Circuit breaker operating mechanism key component fatigue analysis and diagnosis method

Abstract

The invention discloses a fatigue analysis and diagnosis method for key components of a circuit breaker operating mechanism. The fatigue analysis and diagnosis method comprises the following steps: (1) collecting vibration signals in the operating process of a circuit breaker; and (2) tracing the energy storage spring deformation, and finely dividing vibration signals accompanying the operation process into three sub-stages: a tripping trigger stage, an energy storage spring deformation stage and a closing buffer stage. And (3) carrying out adaptive variation mode extraction (AVME) by using the signal in the deformation stage of the energy storage spring, so as to obtain an expected mode. And (4) extracting typical time-frequency domain characteristics of an expected mode as condition attributes of a theoretical information table of key components of the operating mechanism, taking whether fatigue occurs as decision attributes, and establishing a fault original decision information table. And (5) reducing the fault original decision information table by using a rough set theory to obtain a fault reduction decision information table. And (6) training and learning a support vector machine (SVM), and constructing a fault diagnosis model for the key components of the circuit breaker operating mechanism. Experiments show that the method has high accuracy and practicability in the field of circuit breaker fault identification and evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical equipment fault diagnosis, and in particular to a method for circuit breaker fault diagnosis by extracting phased signal modes and then simplifying attributes, which is used in the technical field of high-voltage electrical equipment fault detection. Background Art

[0002] High-voltage circuit breakers play a core role in power systems, requiring a high degree of technical expertise. Their functions primarily encompass two aspects: first, controlling the opening and closing of equipment to maintain stable control of the power grid; and second, protecting the normal operation of power lines to prevent accidents caused by line faults. The reliability of high-voltage circuit breakers is directly related to the safe and stable operation of the power system and is a key factor in ensuring the normal operation of the entire power grid. The operating mechanism is the most important actuator in a circuit breaker, and its performance directly impacts its reliability. Spring-operated mechanisms are currently the most widely used in the field due to their simple principle and ease of maintenance. Circuit breakers that have been in operation for a long time often experience accidents such as improper opening and closing due to spring fatigue, forced shutdown due to energy storage failure, or arcing and explosions caused by prolonged opening. Promptly detecting spring fatigue and implementing appropriate safety measures to prevent these accidents are crucial for the stable operation of circuit breakers and improving the reliability of power systems.

[0003] Vibration signals contain rich information about the movements of various components. Using non-invasive vibration signals to identify mechanical faults in key circuit breaker components has achieved good practical results. Currently, variational mode decomposition (VMD) is a commonly used algorithm for fault diagnosis of key high-voltage circuit breaker components. This algorithm can decompose the signal into frequency bands with different center frequencies, but suffers from over-decomposition, leading to feature loss and modal aliasing. As an improved algorithm for VMD, variational mode extraction (VME), originally applied in the medical field, shares the same theoretical foundation and is currently widely used in fault analysis. However, due to the difficulty in determining relevant initial values ​​such as the center frequency, it still suffers from problems such as low versatility and low recognition. Summary of the Invention

[0004] In order to solve the above problems and achieve the above objectives, the present invention adopts the following technical solutions:

[0005] A fatigue analysis and diagnosis method for key components of a circuit breaker operating mechanism is proposed. This method uses an adaptive variational mode extraction (AVME) algorithm to extract features from circuit breaker vibration signals and simplify their attributes. xThe collaborative optimization mathematical model, minf, utilizes the particle swarm optimization (PSO) algorithm to optimize the VME parameters and complete the AVME algorithm architecture. The circuit breaker vibration signal is finely divided into three stages based on the motion characteristics of the operating components. The energy storage spring's action is traced, the characteristic stages are optimized, and the typical eigenvalues ​​of the desired mode are calculated. Rough set theory is used to simplify the features, addressing the loose coupling between multidimensional monitoring parameters and establishing an effective fault decision-making relationship between the optimized characteristic quantities and component status.

[0006] The specific steps are as follows:

[0007] Step 1: Use a piezoelectric acceleration sensor to collect vibration signals during the circuit breaker operation process.

