(598i) Decentralized Multi-Loop PID Controller Design Directly From Plant Data

Proportional-integral-derivative (PID) controllers are the most widely used controllers in the chemical process industries. As a result, PID controller design methods remain to be an active research topic in the past several decades. In particular, various model-based PID design methods can be found in the literature. There are two steps in the model-based PID designs: an empirical (low-order) model of the process is identified first, which is subsequently used together with a prespecified tuning algorithm to design a PID controller. Although these methods can give good PID design when the underlying process dynamics are reasonably described by the low-order models, the effectiveness of these methods would degrade for higher-order process dynamics owing to the inevitable modeling error. In addition, it is often difficult to collect input-output data for the identification of a controlled plant that has already been set up and is in full operation.

To alleviate these drawbacks, it is an attractive alternative to design PID controllers directly based on a set of process input and output data without resorting to a process model. Toward this end, several mode-free or data-based controller design methods were developed in the literature, such as the iterative feedback tuning (IFT) method, the virtual reference feedback tuning (VRFT) method, the fictitious reference iterative tuning (FRIT) method, and their variants. However, these previous works only discussed the PID controller design for single-loop feedback systems, and no application of the data-based methods to the controller design for multi-loop control systems is reported in the literature. In fact, multivariable systems are frequently encountered in the chemical process industries, and multi-loop PID controllers are still much more favored in most commercial process controls. Application of the data-based controller design method in multi-loop control systems is even more attractive, because the model identification for multivariable systems is more complicated and time-consuming than that for single-loop systems. This motivates our research to extend the data-based controller design method to multi-loop control systems with the specific aim of designing decentralized PID controllers directly from the plant data collected under open-loop or closed-loop operation. Thus, this method can design multi-loop PID controllers without resorting to the availability of process models.

The proposed direct PID controller design approximately solves a model-reference problem, and the design goal of the proposed method is to obtain PID parameters such that the corresponding multi-loop control system behaves as closely as possible to the prespecified reference models for each equivalent single loop. The selection of the reference models considers an appropriate tradeoff between the bandwidth of main loop and the resonant peaks of interaction loops. The optimization problems pertaining to the proposed design are derived, and the associated design issues are addressed. Extensive simulation results show that the proposed multi-loop PID design gives better or comparable control performance than those attained by the model-based PID designs. Consequently, the proposed design is an attractive alternative to the model-based decentralized PID design methods.