters,especially visual information and auditory information are mostly ignored.Compared to the internal state information,visual and auditory information is more direct to equipment status,and more universal.The diagnosis system based on visual and auditory sensors owns more concise information flow mechanism and the hardware topology structure,also has better real-time performance and portability.Based on these considerations,visual and auditory information that represent equipment external status are studied in this dissertation to equipment fault.By studying visual-auditory information fusion model and method in equipment fault diagnosis,the limitations in which fault diagnosis only relied on the internal sense information can be eliminated,and a new approach to solve the complex problems in equipment fault diagnosis is provided.This dissertation started with the research achievement of neurophysiology,to describe the universality and the validity of information fusion in human brain for environment recognition.Further,the feasibility of visual-auditory information fusion are demonstrated based on the information transmission and circulation mechanism.On this basis,the core problems in visual-auditory information fusion are systematically studied.The main contents of this dissertation include:(1)Auditory signal is vulnerable to industrial noise,and the signal-to-noise ratio of is low,aiming at these problems,the fault feature extraction methods of auditory signal are studied.Combined the adaptive decomposition ability of Empirical Mode Decomposition and the separation ability to blind sources of Independent Component Analysis,the EMD-ICA joint processing frame is established,modal-aliasing was avoided effectively in EMD decomposing,meanwhile the restricted problem of signal source number in ICA analysis is solved.(2)To reduce the impact of insensitive features to recognition accuracy rate,a modified distance based feature sensibility assessment and selection method is proposed.In this method,otherness factor is considered,the gathering degree and the between-class distance of different features can be better distinguished,a more reasonable cluster result is obtained,base on it,the input vector dimension and computation time can be reduced,and the classification accuracy will increase.A kind of intelligent fault diagnosis model based on EMD-ICA joint processing technology,the modified feature extraction method and BP neural network is established,and the effectiveness of the model is verified by the simulation experiment for bearing fault diagnosis.(3)Focused on the uncertainty problem of visual-auditory information in fault diagnosis,the information fusion method based on fuzzy-neural network is lucubrated.The principle,structure and algorithm of adaptive neural-fuzzy based inference system(ANFIS)are dissected in depth.Aiming at the problems of too much iteration time,local minimum and so on,ANFIS algorithm is modified herein by the approaches of fletcher-reeves update conjugate gradient method,scale conjugate gradient(SCG)method,and levenberg-marquardt method.The diagnosis experiment based on the modified ANFIS for materials handing process fault diagnosis of sorting manipulator in manganese agent packaging production line is conducted.The result show that,in the diagnosis process based on the fusion of visual and auditory information,the ANFIS modified by LM method owns faster convergence and higher classification accuracy.Compared with the diagnosis results relied on visual signal or auditory signal alone,the effectiveness of visual-auditory information fusion is verified.The principle of LM algorithm is lucubrated,and it further improved by bringing in step adjustment factor,so the degree of dependence on some control iteration parameters is reduced.At last,the improved effect is verified by experiments.(4)Aiming at contradictoriness of visual and auditory information in different signal channels,the evidence combination and inference principle of D-S evidence theory is lucubrated,for the problem of “Evidence Paradox”,the improved evidence theory fusion algorithm based on weight redistribution is proposed,through bringing in the conflict factor,the conflict degree between one evidence source and others could be represented more appropriate,and the rationality of the evidence inference is increased through the fusion jurisdiction redistribution.(5)A layered intelligent diagnosis model is proposed.Firstly,the input feature vector is screened by the feature extraction and selection method which proposed in this dissertation,then the faults are identified primarily by multi-ANFIS corresponding different kinds of data,the results of primary diagnosis are regarded as different evidence,and the ultimate diagnosis decision is provided by D-S evidence inference method.By using this model,the vector of diagnosis space is reduced obviously,and the network structure was simplified.To different diagnosis problems,it has better robustness and flexibility,the uncertainty and the contradictoriness of information features were handled effectively.Finally,the diagnosis effect of the model was verified in “Phone Screen AOI Automatic Detection System”.The experiment date show that by fusing visual and auditory information by the proposed layered intelligent diagnosis model,the equipment fault status which difficult to express can be identified effectively by the fusion of visual-auditory information,meanwhile the real-time performance of the system is guaranteed.