A Valve Stiction Quantification Method Based on Deviation Product Analysis and Neural Network

In response to the widespread issue of valve stiction in process industries, a quantification method based on neural network and deviation product analysis has been proposed. Initially, loop data of varying degrees of stiction, including controller output and process variable, are generated. Through deviation product analysis, dynamic features within the non-stationary time series are extracted. Subsequently, a model capable of accurately identifying the degree of stiction was trained by employing neural network for deep learning on these features, along with a designed loss function that has dynamically adjustable weights. This approach achieved high-precision quantification of valve stiction. Experiments and data analysis have demonstrated the effectiveness of this method in practical industrial applications.