Development and evaluation of control performance assessment indices for Alcoa’s advanced process control applications

摘要

This report details development and evaluation of potential performance measures for Advanced Process Control (APC) applications implemented across Alcoa sites. The final measure would ideally aid in the diagnosis of poor control and enable comparison between the performances of separate controllers.ududIn particular, the work has focused on the development of a suitable control performance index for Honeywell’s Robust Model Predictive Control Technology (RMPCT – Profit Controller) as implemented on an evaporator process located at Alcoa’s Kwinana alumina refinery.ududResearch in the field of controller performance assessment, particularly the performance of multivariate Model-based Predictive Controllers, was investigated. Existing performance indices proposed in the literature were assessed for their suitability to Alcoa’s applications. For the greater part, these methods are not suited to the specific characteristics and functionality of Honeywell RMPCT.ududA CPA metric entitled Event Frequency Performance Index (EFPI) is proposed in this report. It is a composite metric comprising five component metrics each of which are designed to gauge different aspects of RMPCT performance. Its stages of development are described and it is applied to seven periods of RMPCT historical data. The metric results are analysed and compared to general expectations about controller performance for these assessment periods in order to determine the utility of the proposed approach.ududA historical benchmarking method for performance assessment is also proposed. This involves the identification of a period of controller operation that is known to be good and then comparing subsequent assessment periods to this benchmark. This approach is applied to three different aspects of RMPCT performance: CV limit violation, MV movement and economic optimisation. Performance indices using this method are obtained for six periods of RMPCT historical data.