An online detection method for the mechanical fault of circuit breakers of rural distribution lines based on continuous wavelet transform and vibration signal characteristics

Rural distribution lines are widely distributed and dispersed, involving numerous circuit breaker equipment. Detecting mechanical faults on each circuit breaker requires significant time and human resources. It can bring significant interference and power outage risks to the operation of the line. Therefore, an online detection method is studied for mechanical faults in rural distribution line circuit breakers. Continuous wavelet transform technology is used to extract the characteristics of circuit breaker vibration signals. This feature is used to locate the faulty section. The PNN neural network is introduced to identify fault feature samples and achieve online detection of mechanical faults. To test the application performance of this method, a comparative experiment is designed. The results verified that when using this method to extract the vibration signal of circuit breaker mechanical faults, the error is low, and the fault detection accuracy is high.


Introduction
Rural power distribution lines are an important link in rural power supply, and circuit breakers, as protection equipment, play a crucial role in the operation of rural power grids [1].Rural areas have complex and variable climates and harsh working environments, which pose many challenges and challenges to the circuit breakers of rural distribution lines.Firstly, adverse weather conditions may cause the mechanical components of circuit breakers to work in extreme temperatures, humidity, and polluted environments for a long time, increasing the wear and tear of the mechanical components.For example, high temperatures can cause expansion and deformation of internal components in circuit breakers.In contrast, low temperatures can easily cause condensation and freezing, affecting the flexibility and normal operation of mechanical components.Secondly, environmental factors such as dust, humidity, and corrosive gases in rural areas can also cause significant damage to the mechanical components of circuit breakers.A large amount of dust and humidity will adhere to the mechanical components of the circuit breaker, gradually accumulating during operation, leading to poor contact, increased contact resistance, or high-temperature problems.At the same time, corrosive gases can corrode metal parts, causing corrosion and damage to mechanical components of circuit breakers, and posing greater challenges to detecting mechanical faults [2][3].At this stage, to detect the mechanical fault of circuit breakers, some good research results have been proposed in related fields, such as Sun et al. [4] extracted the joint cepstrum coefficient of the circuit breaker closing sound signal as the sound feature vector according to the human ear hearing characteristics, and then used the sparse representation classification algorithm to identify the feature vector, and introduced the divergence concept of linear discriminant analysis into the sparse representation classification objective function to improve the score class performance.Cao et al. [5] combined convolutional neural networks with long and short memory networks to establish a hybrid deep network.On this basis, a signal processing method based on a convolutional neural network is proposed.It uses a convolutional neural network to process the collected signal.It integrates it with the local time domain characteristics of the gated loop element to realize highly sensitive signal detection.
Based on the above research, this paper presents an online detection method for the mechanical fault of circuit breakers of rural distribution lines based on continuous wavelet transform and vibration signal characteristics.

Extraction of circuit breaker vibration signal features and fault section location
We take the transformation decomposition process of a group of vibration signals as an example [6] as follows: where ∋ ( In the formula, signal ∋ ( r ψ is taken as the original signal, and feature extraction is carried out through continuous wavelet transform.The frequency band where the fault feature of the low-frequency component is located is the same as the frequency band where the fault feature of the high-frequency component is located.The lower corner marks "1", "2" and "3" are the number of layers of the continuous wavelet transform, which is a total of 3 layers.The fault extraction process of the vibration signal is summarized as follows: A class of vibration signal features will be decomposed in each continuous wavelet transform process, and the feature relationship can be obtained through the continuous wavelet transform.Formula (2) is derived into Formula (3).For the convenience of subsequent calculation, the signal features are only divided into two categories: low frequency and high frequency.Based on this, the phase space of different signal features is described.
Figure 1 shows the procedure for locating fault segments.

Fault detection model based on PNN neural network
We set the fault mode representation as / ab ββ and the fault characteristic sample is described as

∋ (
12 , ,..., m Xx xx < , then: In the formula, ab  , represent the cost factor, and a  is to mistakenly divide the fault feature sample X belonging to a β to b β , and the opposite is true.G represents the prior probability of H, respectively.
We take the mechanical fault feature vectors selected above as training and test vector sets [7][8].The structure of the PNN fault detection model is shown in Figure 2. The second layer is the hidden layer, in which each neuron node corresponds to a mechanical fault feature sample [9][10].The distance between the input vector λ and the center of each neuron node is calculated to obtain the j neuron of the class i mechanical fault feature sample in this layer, to determine the relationship between input and output.

