Valve Stiction Quantification Based on Riemannian Manifold

摘要

With the modern industrial processes getting advanced and complicated, accurate evaluation for the status of valves in control loops has become increasing significant. Nevertheless, control valves often suffer from stiction nonlinearity, which may bring about the malfunction of control loops, eventually resulting in an unanticipated breakdown or unacceptable performance deterioration. Hence, the recent past two decades have also witnessed a huge growth on valve stiction detection strategies for industry process. However, to the best of author's knowledge, the question of how to quantify further aspects of valve stiction remains unanswered. In particular, most of the existing stiction quantification methods share the same limitation that they may occasionally fail to realize the estimation for the slipjump characteristic of valve stiction, and may be ineffective for the cases in which sticky valves are applied in the cascade control loops. Considering these drawbacks, this paper puts forward a novel quantification method for valve stiction based on the exponential and logarithmic maps attached onto the Riemannian manifold. The key idea is to exploit the relationship between process variables and controller output which constitutes a specific shape (e.g., ellipsoid) on a Riemannian manifold. The method can be capable to deal with three typical stiction features, including deadband, stickband, and slipjump for single control loops and simple cascade loops. The case studies of simulated examples validate that the proposed method is a promising method for quantifying the valve stiction's deadband plus stickband, and slipjump accurately.