Method for inspecting magnetic rings in batches

Abstract

The invention relates to magnetic ring inductance inspection, and discloses a method for inspecting magnetic rings in batches. According to the method, inspection information of a to-be-detected magnetic ring group is obtained by using a batch inspection tool, and the information is input into a batch estimation model, so that an initial inspection result of the to-be-detected magnetic ring group is output. The batch prediction model is constructed based on a convolutional neural network, can predict the inductance value of each magnetic ring through test information, and judges whether the inductance value meets the standard or not. Compared with traditional one-by-one testing, the method has the advantage that the testing efficiency is remarkably improved. According to the method, the production process is optimized, and the overall production cycle is shortened by reducing unnecessary test steps.

Description

Technical Field

[0001] The invention relates to a magnetic ring inductance testing technology, and in particular to a method for batch testing of magnetic rings. Background Art

[0002] Electromagnetic interference is an important issue in electronic product design, and using filters to suppress electromagnetic interference is a common and effective method. Magnetic rings, also known as ferrite magnetic rings or common mode chokes, are one of the key components in filters and are mainly used to suppress high-frequency noise and electromagnetic interference.

[0003] The inductance of a magnetic ring is determined by factors such as its material properties, size, and winding method. During the production process, these factors may fluctuate due to manufacturing process limitations or material inconsistencies, resulting in differences in the actual inductance of each magnetic ring. This inconsistency is a challenge for applications that require precise inductance because it may affect the performance of the final product.

[0004] In order to ensure product quality, it is usually necessary to test the magnetic rings one by one to confirm whether they meet the specifications. Although this step is necessary, it does add additional costs and may extend the production cycle, thus affecting the shipment rate. Summary of the invention

[0005] The object of the present invention is to provide a method for batch testing of magnetic rings, so that the testing of the inductance values ​​of the magnetic rings can be carried out in batches, thereby improving the testing efficiency.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a method for batch testing of magnetic rings, comprising the following steps:

[0007] Step 1: Use batch testing tooling to obtain the inspection information of the magnetic ring group to be tested;

[0008] Step 2: Input the inspection information into the batch estimation model, and the batch estimation model outputs the initial inspection result of the magnetic ring group to be tested;

[0009] Step 3: Select the form of cyclic re-inspection according to the initial inspection results, and obtain qualified magnetic ring groups and unqualified magnetic rings through cyclic re-inspection.

[0010] Furthermore, the batch inspection tooling includes: a carrier plate, a Gaussmeter probe, a metal sheet and a tray, the tray is used to load the magnetic ring, the tray is placed on the carrier plate through a limit block, the metal sheet is embedded in the carrier plate, and is used to connect the magnetic rings in series, and a number of the Gaussmeter probes are evenly distributed on the carrier plate for obtaining inspection information.

[0011] Furthermore, the batch estimation model is trained with a training set collected by batch inspection tooling until the loss function converges; wherein, the position information of the magnetic ring, the magnetic field strength, the magnetic field direction, the measured inductance value and the measured results are integrated into a data set, and the data set is divided into a training set, a validation set and a test set.

[0012] Furthermore, the method of obtaining the inspection information of the magnetic ring group to be tested by using a batch inspection tool comprises the following steps:

[0013] Step 11: Fill the tray with magnetic rings to form a magnetic ring group to be tested, and position the magnetic ring group to be tested on the carrier board so that the metal sheet and the pins of the magnetic ring form a series path, and connect a stable current to the series path;

[0014] Step 12: Use a Gaussmeter probe to obtain the strength and direction of the magnetic field formed by the series path, and use the position coordinates of the Gaussmeter probe and the acquired magnetic field strength and magnetic field direction as the inspection information of the magnetic ring group to be tested;

[0015] The number of magnetic rings in the series path is N, and the N magnetic rings are distributed in a rectangular shape, N=n x ×n y , n x and n y Represent the number of magnetic rings on the x-axis and y-axis respectively; there are M Gaussmeter probes around each magnetic ring, and they are evenly cross-distributed around the magnetic ring. The magnetic field strength is B i , the magnetic field direction is θ i , i is the position number of the Gaussmeter probe.

