Control Loop Performance Monitoring, Diagnosis And Improvement

Control performance monitoring has attracted great attention in both academia and indus-try over the past two decades. In this work, we analyzed the single input single output nonlinear control system and the multi-input multi-output linear control system’s performance by fully extracting the process information contained in routine operating data. Under the assumption of linear process and Gaussian noise, a performance monitoring index for multivariate control system is proposed by comparing the distribution dissimilarity of the process data. In situations where the above assumptions no longer stand, a probability distribution distance based index is proposed to monitor the performance of non-linear control systems. To locate the root cause of the poor control performance, control performance diagnosis technique is introduced after the control performance assessment and monitoring. On the basis of describing function analysis, a new stiction quantification method based on time and frequency domain criterions is proposed, and the existence conditions of limit cycles in cascade control loops were analyzed. To improve the control loop performance, the effect of controller tuning on stiction induced oscillation was discussed for different outer process model structures and controllers. These theoretical analy-sis can provide tuning guidelines for industrial engineers and operators. The main contents of this paper are summarized as follows:1. Most research efforts have been devoted to the performance monitoring of linear con-trol systems, without considering the pervasive nonlinearities (e.g. valve stiction) which present in most industrial control systems. In this work, a novel probability distribu-tion distance based index is proposed to monitor the performance of non-linear control systems. The proposed method uses Hellinger distance to evaluate change of control sys-tem performance. Several simulation examples are given to illustrate effectiveness of the proposed method. For nonlinear control systems or systems which are affected by non-Gaussian distributed noises, the traditional performance measure may fail when the first and second order statistics of the output are similar but higher order statistics change.2. Root cause analysis is the next step task when significant performance degradation is de-tected by the control performance monitoring technique. Valve stiction is one of the most common causes for poor performance in industrial control loops. Most of the current stiction estimation methods use time domain criterion, e.g. Mean Square Error, to jointly identify the stiction and process model parameters. However, stiction induced oscilla-tion is a phenomenon which has some specific characteristics in the frequency domain. Thus, extracting frequency domain information in the routine operation data will provide a more reliable and accurate stiction estimation. Under the framework of Hammerstein model identification and global optimization, a new stiction quantification method based on time and frequency domain criterions is proposed. The stiction parameters estima-tion, especially for the slip jump J, is improved significantly compared to the traditional methods.3. To provide suggested procedures to improve the control performance after locating the root cause of poor performance should always be the first priority of performance moni-toring methods. In this work, the existence conditions of limit cycles in cascade control loops were analyzed. The describing function analysis method was extended to predict behaviours of a cascade control system with a nonlinear element in control valve. The effect of controller tuning on stiction induced oscillation was discussed for different outer process model structures with two types of inner controller:PI and P-only. In terms of oscillation compensation, cascade control structure was compared with single feedback control structure and the results demonstrated that with the change of the control strate-gy and proper tuning of the controller, the cascade control loop can better attenuate the oscillation or can even remove the oscillation completely in comparison with a single feedback control loop.4. Multivariate control strategy especially model predictive control has been widely applied in the process industry in these days. A novel dissimilarity analysis based method is proposed to monitor the control performance of multi-input and multi-output systems. Compared with the minimum variance based index and the determinant based perfor-mance index, the proposed method considers not only the volume of the hyper-ellipsoid defined by the output data covariance matrix, but also the direction of the hyper-ellipsoid. The proposed approach detects changes in the orientation and volume of hyper-ellipsoids formed by the covariance matrices via analyzing the eigenvalues of transformed covari-ance matrices through KL transformation. Furthermore, a new performance index is used to quantify performance change of control systems. Simulation results from a numerical example, the Wood Berry distillation column example and pilot-scale experiment results all demonstrate the effectiveness of the proposed method.