Ranganathan Srinivasan

Approaches for efficient stiction compensation in process control valves

Limit cycles caused due to valve nonlinearity such as stiction can be eliminated with proper valve maintenance. Valve maintenance is undertaken during production stops, which are scheduled once every 6 months to 3 years. The loss of energy and product quality during this intermediate period can be quite high. Stiction compensation algorithms can mitigate this problem to a large extent. In this paper, two novel approaches for stiction compensation are proposed: (a) a simple two-move approach and (b) an optimization based approach much in the spirit of predictive control strategies. Both the approaches are based on a data-driven model for stiction. The merits and demerits of both these approaches are discussed. The results are illustrated using simulation case studies. The two-move approach is also validated on a liquid level system.

Issues in modeling stiction in process control valves

Stiction has been reported as the most commonly occurring nonlinearity in control valves. In the literature, mechanistic and data based models have been proposed to characterize stiction. In this paper, the available models are critically analyzed. The complexities associated with modeling stiction are highlighted. It is shown through experiments on industrial valves that in the presence of static and dynamic friction, the valve behavior is dependent on the rate of the valve input. An approach to model this rate dependent valve behavior - which is not considered in existing data driven models - is proposed.

A Novel attempt to reduce engineering effort in modeling non-linear chemical systems for Operator Training Simulators

Operator Training Simulator (OTS) applications have become the norm of the industry in training operators to achieve efficient process operations. First principles based modeling approach in OTS packages achieves realistic simulations of chemical processes. However modeling the kinetics and thermodynamics accurately require considerable engineering efforts and may involve experimental studies to match the plant behavior. Hybrid models also known as grey-box models replace the unknown/complex equations in first principles models with empirical relationship using functional approximators such as neural networks, polynomials, etc. In this work we explore the use of Kernel Principal Component Analysis (K-PCA) as an approximation technique for certain nonlinear thermodynamics or kinetic functions parameterized using available plant archived data. Simulation results on a complex binary distillation column demonstrate the applicability of the proposed novel approach.