[0016] Furthermore, the batch estimation model is constructed based on a convolutional neural network, and the batch estimation model includes a data preprocessing module, a convolutional feature extraction module, a fully connected layer module, and an output and decision module; the batch estimation model inputs the inspection information into the batch estimation model, and the batch estimation model outputs the initial inspection result of the magnetic ring group to be tested, including the following steps:

[0017] Step 21: The data preprocessing module performs Z-Score standardization processing on the input inspection information to obtain preprocessed data;

[0018] Step 22: The convolution feature extraction module extracts magnetic field features from the input preprocessed data through a convolution layer to obtain a magnetic field feature map.

[0019] Step 23: The fully connected layer module flattens the features of the magnetic field feature map through the fully connected layer, and integrates the features through the fully connected layer to obtain fused features;

[0020] Step 24: The output and decision module obtains the predicted inductance value of each magnetic ring according to the fusion features, determines whether each magnetic ring meets the standard, and outputs the initial inspection result of the magnetic ring group to be tested.

[0021] Furthermore, the convolutional layer formula is: j =f(W j *V+b j ), where O j is the magnetic field characteristic map generated by the jth convolution kernel; W j is the weight of j convolution kernels; b j is a bias term; * represents a convolution operation; f(·) is an activation function, and the activation function f(x)=max(0,x) is selected.

[0022] Furthermore, the formula of the fully connected layer is H = W fc F+b fc , where H is the fusion feature, F is the expansion vector of the magnetic field feature map output by the last convolutional layer, and W fc is the weight matrix of the fully connected layer, b fc is the bias vector of the fully connected layer.

[0023] Furthermore, the output and decision module obtains the predicted inductance value of each magnetic ring according to the fusion feature, determines whether each magnetic ring meets the standard, and outputs the initial inspection result of the magnetic ring group to be tested, including the following steps:

[0024] Step 241: Process the fused features through the output layer to obtain the predicted inductance value; the formula of the output layer is L = W out H+b out , where L is the predicted inductance value, W out is the weight matrix of the output layer, b out is the bias vector of the output layer;

[0025] Step 242: Set the range of qualified inductance values ​​to (L min , L max ), use the formula to judge whether the predicted inductance value of the magnetic ring is qualified. The judgment formula is:

[0026] Step 243: If the predicted inductance values ​​of all magnetic rings in the magnetic ring group to be tested are qualified, the initial inspection result is that all are qualified, otherwise the initial inspection result is that some are qualified, and the position information of the unqualified magnetic rings is output.

[0027] Furthermore, the batch estimation model is optimized using a loss function, and the formula of the loss function is S = αS reg +(1-α)S cls ,in,

[0028] α is a hyperparameter 0<α<1;

[0029] S reg is the regression loss, N is the number of magnetic rings, L iis the real inductance of the ith magnetic ring, is the predicted inductance value obtained by the model;

[0030] S cls is the classification loss, zi is the label of whether the i-th magnetic ring is qualified or not, 1 means qualified, 0 means unqualified, is the pass rate predicted by the model.

[0031] Further, the method of selecting the cyclic retest form according to the initial test result includes the following steps:

[0032] If the initial inspection result is that all products are qualified, a random replacement re-inspection will be carried out; if the initial inspection result is that some products are qualified, a defective replacement re-inspection will be carried out;

[0033] The random replacement re-inspection is to randomly replace 1 to 2 magnetic rings in the magnetic ring group to be tested with magnetic rings that have not been measured, and re-execute steps 1 and 2. If the initial inspection result is all qualified again, a qualified magnetic ring group is obtained. If the initial inspection result changes from all qualified to partially qualified, a defective replacement re-inspection is performed;

[0034] The defective replacement re-inspection is to replace the unqualified magnetic rings in the magnetic ring group to be tested with qualified magnetic rings, and re-execute steps 1 and 2. If the initial inspection result changes from partially qualified to fully qualified, a qualified magnetic ring group is obtained. If the initial inspection result is partially qualified again, the defective replacement re-inspection is re-executed.

