Research And Application Of AdaBoost Ensemble Multi-fault Diagnosis Method Based On Reconstructed LTSA

In the process of modern industry,the probability of multiple faults obviously outranges a single fault in actual industrial processes.If multiple faults occur,it will pose a greater threat to the production safety of industrial processes.It is noted that the composition of multiple faults is quite uncertain,which brings the difficulties to the fault recognition for complex industrial processes.With the fast development of information systems,the collection and storage of data have become easy.How to effectively analyze and use the information in the data to establish multi-fault diagnosis model is the main content of this paper.The details are as follows:In this paper,a novel RPCA-SVME based multiple faults recognition is proposed.First,obtain the principal component information from the original high-dimensional data space.Second,to solve the loss of local feature information,reconstruct the local structural error of the feature space through the inverse mapping matrix,and then align the error to obtain the reconstructed coordinates.Third,based on the OvO ensemble strategy,an SVME classifier is constructed for multiple faults recognition.Finally,a RPCA-SVME model for multi-fault diagnosis is established.As the collected data in the industrial process has the characteristics of nonlinearity and strong coupling,linear manifold learning methods(such as PCA algorithm)cannot effectively extract the embedded manifold structure from data,which will affect the expression ability of the reconstructed feature.To obtain a model with higher performance,an AdBE model based on the RLTSA algorithm is proposed.First,the MLE method is used to estimate the intrinsic dimension for the process data,and the local tangential space of each sample is aligned to obtain a low-dimensional manifold structure embedded in the original data space.Second,to solve the loss of global feature information,an affine matrix is used to inversely map the low-dimensional coordinates to restore the global structure.Finally,to improve the accuracy and calculation speed of the multi-classifier,AdaBoost algorithm is used to enhance the SVM classifier,and the OvR ensemble strategy is used to construct an AdBE classifier.The RLTSA-AdBE model for multi-fault diagnosis is established.For RPC A algorithm and RLTSA algorithm,visualization is made for reconstruction results based on the Circle standard data and the S_curve standard data set to prove the effectiveness.Besides,the MLE algorithm is verified based on the Swiss Roll standard dataset.The proposed RPCA-SVME model and RLTSA-AdBE model are simulated using the TEP.The comparison results show that,the RPCA-SVME can guarantee higher diagnostic accuracy and macro_F1 score,and the RLTSA-AdBE model has a better performance than the RPCA-SVME.