Yoshihiro Ohnishi

Design of a performance-adaptive proportional—integral—derivative controller for stochastic systems

In order to manufacture high-quality products it is necessary to regularly monitor the performance of the control loops that regulate the quality variables of interest. This paper describes a design scheme of performance-adaptive controllers which are based on the above control strategy. According to the proposed control scheme, the output prediction error is monitored regularly and system identification is initiated if this error exceeds a user-defined threshold. Subsequently proportional—integral—derivative (PID) parameters are updated for the new model. Optimal PID parameters are calculated based on the linear quadratic Gaussian (LQG) trade-off curve obtained for the reidentified process model. The behaviour of the proposed control scheme is numerically evaluated by some simulation examples.

Design of a PID Controller with a Performance-Driven Adaptive Mechanism

In this paper, a new design scheme of performance-driven PID controllers whose PID parameters are adjusted based on a control performance criterion. Although a majority of studies have been focused on the derivation of the CPM index, the control parameter tuning method based on the CPM has been hardly studied. Conventional self-tuning controllers are tuned based on the variance of control errors and/or modeling errors. Few adaptive schemes use performance indice as tuning signals, which should be the main driving force in maintaining optimal operation, This paper develops a strategy for the tuning of an adaptive PID controller that is an approximation of a generalized minimum variance controller. The main driving signal for adaptive tuning is the degradation of the controller performance criterion. The effectiveness of the proposed method is numerically evaluated on two simulation examples.

Design of a CMAC-Based PID Controller Using Operating Data

In industrial processes, PID control strategy is still applied in a lot of plants. However, real process systems are nonlinear, thus it is difficult to obtain the desired control performance using fixed PID parameters. Cerebellar model articulation controller (CMAC) is attractive as an artificial neural network in designing control systems for nonlinear systems. The learning cost is drastically reduced when compared with other multi-layered neural networks. On the other hand, theories which directly calculate control parameters without system parameters represented by Virtual Reference Feedback Tuning (VRFT) or Fictitious Reference Iterative Tuning (FRIT) have received much attention in the last few years. These methods can calculate control parameters using closed-loop data and are expected to reduce time and economic costs. In this paper, an offline-learning scheme of CMAC is newly proposed. According to the proposed scheme, CMAC is able to learn PID parameters by using a set of closed-loop data. The effectiveness of the proposed method is evaluated by a numerical example.

PERFORMANCE-DRIVEN ADAPTIVE PID CONTROLLER DESIGN : THEORY AND EXPERIMENTAL EVALUATION

Abstract This paper proposes an adaptive PID controller which is driven by current control performance. The objective of the proposed scheme is to carry out the retuning of PID parameters and system identification only when controller performance deteriorates below a user-specified limit. Experimental evaluations on a pilot-scale process demonstrates the practicality and utility of this idea.

A Design of Nonlinear PID Controller with Neural net based FRIT

Some design schemes of model-free controllers which do not require any system models have been considered in the last decade. FRIT(Fictitious Reference Iterative Tuning) method that directly computes the control parameters from the operating data have been proposed as the one of model-free controllers. FRIT has some useful practical features. One is that it does not require system identification. Another is that the control parameters can be directly computed using only a set of closed loop input/output data and the desired output signal. The calculations of the control parameters needs the optimization of the cost functions. The ordinary approach is the gradient method. However, this calculations derives only linear parameters. Therefore, the applications of FRIT are limited for linear systems. In this paper, a new approach to the discrete FRIT-based nonlinear PID control is proposed. The neural network is utilized for the optimization of FRIT. PID parameters are adequately adjusted corresponding to the nonlinear properties. The conventional schemes by using the neural networks require the information of system Jacobian to update weighting factors. This proposed method can calculate the control parameters without the information of system Jacobian or system parameters except for the information about the time-delay.

Design of Performance-Driven Adaptive PID Controller

This paper proposes an adaptive PID controller which is driven by current control performance. The objective of the proposed scheme is to carry out the retuning of PID parameters and system identification only when controller performance deteriorates below a user-specified limit. Numerical evaluations demonstrates the practically and utility of this idea.

A practical control performance index and PID controller design

In the field of process control, there are cases where the control performance becomes inferior due to the change of surrounding environment. On the other hand, the best control performance is required from the viewpoint of high-quality and energy saving. Therefore, it is important to monitor the process state and evaluate the control performance. The scheme based on minimum variance control (MV-Index) has been proposed to evaluate control performance for deadbeat system. In the deadbeat system, evaluation is considered only step reference value. Therefore, in this paper, the evaluation is considered the reference trajectory specified by user. The effectiveness of the proposed method is verified by a simulation example.

Frequency analysis of control error signal based on minimum variance control

The idea of controller performance assessment is becoming very important in the process control area. One of the main performance monitoring index is based on the minimum variance control benchmark proposed by Harris. However, the low value of this index means only low performance of the controller. So, it is difficult to tune the controlled parameters based on this index. This paper considers the matching of each PID parameters based on frequency analysis of minimum variance output.

Performance-assessment of a weigh feeder using steady-state predictive output

The present study discusses a design method for controlling a weigh feeder. In the present study, a designed control system is assessed, and the control parameters are updated such that the control performance is improved. The proportional-integral (PI) control law is used to obtain a practical control system in industry. In the control system of a weigh feeder, since the steady-state response is more important than the transient response, the PI parameters of the PI control law are decided based on generalized minimum variance control with steady-state prediction (GMVCS). In the design of GMVCS, steady-state prediction is used instead of dead-time ahead prediction, and hence the steady-state control performance is superior to that of conventional methods. Finally, the effectiveness of the proposed method is demonstrated through experimental results.

Design of Neural Networks Based FRIT PID Controllers and Its Applications

Abstract Some design schemes of model-free controllers which do not require any system models have been considered in the last decade. FRIT(Fictitious Reference Iterative Tuning) method that directly computes the control parameters from the operating data have been proposed as the one of model-free controllers. FRIT has some useful practical features. One is that it does not require system identification. Another is that the control parameters can be directly computed using only a set of closed loop input/output data and the desired output signal. The calculations of the control parameters needs the optimization of the cost functions. The ordinary approach is the gradient method. However, this calculations derives only linear parameters. Therefore, the applications of FRIT are limited for linear systems. In this paper, a new approach to the discrete FRIT-based nonlinear PID control is proposed. The neural network is utilized for the optimization of FRIT. PID parameters are adequately adjusted corresponding to the nonlinear properties. The conventional schemes by using the neural networks require the information of system Jacobian to update weighting factors. This proposed method can calculate the control parameters without the information of system Jacobian or system parameters except for the information about the time-delay.