[0008] Step 2: Trace the energy storage spring action process, use a high-speed camera to obtain an image sequence, use the deformation degree of the circuit breaker operating mechanism as the segmentation benchmark, and refine the vibration signal into three sub-stages: trip triggering stage, energy storage spring deformation stage, and closing buffer stage. Take the phased signal of the energy storage spring deformation stage and based on the step convergence factor B x The modified objective function minf is used to process the signal using AVME to obtain the desired mode.

[0009] Step 3: Obtain 23 typical time-frequency domain features of the expected mode, including maximum value, minimum value, peak-to-peak value, mean value, variance, standard deviation, mean square value, root mean square value, form factor, peak factor, pulse factor, margin factor, kurtosis factor, skewness factor, center of gravity frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, spectral kurtosis mean, spectral kurtosis standard deviation, spectral kurtosis skewness, and spectral kurtosis kurtosis, and construct the conditional attribute set of the theoretical information table of the key components of the operating mechanism, and whether the key components are fatigued is used as the decision attribute set.

[0010] Step 4: Based on the rough set theory, attribute simplification is performed, and the feature quantities with strong robustness are selected as the diagnosis basis to establish an effective fault decision relationship.

[0011] Step 5: Divide the sample features into training set, test set, and cross-validation set to train the SVM algorithm and establish an effective fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Flowchart of fatigue analysis and diagnosis method for key components of circuit breaker operating mechanism

[0013] Figure 2 Time domain diagram of closing vibration signal in normal operating state

[0014] Figure 3 Time domain diagram of the elastic deformation stage of the closing vibration signal extracted by AVME in normal operating state

[0015] Figure 4 Time domain diagram of closing vibration signal in fatigue state

[0016] Figure 5 Time domain diagram of the elastic deformation stage of the closing vibration signal extracted by AVME in fatigue state

[0017] Figure 6 Fault diagnosis recognition rate diagram DETAILED DESCRIPTION

[0018] This method takes the closing process of ZN65-12 vacuum circuit breaker as an example to collect the vibration signals of normal and artificially set spring fatigue during the closing process of the circuit breaker.

[0019] Step 1: Select the NI USB-4431 acquisition card, install a CCD high-speed camera, and install the AD50S piezoelectric accelerometer on the circuit breaker housing for pre-test commissioning: Collect data from 15 combinations of switches to diagnose their rationality and verify the accuracy and stability of the equipment and platform.

[0020] Step 2: Set the first coil of the spring's moving end as the tracking target, and use a high-speed camera to capture a sequence of images of the circuit breaker's energy storage spring's deformation process.

[0021] Step 3: Simulate the closing spring fatigue failure by adjusting the pre-compression of the closing spring to achieve artificial spring fatigue failure setting. Select the vibration signal a1(t) of the circuit breaker in normal state, ..., a n (t), vibration signal b1(t),......,b n (t), 80 groups for each case, a total of 160 groups of data, construct the data matrix X(t) = [a1(t); a2(t) ... a 80 (t); b1(t); b2(t)......b 80 (t)].

[0022] Step 4: Trace the deformation of the energy storage spring and scientifically divide the circuit breaker vibration process into the tripping triggering stage, the energy storage spring deformation stage and the closing buffer stage. Correspondingly, the vibration signal a n (t) and b n (t) Segment processing, where a n (t)=[a n1 (t),a n2 (t),a n3 (t)]b n (t) = [b n1 (t),b n2 (t),b n3(t)]n=1,2,3,......,80, that is, X(t) can be refined into: X(t)=[X1(t),X2(t),X3(t)], X1(t=[a 11 (t); a 21 (t)......a 801 (t); b 11 (t); b 21 (t)......b 801 (t)], X2(t)=[a 12 (t); a 22 (t)......a 802 (t); b 12 (t); b 22 (t)......b 802 (t)], X3(t)=[a 13 (t); a 23 (t)......a 803 (t); b 13 (t); b 23 (t)......b 803 (t)], corresponding to three stages respectively. The energy storage spring deformation stage can best reflect the energy change of the spring during the transmission process, so the signal X2(t) is selected for envelope spectrum analysis. The objective function of the PSO algorithm is set, and the calculation formula is:

[0023]

[0024] In the formula is the total energy of the modal component, B(f) is the envelope spectrum of the modal component, max(B(f)) is the maximum value of the envelope spectrum of the modal component, and B is the maximum value of all maximum points in the envelope spectrum greater than The sum of the extreme points, B x is the step size convergence factor, and the calculation formula is:

[0025] B x =[(f e -f b ) 2 +1] (2)

[0026] Where, f e is the expected convergence center frequency of the VME algorithm, f b is the initial convergence center frequency of the VME algorithm.

