The Study Of Error Compensation And Fault-tolerant Control Approaches Against Sensor Faults Based On Data-driven

With the development of modern industrial, production process is becoming larger and more complex, and people’s requirement for the safety and reliability of the system is improved. Because the rapid development of technology, t oday’s industrial s ystems store a large amount of data, which provides a certain amount of data redundancy to enhance the systems’ s security and reliability. In view of this, considering the actual industrial most are nonlinear systems and the sensor is prone to failure, based on datadriven technique, the error compensation based on the inverse and the soft redundancy fault-tolerant control method based on the reliability evaluation of fault detection are proposed respectively for sensor known deterministic faults and unk nown uncertain fault, and the corresponding simulation and semi physical experiment research for the constant bias and precision down common sensor fault is done. Work includes:1) Research on modeling methods of nonlinear neural network and Wiener model based on data-drivenWhen the sensors in industrial systems are failure, the known deterministic fault error most are the static nonlinear characteristics, but industrial objects are mostly expressed as the nonlinear dynamic characteristics. Traditional mechanism modeling methods are difficult to express and implement the corresponding compensation. For two different nonlinear characteristics, combining the respective advantages of neural network and Wiener model, an improved LM-BP neural network and RBF neural network are used as static nonlinear modeling to compensate known faults nonlinear modeling. And then, hybrid modeling methods of Wiener and neural network is proposed for system process to do the research of dynamic nonlinear modeling, laying the foundation for the unknown sensor fault tolerance.2) The research on the error compensation of the known fault sensor based on LM-BP and RBF neural networkConsidering static, deterministic characteristic of the known fault error of the sensor and neural network has good static nonlinear approximation capability, two methods of fault error compensation scheme based on series and parallel inverse are proposed. First, fault sensor and normal sensor output value are collected with the sensing technology, and based on the Levenberg Marquardt backpropagation and radial basis function(RBF) modeling technique, the fault sensor output and normal sensor output, sensor fault and its normal sensor difference between string and parallel inverse loop to compensate for the nonlinear mapping relationship are established respectively, with series parallel connection mode, introduced to the system to compensate the sensor fault deterministic error. It shows that two methods have better fitting precision, semi physical experiment results show that the fault error compensation has the value of the actual usability engineering.3)The research on fault detection, the fault tolerant control method s of the unknown fault of the sensor based on the reliability evaluationConsidering dynamic, uncertain characteristic of unknown fault error and the problem of sensor fault-tolerant security because that it has no fault detection reliability evaluation, putting forward the soft redundancy fault-tolerance control method based on NN-Wiener model of fault detection reliability evaluation function. Firstly, the scheme obtains control quantity of object and the sensor data through simulation and PCS level control experimental platform. Then the simulation and experiment of dynamic NN-Wiener model of the object are established based on these data, and the fault detection module and software redundancy fault-tolerant module are embedded in the simulation and the PCS liquid level control system platform. simulation and experimental results show that the combination between NN-Wiener prediction model and SPRT algorithm can detect reliably all kinds of failures of sensors in PCS, based on fault detection reliability evaluation combined with hard and soft feedback can be smooth handoff, and the fault sensors to achieve security fault tolerance. It also revealed the methods of usability engineering.