Research On Quality-related Fault Root Diagnosis Method Based On Correlation And Causality Analysis

Large-scale industrial production processes involve plenty of different equipment,companying with all kinds of faults or failures occurring from time to time.As a result,accurate and rapid fault diagnosis has an important impact on operation safety and product quality.In the practical production,the faults that could affect produce quality need acquiring fault diagnosis results and taking measures more efficiently and accurately.With the development of the technology of the sensor and data storage,the multivariate statistical process monitoring method belonging to data-driven field has achieved great results in fault diagnosis.This paper mainly studies the quality-related fault diagnosis algorithms of nonlinear industrial processes,providing an available and efficient fault diagnosis algorithm framework for the industrial process monitoring system.And the monitoring results can provide the information basis for the engineers to take the appropriate measures towards the occurring faults in time.So that,the safety of the production could be guaranteed,when the quality of the product and the operating profit could also be ensured.Firstly,the quality-related data modeling and fault detection algorithms are studied and improved in this paper.On the basis of the analysis of the traditional models,kernel sample equivalent replacement theory is combined with three data modeling algorithms respectively,namely kernel partial least square,kernel direct decomposition and orthogonal kernel principal regression.These improved modeling methods are applied to achieve feature space decomposition without the problem of the unclear relationship between kernel samples and process variable samples.The simulations performed on the numerical model and the superheated steam process model proves the accuracy of the improved fault detection algorithms.The data model improved by kernel sample equivalent replacement theory has created great conditions for the classical linear fault diagnosis algorithms to be operated on the nonlinear processes.Based on the improved model,contribution plot method and the method based on reduction of combined index are applied to extract the contribution of the variables directly.These two faulty variable diagnosis methods are equipped concise calculation process,which is able to reduce time consumption greatly,compared with the partial derivation diagnosis algorithm.The simulation results demonstrate that these two methods are able to diagnose faulty variables in the nonlinear processes on the basis of the improved data model,while improving the diagnosis efficiency.Based on the improved data model acquired by the kernel sample equivalent replacement theory,the classic linear fault diagnosis algorithm could be introduced to the diagnosis of nonlinear process.After the construction of the improved data model,contribution plot method and reduction of combined index method are applied to extract the contribution of the variables and diagnose the faults in the nonlinear processes.In comparison with the calculation process of the partial derivation method,the remarkable effect on the computation efficiency is analyzed and explained in this paper.The simulation analysis demonstrates that the data model after the replacement could make the linear diagnosis method available for nonlinear process objects and reduce the amount of computation.Finally,this paper further looks into the diagnosis algorithm of the root cause for the fault occurring on the basement of the transfer entropy theory.This algorithm is able to calculate the amount of information transferred among the faulty variables to acquire the overall causality direction.Then,it analyzes the relationships among faulty variables to determine the root cause variables of the fault occurring.The performances of this method are shown by the simulations on the numerical model and superheated steam process,which proves that the root cause variables could be found accurately with the transferentropy-based method.The root cause diagnosis algorithm has found the fault location successfully.