[0027] Based on the above objective function, the PSO algorithm is used to participate in the VME algorithm parameter optimization process. The specific process is as follows:

[0028] (4.1) Initialize PSO particle swarm parameters: upper and lower limits of target search space μd and l d , two learning factors c1 and c2, the maximum number of algorithm iterations T or convergence accuracy ξ, and the position and velocity of each particle;

[0029] (4.2) Set the particle fitness value fitness, the calculation formula is the same as formula (1) in step 4;

[0030] (4.3) Update the velocity and position of each particle. The calculation formula is:

[0031]

[0032] The symbols in the formula are shown in Table 1.

[0033] Table 1 Description of symbols in PSO

[0034]

[0035] (4.4) The termination condition of the iteration is reaching the maximum number of iterations T or the convergence accuracy ξ.

[0036] Step 5: Perform variational mode extraction on the signal X2(t). Specifically, the steps include:

[0037] (5.1) Reconstruct the energy storage release phase signal X2(t) by μ d (t) and f r (t) constitute, namely:

[0038] X2(t)=μ d (t)+f r (t) (4)

[0039] Where μ d (t) represents the expected mode (IMF), f r (t) represents the residual signal.

[0040] (5.2) around f obtained in step 4 e Find the compact function J1, that is:

[0041]

[0042] Where, represents the partial derivative at time t, δ(t) represents the Dirac function, * represents the convolution operation, u d (t) represents the intrinsic mode function, ω d From formula (6), we can get u d (t) corresponds to the center frequency, represents the L2 norm.

[0043] ω d =2πfe (6)

[0044] (5.3) Select the filter for the residual signal f r (t) is filtered to ensure that the energy of the residual signal is minimized in the frequency band where the desired mode is located. The filter response calculation formula is:

[0045]

[0046] (5.4) Using the filter of formula (7) to get f r (t) is filtered to ensure f r (t) and μ d (t) Minimize the spectrum overlap and obtain the penalty function J2, which is calculated as follows:

[0047]

[0048] (5.5) The expected pattern problem is transformed into a constrained objective function minimization problem, and the calculation formula is:

[0049]

[0050] (5.6) Using the quadratic penalty term and the Lagrange multiplication operator method, the constrained problem is transformed into an unconstrained problem. The augmented Lagrangian function is established and the calculation formula is:

[0051]

[0052] Where λ is the Lagrange multiplier.

[0053] Introducing Parseval's isometric transformation theorem, transforming equation (10) into the frequency domain, the calculation formula is:

[0054]

[0055] (5.7) Update the parameters using the alternating direction multiplier algorithm (ADMM) through multiple iterative suboptimization and The calculation formula is:

[0056]

[0057] When the conditions are met When , the iteration ends.

[0058] Step 6: After obtaining 160 sets of expected modes, calculate the maximum value, minimum value, peak-to-peak value, mean value, variance, standard deviation, mean square value, root mean square value, form factor, peak factor, impulse factor, margin factor, kurtosis factor, skewness factor, center of gravity frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, spectral kurtosis mean, spectral kurtosis standard deviation, spectral kurtosis skewness, and spectral kurtosis kurtosis of each component, and finally form a 160×23 eigenvector matrix.

[0059] Step 7: Set the eigenvector matrix as the conditional attribute set and the corresponding state as the decision attribute set to establish the initial fault decision table. After attribute reduction, the precise fault decision relationship is obtained. Attribute reduction, as the name suggests, involves removing some knowledge or features from the attributes. More precisely, it involves removing irrelevant or unimportant knowledge or features without affecting the classification capabilities of the knowledge base.

[0060] Step 8: Divide the simplified data into a training set, a test set, and a cross-validation set, train the SVM model, and build a fault diagnosis model for the key components of the circuit breaker operating mechanism.

