2023 年 89 巻 925 号 p. 22-00063
Maintenance is one of the important issues in control systems. It is possible to take appropriate measures for maintenance of the control system by detecting change in characteristic of a controlled object. In this paper, three methods are proposed to detect characteristic change of the controlled object by applying the data-driven control FRIT. The method 1 detects the characteristic change of the controlled object based on a P gain value or an I gain value tuned by FRIT. The method 2 retunes parameter values of a reference model that describes desired response defined in FRIT so that the response is fastest within a range where a P gain value or an I gain value does not exceed each upper limit value. Then the characteristic change of the controlled object is detected based on the retuned values. The Method 3 detects the characteristic change of the controlled object by automatically tuning the parameter values of the reference model with maintaining current values for a P gain and an I gain. Performance of the proposed methods was evaluated by applying the methods to a prototyped small heater. An efficiency of the small heater was pseudo-deteriorated by covering a thermocouple with caps of various materials and thicknesses. A relation between the efficiency and the parameter values tuned by the each method at the pseudo-deteriorated condition was approximated by a linear function, and a correlation between the efficiency and the parameter values was evaluated with coefficient of determination. ζs were estimated from each characteristic change detection parameter values with the obtained linear functions, and the errors between the estimated value of ζ and the measured value of ζ were evaluated under two pseudo-deteriorated conditions. The coefficients of determination of the three methods were more than 0.95. The errors were smaller than 4%. It was confirmed that the proposed methods were effective for detecting the characteristic change of the prototyped small heater.
