Research On Sub-health Recognition Algorithms Of RVM Optimized Strong Reconfiguration MDAE

With the improvement of science and technology,the development of modern industry has made great progress,and the production equipment has also been replaced by intelligent technology.In this process,the requirement for equipment itself is constantly improving.Damage of any component will have an important impact on the production process,but will cause economic losses,and some heavy industry fields may pose a threat to life and safety.Sub-health identification of equipment status can largely avoid various losses caused by downtime maintenance.Because the failure of equipment is not an instantaneous cause,but a cumulative process,the research of sub-health identification of equipment status has attracted extensive attention of experts and scholars.After reading and analyzing a lot of in-depth learning and bearing sub-health recognition methods,aiming at the difficulty of extracting features and the unsatisfactory diagnostic effect of traditional bearing sub-health recognition,this paper proposes a RVM optimized strong reconstruction MDAE sub-health recognition algorithm(IMDAE-IFMRVM).The algorithm is deeply studied from two aspects: improving the automatic encoder of edge noise reduction and improving the correlation vector machine.Firstly,aiming at the problems of limited constraints,weak data compression ability and large reconstruction error of marginal denoising automatic encoder,on the one hand,sparse constraints are added on the basis of original marginal denoising constraints to solve the problems of limited constraints and weak data compression ability of marginal denoising automatic encoder,on the other hand,the input of hidden layer is changed to the output of the former hidden layer.In addition,the pretreated data can solve the shortcomings of the edge noise reduction automatic encoder,such as large reconstruction error and weak expression ability of feature data.Then,aiming at the problem that the single kernel function can't deal with the complicated sample data effectively when the information is different and the distribution is uneven in the feature sample data,two methods of adding the weights of the two kernels are used instead of the original single kernel function acting independently.For the determination of the weight coefficients of the kernels,the problem of determining the weight coefficients of the kernels is solved under the joint action of Fisher criterion and maximum entropy criterion.The variance mapped into the sample into the feature space is obtained by the method proposed in this paper.Finally,through the contrast experiment,it is found that the improved algorithm can correctly identify the bearing data status,and the effectiveness of the algorithm proposed in this paper is verified.