Research On Quality-related Process Monitoring Method Based On Autoencoder And Partial Least Squares

Partial least squares(PLS)has been receiving continuous attention in the field of quality-related process monitoring.PLS can construct quality-related latent variables from a large number of process variables.To address the issue of nonlinearity inherent in complex industrial processes,the kernel partial least squares(KPLS)is proposed.However,KPLS can only extract shallow features from process variables,making it difficult to capture more complex input-output relationship.Therefore,a combination of autoencoder(AE)and KPLS is considered to design the novel quality-related process monitoring methods.The main research of this study is as follows.On the one hand,a new quality-related process monitoring method via integrating stacked sparse autoencoder(SSAE)with KPLS(SSAE-KPLS)is proposed.First,an SSAE model is employed to exploit the nonlinearity within process variables.Through SSAE,hierarchical features are learned to extract latent representations of process variables from multiple sparse autoencoder(SAE)layers.Second,the learned hierarchical features from SSAE are used as input,and the final quality variables are used as output.A KPLS model is then built to exploit the nonlinear relationship between the hierarchical features extracted from process variables and the final product quality for process monitoring.Third,Hotelling’s T~2and Q statistics are employed to detect the quality-related and quality-unrelated faults,respectively.Finally,experiments on a numerical example and the commonly used industrial benchmark of the Tennessee Eastman process(TEP)are conducted to illustrate the efficacy and merits of the proposed SSAE-KPLS based quality-related process monitoring method by comparing it with other related methods.On the other hand,a new quality-related process monitoring method based on SAE and comprehensive KPLS(SAE-CKPLS)is proposed.Specifically,the SAE is employed to compress the original process variables to a reasonable dimension in a nonlinear way.Then,taking the quality variables as the output,KPLS models are established for the hidden layer and residual of SAE,respectively.Accordingly,the original process variable space is decomposed into four subspaces:quality-related and process-related subspace,quality-unrelated but process-related subspace,quality-related but process-unrelated subspace,quality-unrelated and process-unrelated subspace.Construct appropriate statistics for the first three subspaces to achieve a comprehensive process monitoring scheme.Finally,a hot rolling mill process and the TEP are used to prove the superiority of the proposed method by comparing it with other classical methods.