Research On Data-driven Fault Diagnosis Of Heating Boiler

With the increasing demand of society,modern heating enterprises are developing in the direction of informatization,scale and intelligence.At the same time,higher requirements are put forward for the reliability and safety of the heating boiler control system.Therefore,it is an important link to ensure the safe and stable operation of the system to carry out real-time fault monitoring of the boiler operation process,timely identify the fault state points in the production process and traceability of the fault cause variables.Based on the multivariate statistical analysis method in data-driven,this paper conducts fault diagnosis research on the multivariate,multimodal and nonlinear process characteristics of heating boiler industrial process.The main research work of this paper includes the following aspects.Firstly,a fault detection method based on cross-section PCA is proposed for the multi-modal commonly existing in industrial processes.Cluster analysis of industrial process data is carried out based on modal recognition value,and the continuous industrial process data is classified by cross segmentation,which effectively realizes the division of stable mode and transition mode.For the fault detection of stable mode,the applicability of the steady-state fault model is improved by matching the data characteristics with the corresponding fault detection model.In view of the transition characteristics of gradual transition mode,a comprehensive determination method of multi-stage steady-state sub-model is proposed to realize the fault detection of transition mode.The fault monitoring results of the actual process of the heating boiler are used to verify that compared with the traditional PCA method,this method reduces the omission and false alarm,and the accuracy of fault detection is improved.Secondly,in view of the low accuracy of anomaly detection in complex industrial process monitoring by kernel principal component analysis method,an improved KPCA fault detection method based on PSO algorithm is proposed from two aspects of kernel function form and parameters.Aiming at the selection of kernel function form,Combining the generalization ability of global kernel function and the learning ability of local kernel function,a new hybrid modeling method combining polynomial kernel and Gaussian kernel is proposed,so that the improved hybrid kernel function has better comprehensive ability.Based on the improved KPCA method,PSO algorithm is introduced to optimize the parameters of the improved kernel function to further improve the linear separability of the mixed kernel function.The fault detection experiment of TE chemical process shows that the accuracy of fault detection is improved.Finally,under the background of a boiler actual operation project in Dalian,combined with its specific process requirements,a heating boiler fault monitoring and diagnosis platform based on GUI is designed.With the help of the powerful computing power of the computer,the theoretical method of fault detection and diagnosis proposed in this paper is tested on the spot.The results show that the method in this paper can accurately and efficiently identify and diagnose the faults occurring in the boiler operation,and has higher accuracy than the traditional method.This study has certain reference value for ensuring the continuous,stable and safe operation of complex production processes.