[0061] Step 9: The prediction results of this fault diagnosis model are as follows Figure 6 As shown in the figure, the recognition accuracy reaches 98%, which is of practical significance.

Claims

1. A fatigue analysis and diagnosis method for key components of a circuit breaker operating mechanism, characterized in that: Including steps: Step 1: Install the signal detection device and collect 80 sets of vibration signals of the circuit breaker in normal state and spring fatigue state; Step 2: Tracing the deformation of the energy storage spring, the vibration signal accompanying the operation process is refined and divided into three sub-stages: the trip trigger stage, the energy storage spring deformation stage, and the closing buffer stage. Adaptive variational modal extraction is performed on the signal of the preferred characteristic stage—the energy storage spring deformation stage—to obtain the desired mode. This mode is closely related to the signal itself and better reflects the stage characteristics of the signal, making signal analysis more flexible. Step 3: For the extracted expected mode, obtain 23 typical time-frequency domain characteristic indicators, including the maximum value, minimum value, peak-to-peak value, mean value, variance, standard deviation, mean square value, root mean square value, form factor, peak factor, impulse factor, margin factor, kurtosis factor, skewness factor, center of gravity frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, spectral kurtosis mean, spectral kurtosis standard deviation, spectral kurtosis skewness, and spectral kurtosis kurtosis, and construct a conditional attribute set. Whether the key components of the circuit breaker operating mechanism are fatigued is used as the decision attribute set. Step 4: Modify and simplify attributes to reduce redundancy, obtain a simplified fault decision table, and obtain diagnostic rules from the table; Step 5: Divide the 160 sets of collected data into training set, test set, and cross-validation set, and use the support vector machine model for state identification.

2. The method according to claim 1, characterized in that: In step 2, an adaptive modal decomposition method is proposed to extract the desired modal components. The specific method is: the VME algorithm requires the preprocessing parameter center frequency ω d , the particle swarm optimization (PSO) algorithm is used to adaptively optimize the parameters, and an objective function based on the step-length convergence factor correction is designed. By searching for the minimum value of the objective function, the parameter selection problem of the artificial mode is solved, so that the initial center of gravity frequency is as close as possible to the convergence center frequency. The step-length convergence factor calculation formula is: B x =[(f e -f b ) 2 +1] (1) where f e is the expected convergence center frequency of the VME algorithm, f b is the initial convergence center frequency of the VME algorithm. The objective function calculation formula is: in is the total energy of the modal component, B(f) is the envelope spectrum of the modal component, max(B(f)) is the maximum value of the envelope spectrum of the modal component, and B is the maximum value of all maximum points in the envelope spectrum greater than The sum of the extreme points of . By correcting the step-size convergence factor, the difference between the expected center frequency and the initial center frequency is gradually reduced, highlighting the essential characteristics of the fault state, which is conducive to extracting significant features.

3. The method according to claim 1, characterized in that: In step 2, the energy storage spring deformation is traced, with the first coil of the energy storage spring's moving end as the tracking target. High-speed cameras are used to capture the vibration signal accompanying the operation process, dividing it into three phases: the trip trigger phase, the energy storage spring deformation phase, and the closing buffer phase. The trip trigger phase is from coil energization to core impact tripping; the energy storage spring deformation phase is from the beginning of spring deformation to maximum deformation; and the closing buffer phase is from maximum spring deformation to the end of oscillation.

4. The method according to claim 1, characterized in that: In step 3, 23 typical time-frequency domain characteristic indicators such as the maximum value, minimum value, peak-to-peak value, mean value, variance, standard deviation, mean square value, root mean square value, waveform factor, peak factor, pulse factor, margin factor, kurtosis factor, skewness factor, center of gravity frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, spectral kurtosis mean, spectral kurtosis standard deviation, spectral kurtosis skewness, and spectral kurtosis kurtosis of the signal are obtained to more comprehensively analyze the fault data characteristics.

5. The method according to claim 1, characterized in that: In step 4, using rough sets to simplify the attributes of the 23 characteristic indicators can better remove redundant components, select characteristic quantities with strong robustness as the diagnosis basis, obtain more accurate fault decision-making relationships, and better achieve the purpose of accurate identification.

Images