Method for on-line observation of contactor contact wear

Abstract

The application relates to a contactor contact wear online observation method, which adopts a BP neural network model to design a moving iron core displacement observer and an electromagnetic suction force observer according to the binary one-to-one mapping relationship of electromagnetic mechanism coil current i , magnetic circuit magnetic chain Ψ , moving iron core displacement x , electromagnetic suction force F x ; the electromagnetic suction force and the moving iron core displacement are directly observed according to the coil current and the coil voltage in the contactor closing process; the counterforce integral waveform is obtained after the electromagnetic suction force F x , moving iron core displacement x is calculated v f ; the counterforce step is indirectly identified by observing the counterforce integral waveform, the integral operation converts the step signal into a turning point signal, so that the contact wear is indirectly identified. The technical scheme directly observes the current contact wear condition through the designed contact wear observer in the contactor closing process, is used for guiding contact replacement and predictive maintenance of the contactor, and ensures reliable operation of the contactor during the working period.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of switching electric appliances, in particular to a contactor contact wear online observation method. BACKGROUND

[0002] The power grid system is increasingly intelligent and complex, and the contactor is one of the core components, and its performance indicators directly affect the safety and stability of the entire power grid control system. The arc during the contactor switching process continuously erodes the contact, causing its electrical life to end and leading to failure. In view of this problem, a large number of scholars have done a lot of research on ontology design and life prediction. At present, in the ontology design aspect, the electrical life of the contactor can be improved by replacing the erosion-resistant contact material and improving the arc extinguishing ability, but since the arc erosion is always difficult to avoid, the contact is continuously worn, so real-time observation of the contact wear condition has important engineering significance.

[0003] At present, in the observation of contactor contact wear, the commonly used method is to obtain the direct relationship between the contactor operating parameters such as the attraction time, arc time, arc energy and contact resistance and the contactor contact service time through experiments, and to indirectly observe the contactor contact wear condition by using feature extraction and artificial intelligence algorithms. This method constructs a prediction algorithm from the direct mapping relationship between the remaining service time and the operating characteristics, without prior knowledge, but the training process requires a large amount of data. In addition, some scholars use statistical methods to observe the contact wear condition, by statistically analyzing the cumulative arc energy of the contactor contact to make it meet the random model, and then determining the distribution function of the contactor remaining service time, and then determining the contact wear condition. This method is suitable for large quantities of simple component products, and requires long-term and large-scale experiments and statistical analysis, and the method does not consider the complex operating environment and the performance degradation characteristics of the material, resulting in low observation accuracy. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a contactor contact wear online observation method, which directly observes the current contact wear condition during the contactor closing process by designing a contact wear observer, to guide the contact replacement and predictive maintenance of the contactor, to ensure the reliable operation of the contactor during its working period.

[0005] To achieve the above purpose, the present application adopts the following technical scheme: a contactor contact wear online observation method, according to the one-to-one mapping relationship of the electromagnetic mechanism coil current i, the magnetic circuit flux linkage ψ, the moving iron core displacement x, the electromagnetic force F x , a BP neural network model is used to design a moving iron core displacement observer and an electromagnetic force observer; the electromagnetic force and the moving iron core displacement are directly observed from the coil current and the coil voltage during the contactor closing process; the electromagnetic force F x, the inverse force integral waveform v obtained after the operation of the moving iron core displacement x f ; the inverse force step is indirectly identified by observing the inverse force integral waveform, the integral operation converts the step signal into a turning point signal, so as to realize the indirect identification of the contact wear.

[0006] In a preferred embodiment, after the step signal is converted into the turning point signal, the turning point identification is performed by using the turning point identification method based on function approximation; specifically, for the discrete data containing the turning point, a function with the turning point is used to perform errorless approximation; if there is a function that can perform approximation on the original data, and the approximation error reaches the minimum, then the turning point of the function is the turning point of the original data.

