Research On Fault Diagnosis Method Of Spring Operating Mechanism Circuit Breakers Based On Feature Extraction By Time Domain Segmentation

The operation state of high voltage circuit breakers(HVCBs)is related to the reliability of power grid directly.Safety and reliability of HVCBs operating is an important factor to ensure the reliable operation.When the power system is in normal operation,HVCBs control power system and switching operation of high voltage equipment.When the system fails,the short-circuit current and overload current in the high-voltage line can be cut off by relay protection device to prevent the fault scope from widening.Failure of high voltage circuit breaker in normal operation will cause very serious harm to the power system,even cause power supply interruption,and cause serious losses to the national economy.By analyzing the vibration signal produced by the action of HVCBs,many mechanical faults such as insufficient energy storage of spring and loosening of screw can be found.Therefore,fault diagnosis of HVCBs based on vibration signal is of great significance.To improve the efficiency of feature extraction for mechanical vibration signals of HVCBs,in order to obtain the mechanical condition,an efficient feature extraction technique is used to analyze and process the vibration signals of HVCBs.Its main research contents include signal acquisition,feature extraction,feature analysis and state recognition.Firstly,a reasonable vibration signal acquisition system platform is built to collect vibration signals under different states of HVCBs.After that,although the existing features can accurately describe different state signals effectively.However,the frequency domain distribution of fault features of HVCBs vibration signals is widespread and the actual work is affected by installation and so on.It is difficult to extract relevant features from specific frequency domain.Because there are differences in the amplitude,attenuation degree and vibration start time of different fault state signals in circuit breaker vibration signals.Therefore,time domain features can be extracted directly from the original vibration signals to analyze the fault state of HVCBs.Then,in order to improve the quality of features,aiming at the subjectivity and blindness of the original feature set,the weights of extracted features are evaluated by using Gini importance degree and feature evaluation index based on scatter matrix.After analyzing the relationship between features and categories,according to combination effect of features,selecting the scale of feature set and experimenting with several groups of test samples.It is proved that the feature dimension reduction method adopted in this paper can effectively reduce the computational complexity and improve the accuracy of fault diagnosis.Finally,a hierarchical hybrid classifier combining single-class classifier and multi-class classifier is constructed.Firstly,normal and fault states are distinguished by single-class support vector machine;if fault states are identified,random forest is used to identify fault types,and then single-class support vector machine is used to correct random forest recognition results.Experiments show that feature extraction based on time domain segmentation of original signals is efficient,feature selection can reduce the computational complexity,and hierarchical hybrid classifier can effectively identify unknown faults without training samples and has high recognition accuracy.