∋ ( ∋ (
where ij λ represents the j center of the type I mechanical fault feature sample; δ represents the smoothing parameter; ij ξ stands for its output; m is used to represent the number of training samples for failure modes.The third layer is the summation layer, and in the model, there is a one-to-one correspondence between neurons and mechanical fault categories of circuit breakers.This layer will output a weighted average of neurons belonging to the same hidden layer, as shown below: where ϕ represents the number of class I neurons; i ϖ stands for its corresponding output.

Preparation for experiment
The rural distribution line system simulation model was built in EMTDC to verify the superiority of the proposed method of locating rural distribution line fault segments based on empirical wavelet transform.The simulation model is shown in Figure 3.The system is a 10 kV hybrid line distribution network, and distributed power is connected to Line 2, Line 3, and bus, respectively, with a capacity of 0.5 MW and a permeability of 16.5%.The three lines are 16 km, 18 km, 20 km, and 10 km.Each line is provided with a mechanical fault point of the circuit breaker.Table 1 shows the six typical sets of fault parameter settings.
Table 1.Fault parameters of rural distribution lines.

Fault section location analysis
According to the above experiment preparation and parameter setting, the fault section is under different conditions.Due to a small amount of noise in the collected power signal and the decision index of the running state, the actual signal will be distorted.This project aims to introduce multiple noise levels into the existing power system and test the adaptive ability and accuracy of the proposed multi-source power system.The positioning results are shown in Table 2.  4 that under different load fluctuation conditions, the positioning accuracy of the proposed method is much higher than that of the comparison method, and the average accuracy is always above 70%.The comparison of experimental results shows that the proposed method can accurately locate the fault section of the distribution network and can effectively improve the location accuracy of different fault types.According to Figure 5, the proposed method can effectively extract the circuit breaker vibration signal characteristics when the circuit breaker is in normal operation and mechanical fault.The characteristics of the extracted circuit breaker vibration signal can effectively present the changing trend of the circuit breaker vibration signal.

Conclusion
Based on continuous wavelet transform and vibration signal characteristics, the online detection method of breaker mechanical fault in rural power distribution lines has important research significance and broad application prospects in the rural power distribution field.This method can use continuous wavelet transform to extract rich time-frequency characteristics from circuit breaker vibration signals and more accurately describe the characteristics of vibration signals.Through this feature, it is possible to achieve online detection of mechanical faults to detect and diagnose potential problems in advance.
signal.k represents the frequency band to which the feature belongs.r represents a high-pass filter.y is a low-pass filter.t represents coefficient.In the application process, continuous wavelet transform must focus on the whole frequency band and decompose the fault diagnosis features carried in the original signal to the frequency band as follows:

Figure 1 .
Figure 1.Fault section location steps based on continuous wavelet transform.

Figure 2 .
Figure 2. PNN fault detection model structure.The first layer of this structure serves as the input layer and can receive vibration signals from circuit breakers.The second layer is the hidden layer, in which each neuron node corresponds to a mechanical fault

Figure 3 .
Figure 3. Simulation model of rural distribution lines.The system is a 10 kV hybrid line distribution network, and distributed power is connected to Line 2, Line 3, and bus, respectively, with a capacity of 0.5 MW and a permeability of 16.5%.The three lines are 16 km, 18 km, 20 km, and 10 km.Each line is provided with a mechanical fault point of the circuit breaker.Table1shows the six typical sets of fault parameter settings.Table1.Fault parameters of rural distribution lines.

Figure 5 .
Feature extraction results of circuit breaker vibration signal.(a) Vibration signal in normal operating state; (b) Vibration signal in jammed operating state;(c) Vibration signal feature extraction results of normal operating state; (d) Vibration signal feature extraction results of the mechanism card operating state.Taking the normal operating state of the circuit breaker and the operating state of the mechanism jammed mechanical fault as examples, the vibration signal features of the circuit breaker under these two operating states are extracted by the method presented in this paper.The feature extraction results are shown in Figure 5.

Table 2 .
Results of locating fault segments.