| Pneumatic control valves,as the terminal actuating element of industrial process automation,are essential equipment for industrial sites.Due to the complex and diverse industrial site environments,pneumatic control valves are prone to a variety of failures,resulting in significant losses to personal safety and property.In this regard,early fault diagnosis and prediction of pneumatic control valves and their components can effectively avoid unnecessary losses caused by planned downtime.In recent years,datadriven fault analysis methods have been widely used in various electromechanical equipment.In view of this,this paper adopts a data-driven fault diagnosis and prediction method to carry out research on valve positioner,a key component of pneumatic regulating valve.The main research are as follows.（1）The working mechanism and common failures of pneumatic control valves and its key components,valve positioners,are analyzed to clarify the common types of failures,the causes of their formation and the maintenance measures after the failures are generated,so as to lay the foundation for the subsequent construction of the pneumatic control valve positioner experimental platform.Then the corresponding experimental platform is built according to the working principle of the pneumatic regulating valve positioner,which is used to obtain its operating status data.（2）Five common mechanical faults are artificially injected into the valve positioner on the pneumatic regulating valve positioner experimental platform,followed by the valve positioner fault simulation experiments under each fault condition,and the operating state data of the valve positioner under the fault condition are obtained through the acquisition module arranged.Then,the data pre-processing is performed on the obtained operating condition data,and the noise reduction is completed by using wavelet threshold denoising on the data set,and the numerical normalization method is used to eliminate the difference in magnitudes of the multi-source data,and the sample size of the data set is enriched by sliding window data expansion.（3）A fault diagnosis method based on empirical modal decomposition and multibranch convolutional neural network is proposed for the fault diagnosis model applicable to valve positioners.The empirical modal decomposition method is used to perform empirical modal decomposition on the multi-source data set so that the feature information of the data set can be better extracted by the convolutional neural network,and the convolutional neural network is improved into a multi-branch structure so that it can learn the data collected by each sensor individually,and finally the final fault diagnosis results are output by fusing the features through the output layer.Finally,the advantages of the proposed new model are verified by comparing the traditional fault diagnosis methods.（4）A temporal convolutional neural network prediction method based on improved particle swarm algorithm optimization is proposed for the state parameter prediction model applicable to valve positioners.The model uses automatic parameter search to obtain the optimal parameters of the prediction model in order to improve the accuracy of condition data prediction.For the proposed method,the root mean square error and the average absolute percentage error are used as evaluation indicators,and the prediction accuracy of the model is verified by using the operating condition data of the valve positioner under normal operating conditions.The validation results show that the proposed prediction model can effectively improve the prediction accuracy of the valve positioner condition operation data. |
