Limits of control performance: a model-based approach

基本控制已经成熟到许多先进的控制算法现在是大多数DCS系统中的标准产品的程度。对于许多关键回路，期望在给定特定类型的控制策略（例如比例、积分和微分（PID）或基于模型的预测控制（MPC））的情况下获得最高水平的控制性能。控制性能的这种基本限制是由于特定的过程，执行器和传感器的特性，并可以表示在由于波德的积分定理的灵敏度和互补的灵敏度函数的界限。了解这些类型的因素及其对性能的影响，可以达到一个“精心设计”的控制器是重要的因素，可以最容易地描述，如果过程模型是已知的。因此，本研究建议的具体重点将是探索控制性能的限制，从基于模型的角度来看，不同于基于历史数据的角度。基于数据的性能监控方法现在已经很好地建立起来，其重点是从常规过程数据计算性能指标，并将其与基准进行比较，例如最小方差基准或其他使用指定的基准。 最佳可实现控制的主题给出了一些知识的过程动力学和控制器类型，例如PI（D）控制器，将是本研究建议的主要焦点。超过90%的控制器在工业中使用的（PID）类型，但问题的最佳可实现PI（D）控制给定类的一阶加时滞或二阶加时滞过程是一个开放的问题。这个建议是探讨这些基本的限制PI（D）控制器时，适用于最常见的工业过程。

Basic control has matured to the point where many advanced control algorithms are now standard products in most DCS systems. For many critical loops it is desired to obtain the highest level of control performance given a specific type of control strategy, e.g. Proportional, Integral and Derivative (PID) or Model-based Predictive Control (MPC). Such fundamental limits to control performance are due to specific process, actuator and sensor characteristics and can be expressed in terms of the bounds due to Bode's integral theorem on the sensitivity and complementary sensitivity functions. An understanding of these types of factors and their impact on the performance that can be attained with a 'well designed' controller are important factors and can most easily be described if the process model is known.Therefore the specific focus of this research proposal will be to explore the limits of control performance from a model-based perspective, unlike the historical data-based perspective. The data-based performance monitoring approach is now well established and its focus has been on computing performance metrics from routine process data and comparing them to a benchmark, for example the minimum variance benchmark or other used-specified benchmarks.  The topic of the best achievable control given some knowledge of the process dynamics and controller type, for example a PI(D) controller, will be the main focus of this research proposal.Over 90% of the controllers used in industry are (PID) type and yet the issue of the best achievable PI(D) control for a given class of first order plus time-delay or a second order plus time-delay process is an open problem. This proposal is to explore these fundamental limitations of PI(D) controllers when applied to the most common industrial processes.