VALVE STICTION DETECTION THROUGH IMPROVED PATTERN RECOGNITION USING NEURAL NETWORKS

Valve stiction is a very commonly occurring fault within control valves that is difficult to detect and diagnose.Many stiction detection methods in literature have shown to either be lacking in detection accuracy, or require too much information which renders it difficult for use in a wide number of process types. In this paper,anon-invasive method for detecting valves suffering from stiction using a multilayer feed-forward artificial neural networks (ANN) is proposed. The detection and differentiation of whether a valve is suffering from a stiction problem is done through a simple class-based diagnosis. The model uses transformation of PV (process variable) and OP (controller output variable), which can be easily selected from routine operational data. Samples used for training are generated from a data-driven stiction simulation using Choudhury’s model