[0007] By observing and analyzing the waveform of v f (t), a piecewise quadratic function is used to approximate it, so the fitting function is:

[0008]

[0009] The constraint condition is:

[0010] x0≤w slip ≤x n (2)

[0011] The optimization goal is:

[0012] min mse(f(x j )-y(x j ))x j ∈(x0,x n )j∈(0,n)(3)

[0013] In the formula, p1, p2, p4 and p5 are constraint function coefficients; w is the independent variable of the function; w slip is the independent variable of the function segment point; z slip is the dependent variable of the function segment point; x0 and x n are the independent variable ranges of the original discrete data; mse represents the mean square error; f is the fitting function, f(x j ) is the output of the fitting function when the original discrete data is the variable, y(x j ) is the original discrete data; j is the index of the original discrete data.

[0014] By sequentially traversing the constraint conditions, the optimization problem containing the constraint conditions is converted into multiple optimization problems without constraint conditions, and for a single optimization problem without constraint conditions, the Nelder-Mead simplex optimization algorithm is used to solve, and then the optimization problem containing the constraint conditions is solved.

[0015] In a preferred embodiment, the turning point identification specifically includes the following steps: first, the discrete data (x jy(x j )) into this algorithm, the segment points (w slip ,z slip ) of the fitting function f are set to the original discrete data points (x j ,y(x j )) in turn, the Nelder-Mead optimization algorithm is used to optimize the parameters p1, p2, p4, p5 of the fitting function at each segment point, so that the original discrete data is approximated, when the optimal approximation is reached, the mean square error loss j of the fitting function f(x j ) and the original discrete data y(x j ) is calculated and recorded, then it is judged whether all discrete data is traversed, if not all discrete data is traversed, the segment points (w slip ,z slip ) of the fitting function are set to the next discrete data (x j+1 ,y(x j+1 )), the above process is repeated until all discrete data is traversed; after traversing, the index j of the minimum value of the mean square error loss is found, the original discrete data point x j corresponding to this index is the inflection point of the original discrete data.

[0016] In a preferred embodiment, a flux observer is constructed for indirect measurement of the coil flux, and the contactor working process follows the voltage balance equation:

[0017]

[0018] In the formula: u is the coil voltage; i is the coil current; R is the coil resistance; ψ is the magnetic flux of the magnetic circuit

[0019] Rewrite formula (4), that is, construct a voltage integral type flux observer, which can indirectly obtain the magnetic flux of the magnetic circuit by integrating the coil voltage and coil current:

[0020] ψ = ∫(u-iR)dt (5)

[0021] For a structure-fixed electromagnetic mechanism, the coil current i, the magnetic flux of the magnetic circuit ψ, the moving iron core displacement x, and the electromagnetic suction force F x have nonlinear mapping relationships of (i, ψ)→x and (i, ψ)→F x ; based on this mapping relationship, a suction force observer and a displacement observer of the contactor are established; a multilayer BP neural network is used for the design of the observer;

[0022] Finite element simulation experiments are conducted on the electrical, magnetic, and mechanical parameters of the contactor to obtain (i, ψ)→x and (i, ψ)→F xThe binary mapping data takes the static data of the electromagnetic system (i, ψ) as a binary input, and the electromagnetic force F x and the moving iron core displacement x as the expected output, and performs offline training on the BP neural network. After the training is completed, the displacement observer and the force observer are constructed. During the operation of the observer, the (ψ, i) data pair is input, and the flux observer is matched. The neural network model is mapped to output x and F x , and finally the purpose of estimating the displacement of the moving iron core and the electromagnetic force in real time through the easily measured coil voltage and coil current of the contactor is achieved.

[0023] In a preferred embodiment, the contactor moving iron core meets the d'Alembert mechanical motion equation during movement:

[0024]

[0025] In the formula: m is the equivalent mass of the moving part; v is the moving iron core speed;

[0026] According to formula (6), the counterforce is calculated:

[0027]

[0028] As can be seen from formula (7), the counterforce observation needs to know the electromagnetic force F x , the moving iron core displacement x, and the equivalent mass m of the moving part; F x and x are observed by the displacement observer and the force observer, and then the counterforce observation is completed; the F f signal is integrated by formula (8), and the counterforce integral is obtained:

[0029]

