Nonlinear Sensor Fault Diagnosis Method Based On Deterministic Learning

In the field of automatic control,sensors are the main devices for information acquisition.With the continuous improvement of the level of automation technology,many large-scale automation projects are increasing and various sensors for parameter measurement and state control are also changing.It should not be overlooked that these sensors consisting of precision components are often in a harsh working environment,and failure is inevitable.Once the sensor is degraded,faulted or failed,it will have serious impact on subsequent monitoring,control,fault diagnosis and other systems,resulting in misdiagnosis,false alarms,and even incalculable losses.Therefore,there is a very important direction for the research of sensor fault diagnosis.In this paper,a nonlinear sensor fault diagnosis method based on adaptive learning and neural networks is proposed.In the process of sensor fault diagnosis,the following steps are included: Establish a single output uniform energy standard model for the system.Normal Pattern learning training and state observation.For the training system to conduct learning training and state observation during normal operation,the learning training method adopts Lyapunov-based learning method and realizes the convergence of RBF neural network and the convergence of RBF neural network according to the theory of adaptive learning.The internal dynamic approach is approached in fault mode.Using high-gain observer with neural network for learning training and state observation.Using high-gain observer of neural network for second learning training and state observation.Learning training and state estimation of failure mode.Establish a pattern library including normal mode and failure mode.Establish a dynamic estimator.Compare the state vector in the dynamic estimator with the state vector of the system under test,and construct the residual for the residual differential evaluation for fault isolation.The method is used for fault diagnosis of relatively complex nonlinear systems,and can learn and quickly simulate unknown system modes,thereby performing fault finding and approximation.