Smart Approaches for Detecting and Quantifying Stiction of Control Valve Using Signal Processing and Machine Learning Techniques

Variations in process variables is one of the major concern in industrial plants. This varation can have a significant impact on both the quality of the product and the performance of equipment, resulting in increased operating and maintenance costs as well as potential safety issues. One of the common sources of process variation is control valve stiction. Therefore, early detection and quantification of valve stiction is crucial for maintaining product quality and reducing maintenance costs. Diagnosing stiction in a smart valve equipped with a positioner is a relatively simple process. However, for an ordinary control valve, the diagnostic process is not straightforward due to absence of control valve output. As there are large number of ordinary control installed in plants, our focus in this research on developing smart algorithms for detecting and quantifying stiction for this specific type of control valves. The model estimates control valve output (MV) using an inverse model and smart search methods. In order to implement the proposed approach, we used state-of-the-art signal processing and machine learning techniques to develop a trustworthy strategy for stiction analysis for ordinary control valve which is applicable also to variable operating setups case. Testing on publicly available industrial data showed that the proposed approach functioned well for all tested scenarios to detect the stiction and provided a good estimate of the stiction value compared to other avalible techniques.