[0030] After the integral operation, the counterforce step signal becomes a turning point signal, and the counterforce step time strictly corresponds to the turning point time. The turning point time t ct is extracted, and the current contact wear value can be obtained by means of the turning point time and the initial contact opening distance x a During the closing process of the contactor, the coil voltage u and coil current i signals of the contactor are collected, the magnetic flux ψ of the magnetic circuit is indirectly observed by means of the integral flux observer, the (i, ψ) data pair is input into the moving iron core displacement observer and the electromagnetic force observer, and the moving iron core displacement x and the electromagnetic force F x are observed. x The moving iron core displacement is differentiated to obtain the moving iron core speed v; the electromagnetic force F x is divided by the equivalent mass m of the moving iron core to obtain the electromagnetic acceleration a x ; the electromagnetic acceleration is integrated to obtain the electromagnetic speed v x ; and the electromagnetic speed v fSince the counter force F f At the moment of contact just closing t ct There is a step, so the counter force integral v f At the moment of contact just closing, there is an inflection point; through the inflection point identification, the moment of contact just closing t ct And t ct Corresponding to the closing moving iron core displacement x waveform, the current contact opening distance x a , then the current contact wear value x a0 Can be observed by using the current contact opening distance minus the initial contact opening distance x ct.ero .

[0031] Compared with the prior art, the present application has the following beneficial effects: the present application can directly observe the contact wear from the coupling relationship analysis between the electric, magnetic and mechanical physical quantities in the contactor operation process, which plays a crucial role in the stable and reliable operation of the contactor. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 The inflection point identification algorithm flowchart of the preferred embodiment of the present application;

[0033] Figure 2 The principle diagram of the suction force and displacement observer of the preferred embodiment of the present application;

[0034] Figure 3 The contact wear observation principle diagram of the preferred embodiment of the present application. DETAILED DESCRIPTION

[0035] The present application will be further described below in combination with the drawings and embodiments.

[0036] It should be noted that the following detailed description is all exemplary, and is intended to provide further description of the present application. Unless otherwise specified, all technical and scientific terms used in the present application have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0037] It should be noted that the terms used herein are only for the purpose of describing the specific embodiments, and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, they indicate the presence of a feature, step, operation, device, component and / or their combination.

[0038] The present application proposes a contactor contact wear online observation method. According to the electromagnetic mechanism coil current i, the magnetic circuit flux ψ, the moving iron core displacement x, the electromagnetic suction force F xThe binary one-to-one mapping relationship, the BP neural network model is adopted to design the moving iron core displacement observer and the electromagnetic suction observer, and the electromagnetic suction and the moving iron core displacement can be directly observed according to the coil current and the coil voltage during the contactor closing process. x , the moving iron core displacement x is operated to obtain the reaction force integral waveform (denoted as v f ). Since the reaction force F f existed in the moving iron core at the moment when the contact is just closed, the moving iron core displacement at the moment can directly observe the contact opening distance x a , and the contact opening distance can reflect the contact wear condition. Since multiple differentiations are easy to produce peak interference, the present application does not directly identify the reaction force step, but indirectly identifies the reaction force step through the observation of the reaction force integral waveform, and the step signal is converted into the inflection point signal through the integral operation, so as to realize the indirect identification of the contact wear.

[0039] Inflection point identification principle:

[0040] After the step signal is converted into the inflection point signal, a kind of inflection point identification algorithm is needed, and the present application proposes an inflection point identification method based on function approximation, the core idea of which is that for the discrete data containing the inflection point, a function with the inflection point can be used for errorless approximation; if there is a function that can approximate the original data, the approximation error is minimum, and the inflection point of the function is the inflection point of the original data. Using this idea, the inflection point identification problem can be converted into a function extreme value problem, and then the existing optimization method is used to solve it.

[0041] Through the observation and analysis of the v f (t) waveform, it can be considered that a piecewise quadratic function can be used for approximation, so the fitting function is:

[0042]

[0043] The constraint condition is:

[0044] x0≤w slip ≤x n (2)

[0045] The optimization target is:

[0046] min mse(f(x j )-y(x j ))x j ∈(x0,x n )j∈(0,n)(3)

[0047] In the formula, p1, p2, p4 and p5 are constraint function coefficients; w slip is a function segmentation point independent variable; z slip is a function segmentation point dependent variable; x0 and xn is the original discrete data independent variable range; mse represents mean square error; f is a fitting function, f(x j ) is the fitting function output when the original discrete data is the variable, and j is the original discrete data index.

