Ivestigatin Of Fault Detection And Insulation Of Sensors Commonly Used On Chemical Industry Static Equipment

There are many sensors located in chemical production process, which are used for monitoring and controlling the production process. Once the sensor failure occurs and the failure can not be solved in time, the wrong decision of device state will be caused and normal production is disturbed, also may induce a heavy accident. Therefore it is necessary to carry out the study on sensor failure detection, it also will help to realize failure detecting automation of device.The paper carries on the study on failures of sensors correlating with static chemical device. Firstly, the type and features of common used sensors are analyzed, and considering the running character of chemical equipments the sensor signal features are concluded. The sensor failure types is classified, and the failure model is also build. The sensor failure detecting method of combining single sensor failure detection with that of multiple sensors is presented based on the study of the domestic and foreign failure detecting ways.Single sensor failure detection is mainly through setting up output model of sensors, and the residual of output signal with predicting signal from sensing model will be compared with threshold to judge an occur of failure, if there is a real failure, the predict signal will take the place of actual failure signal to realize failure separation. The dynamic neural network is adapted to model a sensor failure on two cases. One is to set up a failure model with only the previous output the sensor, and the other one is to use both the sensor's previous output and the signal from adjacent sensor. These models can detect the deviation failure, broken circuit failure and the drift failure through simulation. The Support Vector Machine has better generalization ability, therefore it is also used to setup sensor prediction model, and these models can obtain good failure detection.Since chemical engineering is a dynamic stable process, the signals of different monitoring position have a certain correlation, which can be used to set up a model of the sensor system. When failure occurs on one of these sensors, the change of the model will beyond predefined threshold, thus the failure can be determined.Simulations is carried out through Matlab software combined with Support Vector Machine toolbox and neural network toolbox, which show that the proposed method can effectively fulfill the common sensors of chemical equipments fault detection and isolation...