Research On The Method Of Mechanical Fault Diagnosis Of High Circuit Breaker Based On Control Loop Detection Of Operating Mechanism

The high-voltage circuit breaker is used to cut off the fault electricity to isolate the circuit equipment when the power system breaks down,so as to avoid the further expansion of the accident to affect the industrial production and residents' life.The traditional fault diagnosis of circuit breaker has some problems,such as complex process,slow training speed and low accuracy.There are pretty problems in regular maintenance,such as time-consuming,frequent operation and too much disassembly,which will reduce the reliability of the circuit breaker.High voltage circuit breaker contains rich vibration signals in the process of mechanical operation.Vibration signals' characteristic value is stable and reliable.Effective feature vector can be extracted from the vibration signal,so as to accurately determine whether the circuit breaker has fault and the type of fault.Many scholars have benefited from the advantages of convolutional neural network,such as strong local sensing ability and common weights,to speed up the training process.In this paper,the convolution neural network is used to extract the characteristics of vibration signals of circuit breakers.Firstly,convolution operation is carried out by preset step sliding filter.Convolution operation can traverse the whole sample to obtain fault information in the sample.Then the fault information is passed to the pooling layer.Nonlinear activation in the pool layer is helpful to reduce the number of data points and avoid over fitting,so as to speed up the training.It also acts as a smooth process from which unnecessary noise can be eliminated.Finally,the obtained local information is transferred to the full connection layer to connect all previous feature maps and obtain the complete feature vector.The hybrid model of convolution neural network and long-short term memory neural network established in this paper gives full play to the advantages of the two models,which can not only improve the training speed,but also ensure high diagnosis accuracy.The method based on shallow neural network,such as support vector machine(SVM),whose training speed is slow,that can not meet the need of high-voltage circuit breaker real-time monitoring.In view of the three most common mechanical faults of vacuum circuit breaker,such as tripping closing electromagnet jam,the principle axles jam and the half shaft jam,several experiments are tested respectively.The fault diagnosis of the same vibration signal samples of high voltage vacuum circuit breaker is carried out by the proposed method.The results show that the proposed method has better diagnosis result in Receiver Operating Characteristic Curve(ROC),Precision-Recall Curve(PRC),Kolmogorov-Smirnov Curve(KS),Cumulative Gain Curve(CG)and Lift Curve,which verifies the possibility of this method.