[0048] For the optimization problem with a single constraint condition, the application converts the optimization problem with the constraint condition into multiple unconstrained optimization problems by sequentially traversing the constraint condition, and for a single unconstrained optimization problem, the application uses the Nelder-Mead simplex optimization algorithm to solve, and then solves the optimization problem with the constraint condition.

[0049] The flow of the inflection point identification algorithm is shown in Figure 1 : First, input the discrete data (x j , y(x j )) into the algorithm, and sequentially set the segmentation points (w slip , z slip ) of the fitting function f as the original discrete data points (x j , y(x j )). At each segmentation point, use the Nelder-Mead optimization algorithm to optimize the fitting function parameters p1, p2, p4, and p5 to make them approximate the original discrete data. When the optimal approximation is reached, calculate and record the mean square error loss j between the fitting function f(x j ) and the original discrete data y(x j ). Then, determine whether all discrete data has been traversed. If not, set the segmentation points (w slip , z slip ) of the fitting function as the next discrete data (x j+1 , y(x j+1 )), and repeat the above process until all discrete data has been traversed. After traversal is completed, find the index j of the minimum value of the mean square error loss. The original discrete data point x j corresponding to this index is the inflection point of the original discrete data.

[0050] Principle of flux observer, displacement observer and suction force observer:

[0051] The dynamic flux of the contactor is difficult to measure directly, so a flux observer needs to be constructed for indirect measurement of the coil flux. The contactor working process follows the voltage balance equation:

[0052]

[0053] In the formula: u is the coil voltage; i is the coil current; R is the coil resistance; ψ is the magnetic circuit flux

[0054] Rewrite formula (4), that is, to construct a voltage integral type flux linkage observer, which can be indirectly obtained by collecting coil voltage, coil current and integral:

[0055] ψ = ∫(u-iR)dt (5)

[0056] For the structure of the electromagnetic mechanism, the coil current i, the magnetic circuit flux linkage ψ, the moving iron core displacement x, the electromagnetic suction force F x There is a nonlinear mapping relationship between (i, ψ)→x, (i, ψ)→F x Based on this mapping relationship, the suction force observer and the displacement observer of the contactor can be established. The displacement observer and the suction force observer are designed by using the multilayer BP neural network with strong fitting ability, and the principles are as shown in Figure 2

[0057] The finite element simulation experiment is carried out on the electrical, magnetic and mechanical parameters of the contactor, and the binary mapping data of (i, ψ)→x, (i, ψ)→F x The static data of the electromagnetic system (i, ψ) are taken as binary inputs, and the electromagnetic suction force F x And the moving iron core displacement x are taken as expected outputs, and the BP neural network is trained offline. After the training is completed, the displacement observer and the suction force observer can be constructed. The observer runs with (ψ, i) data pair as input, cooperates with the flux linkage observer, and outputs x, F x Through the neural network model, the purpose of estimating the displacement of the moving iron core and the electromagnetic suction force in real time through the easily measured coil voltage and coil current of the contactor is realized.

[0058] Contact wear observation principle:

[0059] The contact wear of the contactor can be divided into mechanical wear and electrical wear. The electrical wear is the main factor of the contact wear during the operation of the contactor. The root cause of the electrical wear is the erosion of the arc generated during the operation of the contactor to the contact material. Generally speaking, the contact erosion amount is positively correlated with the contact opening distance, so the contact wear can be observed by observing the contact opening distance. The contact opening distance of the contactor can be represented by the counterforce step position, so the counterforce step observation algorithm needs to be designed to indirectly realize the contact wear observation.

