8th International IFAC Symposium on Dynamics and Control of Process Systems COMPARISON BETWEEN TWO FRICTION MODEL PARAMETER ESTIMATION METHODS APPLIED TO CONTROL VALVES Rodrigo Alvite Romano and Claudio Garcia Polytechnic School, University of São Paulo, Brazil Abstract: Friction in the control valve is one of the major sources of oscillations and ab- normal operations in process control loops. Models of control valves can be used to diag- nose abnormal operations as well as to compensate for such undesirable effects. This work describes two methods used to estimate the parameters of a Karnopp friction model applied to control valves. The methods were tested using simulated data of three valves with different friction levels. The advantages of each one are emphasized. Copyright © 2007 IFAC Keywords: Parameter estimation, Control valves, Friction, Karnopp model. 1. INTRODUCTION Abnormal situations in process plant operations are commonly encountered and can occur due to: poor controller tuning, oscillatory disturbances, poor process and control system design and control valve friction (Choudhury et al., 2005). Control valves are the most employed actuators in chemical engineering control applications. However, as they are the only moving part of the loop, they are prone to friction, which causes oscillations or steady- state errors in the stem position. Such undesirable occurrences affect the overall profitability of the process (Kaiyhan and Doyle, 2000). Friction models can be used for detection and to deal with friction effects through model based compensa- tion (Armstrong-Hélouvry et al., 1994; Kaiyhan and Doyle, 2000). Several friction models were proposed in the litera- ture for different applications: from the models based on one (Stenman et al., 2003) or two parameters (Choudhury et al., 2005), to complex models with seven parameters (Armstrong-Hélouvry et al., 1994). Among them, a good compromise between relative simplicity and correct representation of most friction effects is the model proposed by Karnopp (1985). Nevertheless, as stated by Ravanbod-Shirazi and Besançon-Voda (2003), there are still insufficient methods for parameter estimation of the Karnopp model. In this paper, two promising methods for estimating the parameters of a Karnopp model, applied to a con- trol valve, are compared. The data set used in the analysis was derived from simulations. The main contributions of this paper are: [1] the ad- aptation of the method proposed by Ravanbod- Shirazi and Besançon-Voda (2003) for single action pneumatic actuators and [2] a comparison between this method and a novel analytic one proposed in Garcia (2007). The paper is organized as follows: the pneumatic actuator and the Karnopp friction model is presented in section 2. Two identification methods for control valves are briefly described in section 3. Both methods are used to identify the parameters of three valves with different friction levels on section 4. Finally, in section 5, conclusions are drawn. 2. THE VALVE MODEL A dynamic model for the control valve is given by the force balance equation: () () () () ext spring f mxt F t F t F t ⋅ = - +  (1) where: m: mass of the valve moving parts. x(t): stem position. Preprints Vol.2, June 6-8, 2007, Cancún, Mexico 303