The Study And Application Of Soft Closed-loop Fault-tolerant Control Approach Against Sensor Faults Based On Multi-model

With the increasing of the industrial system complexity, the requirements of system safety and reliability are becoming higher and higher. Meanwhile, the rich data reserve of the modern industrial systems and the rapid development of control technology also provide a certain security guarantee to improve the system safety and reliability. In view of this, the paper proposes mono-model and multi-model programs for sensor fault detection and soft closed-loop fault-tolerant control based on the data driven technology. The fact that the industrial systems are mostly nonlinear and sensors are extremely faults is considered in this paper. In addition, the corresponding simulation and experimental studies were carried out aiming at all kinds of sensor fault. Based on the proposed plan, the following works have been done in the paper:1) The problem of sensor fault detection and soft closed-loop fault-tolerant control was studied based on mono-model.Considering some problem, such as nonlinear coupling, time-varying, retardation existing in industrial control systems, difficult establishment of a precise mechanism model, sensor fault and propagation effect in the closed-loop system which can lead to declining the system performance and so on, mono-model sensor fault detection and soft closed-loop fault-tolerant control was studied based on data driven. Firstly, based on the data-driven approaches, the RBF, LS-SVM and KPLS prediction models of nonlinear systems are established using the historical data of process operation, and the residual error signals are obtained via the difference between predicted mono-model output and the actual sensor output. Secondly, the historical and current residual error signals are analyzed for fault detection based on the SPRT algorithm. When a sensor fault is detected, the predicted output of the model is used to replace of the actual output of the physical sensors so as to achieve fault-tolerant control of sensor faults by the mono-model prediction soft closed-loop. The results of the simulation experiment show that the sensor faults can be detected reliably through combination of the data-driven prediction models and SPRT, and safe fault-tolerant control for sensor faults can be achieved with the help of the soft closed-loop switch during the course of the existing fault.2) The problem of sensor fault detection and soft closed-loop fault-tolerant control was studied based on Multi-model.Considering some problem, such as the limitations and applicability of method of mono-modeling, the unreliable results of fault detection and soft closed-loop fault-tolerant control due to the effect of modeling error for the real system and so on. the multi-model sensor fault detection and soft closed-loop fault-tolerant control was studied based on data driven. Firstly, based on the data-driven approaches, the RBF, LS-SVM and KPLS prediction models of nonlinear systems are established, and constructing three residual error signals combing the actual sensor output. Secondly, the SPRT algorithm is used to detect the fault, wherein, two or more channels of residual error signals above the threshold value are determined to the occurrence of a sensor fault. When a sensor fault occurs, the optimal integration of the output values of multiple predicted models are used to replace the actual sensor output, thus achieving tolerant control of the sensor faults via soft closed-loop. The results of the simulation experiment show that the multi-model method has higher reliability in fault detection and the fault-tolerant control performance is better.3) The experimental of sensor fault detection and soft closed-loop fault-tolerant control was studied based on PCS platform.Considering the differences between the industrial actual systems and the simulation systems of the models, and in order to ensure the engineering availability of the fault detection and fault-tolerant control algorithms, the experiment platform for sensor fault detection and fault tolerant control is built up through PCS equipment of our school advanced control laboratory. The real ultrasonic sensor fault detection and fault tolerant control in the liquid level control system is studied based on the platform. The experimental results show that various types of sensor faults in PCS could be detected in a reliable and timely manner through combining the data-driven predictive models and SPRT, and safe fault-tolerant control of sensor faults can be achieved by soft closed-loops with predicted model. Compared with the mono-model approach, the multi-model approach method have more reliable detection result, high safety and performance of fault-tolerant, which reveal the method have strong applicability in project engineering.