[0060] The moving iron core of the contactor meets the d'Alembert mechanical motion equation during the movement:

[0061]

[0062] In the formula: m is the equivalent mass of the moving part; v is the speed of the moving iron core;

[0063] According to formula (6), the counterforce can be calculated: ​

[0064]

[0065] From formula (7), it is known that the reaction force observation needs to know the electromagnetic force F x , the moving iron core displacement x and the equivalent mass m of the moving part. F x and x are observed by means of the displacement observer and the force observer, and then the reaction force observation is completed. In theory, the reaction force step signal can be obtained by twice differentiating the displacement waveform according to formula (7), but the displacement observation signal has errors and disturbances, so that the signal accuracy is low after multiple differentiations. Therefore, the formula (8) is used for integral operation on the F f signal, and the reaction force integral is obtained.

[0066]

[0067] After the integral operation, the reaction force step signal becomes the inflection point signal, and the reaction force step time strictly corresponds to the inflection point time, so that the inflection point time t ct can be extracted by means of the inflection point identification algorithm, and the current contact wear value x a can be obtained by means of the inflection point time and the initial contact opening distance x

[0068] The finally constructed contact wear observation principle is shown in Figure 3 .

[0069] In the process of closing the contactor, the coil voltage u and coil current i signals of the contactor are collected, the magnetic flux ψ of the magnetic circuit is indirectly observed by means of the integral magnetic flux observer, the (i, ψ) data pair is input into the moving iron core displacement observer and the electromagnetic force observer, the moving iron core displacement x and the electromagnetic force F x are observed, the moving iron core velocity v is obtained by differentiating the moving iron core displacement, the electromagnetic acceleration a x is obtained by dividing the electromagnetic force F x by the equivalent mass m of the moving iron core, and the electromagnetic velocity v x is obtained by integral calculation on the electromagnetic acceleration. Since the force of the moving iron core meets the superposition theorem, the reaction force integral v x can be analyzed by subtracting the total velocity v from the electromagnetic velocity v f . Since the reaction force F f exists a step at the moment t ct when the contacts are just closed, the reaction force integral v f has an inflection point at the moment when the contacts are just closed. The moment t ct when the contacts are just closed can be obtained by means of the inflection point identification algorithm, and the current contact opening distance x a can be observed by corresponding the t ct to the closing moving iron core displacement x waveform, and the current contact wear value x ct.ero can be observed by subtracting the initial contact opening distance x a0 from the current contact opening distance.。

Claims

1. A method for online observation of contactor contact wear, characterized in that: According to the electromagnetic mechanism coil current i, magnetic circuit flux ψ, moving iron core displacement x, electromagnetic attraction F x The binary one-to-one mapping relationship is used to design the moving iron core displacement observer and the electromagnetic suction force observer using the BP neural network model; During the contactor closing process, the coil voltage u and coil current i signals of the contactor are collected, and the magnetic flux ψ is indirectly observed with the help of the integral flux observer. The (i, ψ) data is input into the moving iron core displacement observer and the electromagnetic suction force observer to observe the moving iron core displacement x and the electromagnetic suction force F. x , differentiate the displacement of the moving iron core to get the moving iron core velocity v; the electromagnetic attraction F x Divide by the equivalent mass m of the moving iron core to get the electromagnetic acceleration a x , integrating the electromagnetic acceleration can give the electromagnetic velocity v x ; Use electromagnetic velocity v x Subtracting the total velocity v, we can analyze the reaction force integral v f , due to the reaction force F f At the moment when the contacts are just closed, t ct There is a step, so the reaction force integral v f There is an inflection point when the contacts just close; The reaction force step is indirectly identified by observing the reaction force integral waveform, and the integral operation converts the step signal into an inflection point signal; The contact closing moment t can be obtained by inflection point identification ct , and t ct Corresponding to the closing moving iron core displacement x waveform, the current contact opening distance x can be observed a , then use the current contact distance minus the initial contact distance x a0 The current contact wear value x can be observed ct.ero ; After converting the step signal into an inflection point signal, the inflection point identification method based on function approximation is used to identify the inflection point. Specifically, for discrete data with an inflection point, a function with an inflection point is used to perform a zero-error approximation on it. If there is a function that approximates the original data with the minimum approximation error, then the inflection point of this function is the inflection point of the original data. By v f (t) The waveform is observed and analyzed, and a piecewise quadratic function is used to approximate it, so the fitting function is: The constraints are: x0≤w slip ≤x n (2) The optimization goal is: min mse(f(x j )-y(x j ))x j ∈(x0,x n )j∈(0,n)(3) Where: p1, p2, p4 and p5 are the coefficients of the constraint function; the w parameter is the independent variable of the function; w slip is the function segmentation point independent variable; z slip is the dependent variable of the function segmentation point; x0, x n is the range of the original discrete data independent variable; mse represents the mean square error; f is the fitting function, f(x j ) is the fitting function output when the original discrete data is used as the variable, y(x j ) is the original discrete data; j is the original discrete data index; By traversing the constraints in sequence, the optimization problem with constraints is transformed into multiple optimization problems without constraints. For each unconstrained optimization problem, the Nelder-Mead simplex optimization algorithm is used to solve it, and then the optimization problem with constraints is solved.

