Research On Sensor Fault Diagnosis Method Based On Particle Swarm Optimization Probabilistic Neural Network

As an important part of data collection system,sensor is available in widespread applications.However,most sensors are prone to failure due to the poor working environment.The accuracy and reliability of sensor measurement signal is the key factor for the normal operation of control system.Therefore,a probabilistic neural network model based on particle swarm optimization is established in this paper for fault diagnosis of a single sensor.In this thesis,the basic principle of the sensor fault is firstly introduced.Seven kinds of sensor failure models were established in view of the three common kinds of the sensor faults and considering the multiple faults occurred at the same time.Secondly,the fault injection method was employed to simulate the sensor fault.The acquisition platform for sensor fault signal is built based on the laboratory to collect the sensor fault signal.Thirdly,taking the advantage of wavelet transform in handling the global characteristics of non-stationary signals and multiscale permutation entropy(MPE)in representing the local characteristics of signals at multiple scales,the method of wavelet transform and multiscale permutation entropy was integrated to extract the fault characteristics of sensors.Finally,the multiscale permutation entropy of the six scales was taken as the characteristic value,and the fault type of the sensor was taken as the output.The probabilistic neural network(PNN)model and the PSO-PNN model optimized by particle swarm optimization(PSO)were established.Both models were trained by the signals collected from the acquisition platform,and the prediction effects of the two models were compared.The experimental results show that the prediction accuracy of PSO-PNN was significantly higher than that of PNN,reaching 96.67%,which reflects the effectiveness of PSO-PNN for sensor fault diagnosis.