ZhuoHe Intrumentation HomePage > Technical article > >Comparison of two approaches to automated PI controller tunin

Comparison of two approaches to automated PI controller tunin

2016-05-23 13:34   Article Source:Unknown   View Times:
    Automated controller tuning is of significant in-terest to control engineers since it can lead to in-creased performance while saving time and energy. The original work on tuning focused on the development of procedures and recommendations to be used manually by operators. However, more sophisticated methods of autotuning have since been proposed. For example, automatic tuning is now used extensively in PID control.
   PID controllers have been the most commonly used controllers in industrial control practice for the past 60 years, even though great progress in control theory has been made over this period.This is because of their simple structure, ease of design and implementation, and transparent interpretation. A poorly tuned control system may waste energy and cause excessive and unnecessary wear of actuator. High performance is always the design target in industrial control applications and recently many modern control techniques that improve PID tuning have been reported .In this research, automated PI controller tuning tech-niques are studied for an industrial weigh belt feeder.
  The weigh belt feeder used in this research(see Fig. 1) was designed and manufactured by Mer-rick Industries, Inc. of Lynn Haven, Florida. It is a process feeder that is typically used in a food, chemical, or plastics manufacturing process. To ensure a constant feed rate in industrial operation, a PI control law is designed and implemented in the Merrick controller. In current practice the PI tuning process is performed manually by an engi-neering technician. However, for better and more consistent quality, it is desired to use automated PI tuning.

   The dynamics of the weigh belt feeder are domi-nated by the motor. To protect the motor, the con-trol signal is restricted to lie in the interval V. The motor also has significant friction. In addi-tion, the sensors exhibit significant quantization noise. Hence the weigh belt feeder exhibits non-linear behavior . The system nonlinearities make standard tuning methods difficult to apply. For example, in attempting to apply the Ziegler-Nichols tuning method to the weigh belt feeder, the system saturated before the ultimate gain was obtained. Relay feedback autotuning, another widely used method, does not perform well for highly nonlinear systems or systems with large disturbances. When relay feedback tuning was at-tempted, because of the sensor quantization noise and the motor friction, the desired square wave with symmetric positive and negative half cycles could not be achieved, even after considering the relay hysteresis and compensating for part of the load disturbance. There are many other methods of PID controller tuning in the literature, usually based on knowledge of the process parameters. Due to the nonlinearities of the weigh belt feeder, its process parameters change with time and set point. For example, friction is highly nonlinear and depends on multiple parameters that vary dur-ing the process @9#. In an effort to design PI con-trollers for this type of process, two non-model-based PI tuning methods were studied in this research: unfalsified PI control and fuzzy PI con-trol. Both methods do not require an explicit plant model, hence they are suitable for control design   for nonlinear plants that are difficult to model.
   The unfalsified design concept points out that the control law can be obtained directly from a set of candidate controllers by using stored sensor output signals and actuator input signals. Consequently, it is not necessary for a controller to actually be inserted in the feedback loop in order to be falsified. Thus the adaptive unfalsified con-trol processes may be significantly less susceptible to poor transient response than other processes that require inserting controllers in the loop. An-other particularly attractive feature of unfalsified control is the ability to find control laws that can meetmultiple objectives. For the weigh belt feeder, the objectives are to design PI controllers to minimize the transient and steady-state errors of the feed rate and to avoid actuator saturation at the
same time. To reduce the computational time used for the unfalsification, a genetic algorithm was adopted to perform the unfalsification.
   Fuzzy logic control provides a formal method-ology for representing, manipulating, and imple-menting a human’s heuristic knowledge about how to control a system. It has been found particularly useful for controller design when the plant model is unknown or difficult to develop. It does not need an exact process model and has been shown to be robust with respect to distur-bances, large uncertainty, and variations in the process behavior. Two types of fuzzy logic controllers~FLC’s!, PI-like FLC’s ~including gain scheduled and self-tuning PI-like FLC’s! and PI FLC’s, were designed for the weigh belt feeder. PI-like FLC’s do not have explicit propor-tional and integral gains; instead the control signal is directly deduced from the knowledge base and the fuzzy inference. In contrast, PI FLC’s are com-posed of the conventional PI control system in conjunction with a set of fuzzy rules~knowledge base!and a fuzzy reasoning mechanism to tune the PI gains online. Since the PI FLC’s outperform the PI-like FLC’s for the weigh belt feeder and they have explicit P and I gains, only PI FLC’s are compared with unfalsified PI controllers.
  Both unfalsified PI control and fuzzy logic PI control are effective solutions to systems with time-varying parameters and other uncertainties. In this paper, a detailed comparison of the two methods from the point of view of allowed design specifications, process knowledge requirements, computational requirements, controller develop-ment effort, transient performance and the ability to handle motor saturation is given. This compari-son can help a control engineer choose the most suitable type of controller for a given application.
The paper is organized as follows. Section 2 de-scribes unfalsified PI control design with genetic algorithm implementation for the weigh belt feeder. Section 3 describes fuzzy logic PI control
for the weigh belt feeder. Section 4 presents a de-tailed comparison of the two control design meth-ods. Finally, Section 5 presents some conclusions.

Respective topics:
  • Next:Product Feeders