2. The method for online observation of contactor contact wear according to claim 1, characterized in that: Inflection point identification specifically includes the following steps: First, the discrete data (x j ,y(x j )) Input this algorithm and fit the segment points (w slip ,z slip ) are set as the original discrete data points (x j ,y(x j )), use the Nelder-Mead optimization algorithm to optimize the fitting function parameters p1, p2, p4, p5 at each segment point to make it close to the original discrete data. When the optimal approximation is reached, calculate and record the fitting function f(x j ) and the original discrete data y(x j )’s mean square error loss j , and then determine whether all discrete data have been traversed. If not, the segmentation points (w slip ,z slip ) is set as the next discrete data (x j+1 ,y(x j+1 )), repeat the above process until all discrete data are traversed; after the traversal is completed, find the index j of the minimum value of the mean square error loss, and the original discrete data point x corresponding to this index j This is the inflection point of the original discrete data.

3. The method for online observation of contactor contact wear according to claim 1, characterized in that: A flux observer is constructed to indirectly measure the coil flux. The contactor working process follows the voltage balance equation: Where: u is the coil voltage; i is the coil current; R is the coil resistance; ψ is the magnetic flux By rewriting Equation (4), a voltage-integrated flux observer is constructed. The magnetic flux can be indirectly obtained by simply collecting the coil voltage and coil current and integrating them: ψ=∫(u-iR)dt (5) For an electromagnetic mechanism with a fixed structure, its coil current i, magnetic flux ψ, moving iron core displacement x, and electromagnetic attraction F x There exists (i,ψ)→x, (i,ψ)→F x The nonlinear mapping relationship of contactor is established based on the mapping relationship; the suction observer and displacement observer of contactor are designed by multi-layer BP neural network; Finite element simulation experiments are performed on the electrical, magnetic and mechanical parameters of the contactor to obtain (i, ψ)→x, (i, ψ)→F x Binary mapping data, taking the static data of the electromagnetic system (i, ψ) as binary input, the electromagnetic attraction F x and the displacement of the moving iron core x as the expected output, and the BP neural network is trained offline. After the training is completed, the displacement observer and the suction observer are constructed. When the observer is running, the (ψ, i) data pair is used as input, and the flux observer is used to map the output x and F through the neural network model. x , and ultimately achieve the goal of real-time estimation of the displacement of the moving iron core and the electromagnetic attraction through the coil voltage and coil current that are easily measured by the contactor.

4. The method for online observation of contactor contact wear according to claim 1, characterized in that: The moving iron core of the contactor complies with the D'Alembert mechanical motion equation during its movement: Where: m is the equivalent mass of the moving part; v is the speed of the moving iron core; The reaction force is calculated according to formula (6): From formula (7), we can see that the reaction force observation requires the known electromagnetic attraction force F x , the displacement of the moving iron core x and the equivalent mass m of the moving parts; with the help of the displacement observer and the suction observer, F x and x, and then complete the reaction force observation; f The signal is integrated using formula (8) to obtain the reaction force integral: After the integral operation, the reaction force step signal becomes the inflection point signal, and the reaction force step moment strictly corresponds to the inflection point moment. The inflection point moment t is extracted. ct , with the help of the inflection point moment and the initial contact opening distance x a The current contact wear value can be obtained.

Images