CONTROL PERFORMANCE EVALUATION AND DIAGNOSIS

Abstract

<p>A working definition of "good" control performance was considered and the criteria for its measurement were then determined. This study is focused on a time series approach to control performance evaluation and diagnosis, in which statistical tools such as the auto correlation and cross correlation functions and the power spectrum, as well as the input and output variances are used. This technique allows the use of normal operating data for control system performance evaluation, thus requiring minimal effort.</p> <p>Several simulation and industrial cases were investigated for this research, including SISO feedback and feedforward-feedback strategies as well as MIMO applications. The basis of this approach is the comparison of the existing controller statistical properties to that of a theoretical optimum. In this manner, it is possible to ascertain whether the potential for controller improvement exists and is warranted. The diagnostic procedure then allows for the determination of the likely cause of inadequate control performance, with the ability to distinguish between poor tuning and model mismatch and between poor feedforward or feedback control or the specific controlled or manipulated variable in a MIMO system.</p>