Abstract
This work relates to the reliable generalized predictive control issues in the case when actuator or sensor failures take place. The experimental results that form the basis from which the conclusions are drawn from have been obtained in the position control of a servo drive task, and extend the results from the prior research of the author, dedicated to velocity control problems. On the basis of numerous experiments, it has been shown which configuration of prediction horizons is most advantageous from the control performance viewpoint in the adaptive generalized predictive control framework, to cope with the latter failures, and related to a minimum performance deterioration in comparison with the nominal, i.e., failure-free, case. This case study is the main novelty of the presented work, as the other papers available in the field rather focus on additional modifications of the predictive control framework, and not leaving possible room for optimization/alteration of prediction horizons’ values. The results are shown on the basis of the experiments conducted on the laboratory stand with the Modular Servo System of Inteco connected to a mechanical backlash module to cause actuator/sensor failure-like behavior, and with a magnetic brake module to show the performance in the case of an unexpected load.
Highlights
The paper extends the results presented in [6], where actuator/sensor failure has been considered in the case of velocity control to the position control task, leading to different results, as per varied impact of faults/failures on the behaviour of the control system
This paper extends the results presented in [5], where the generalized predictive control (GPC) controller has been implemented to work in real-time mode
The command input is fed to the servo drive from an input-output card used by Simulink Coder in order to work in a real-time regime
Summary
The considered experimental setup consists of a DC Buehler motor [5,6,21]: 12 V, 77 W, 250 mNm, speed 3000 rpm, current 4.7 A, the tachogenerator and the inertia load (brass cylinder, 2 kg, diameter 66 mm, length 68 mm), as shown in Figure 1 [22,23]. It is assumed that the ZOH-discretized model of this plant is taken into consideration when implementing the GPC algorithm with the sampling period of TS = 0.1 s, with voltage input and velocity output, and with the sum operator at the output exchanged for interior integration property, with the encoder output fed back to the controller. Having performed a series of experiments in a position control task (input u—voltage, output y—rotation angle of the shaft) with N = 10,000, initial estimates in RLS scheme close to 0, unity forgetting factor and the initial value of a diagonal covariance matrix set at 100 times the unity matrix, the results from Table 1 have been obtained. Nu , Ny output signal (rotation angle of the shaft) rotational velocity of the shaft calculated control signal applied/constrained control signal reference signal disturbance signal affecting model’s output time-shift operator (discrete-time domain) sampling period control and output prediction horizons
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have