[0035] Beneficial effects: The present invention provides a method for batch inspection of magnetic rings. The method obtains the inspection information of the magnetic ring group to be tested by using batch inspection tooling, and inputs this information into a batch estimation model, thereby outputting the initial inspection results of the magnetic ring group to be tested. The batch estimation model is built based on a convolutional neural network, which can predict the inductance value of each magnetic ring through the inspection information and determine whether it meets the standard. Compared with the traditional one-by-one testing, this method significantly improves the inspection efficiency. This method also optimizes the production process and shortens the overall production cycle by reducing unnecessary testing steps. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0037] Figure 1 is a flow chart of the inspection method of the present invention;

[0038] Figure 2 It is a structural schematic diagram of the batch detection tooling in the present invention;

[0039] Figure 3 It is a schematic diagram of the structure of the batch inspection tooling after the tray is removed in the present invention;

[0040] Figure 4 It is a structural schematic diagram of the tray in the present invention;

[0041] Figure 5 It is a structural schematic diagram of the magnetic ring in the present invention.

[0042] Explanation of the reference numerals: 1. carrier board; 2. limit block; 3. metal sheet; 4. Gauss meter probe; 5. tray; 51. tray body; 52. placement slot; 53. lifting plate; 54. giving way structure; 6. magnetic ring; 61. pin. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0044] Embodiment 1

[0045] This embodiment provides a batch inspection tool for inspecting magnetic rings, such as Figure 2-Figure 5 As shown, the batch inspection tooling includes: a carrier plate 1, a Gaussmeter probe 4, a metal sheet 3 and a tray 5. The tray 5 is used to load the magnetic ring 6. The tray 5 is placed on the carrier plate 1 through a limit block 2. The metal sheet 3 is embedded in the carrier plate 1 and is used to connect the magnetic ring 6 in series. Several Gaussmeter probes 4 are evenly distributed on the carrier plate 1 for obtaining inspection information.

[0046] Specifically, Figure 5 For example, the middle magnetic ring 6 is in the form of a double coil, and the magnetic ring 6 has two sets of pins 61; the magnetic ring 6 is placed Figure 4 In the tray 5, the tray 5 includes an integrally formed tray body 51 and a lifting plate 53. The tray body 51 is provided with a placement slot 52 and a clearance structure 54. The lifting plate 53 is arranged in the placement slot 52. The magnetic ring 6 is mounted on the lifting plate 53 and is limited by the placement slot 52. The placement slots 52 are distributed in a rectangular array on the tray body 51. The example shows a 3×4 sample. The clearance structure 54 is used to make way for the Gauss meter probe 4 and is evenly arranged around the placement slot 52. In the example, the four corners of the clearance structure 54 are provided with a clearance structure 54. The setting position of the lifting plate 53 relative to the tray body 51 must meet the requirement that after the tray body 51 is placed on the carrier board 1, the pin 61 contacts the metal sheet 3, but the magnetic ring 6 does not break away from the limitation of the placement slot 52.

[0047] Placing the tray 5 on the carrier board 1 requires positioning the tray 5 to ensure that the pins 61 of the magnetic ring 6 are smoothly docked with the metal sheet 3 to form a series path. The positioning of the tray 5 is achieved by the limit block 2.

[0048] The distribution form of the Gaussmeter probe 4 on the carrier board 1 needs to match the distribution form of the yield structure 54. That is to say, the installation position of the Gaussmeter probe 4 on the carrier board 1 is determined. The placement position of the magnetic ring 6 on the carrier board 1 is also accurately positioned, which greatly reduces the difficulty of obtaining the position of the magnetic ring 6.

[0049] After the metal sheets 3 at both ends of the series path are powered, the series path generates a magnetic field, and the Gaussmeter probe 4 obtains its own position coordinates and the collected magnetic field strength and direction as a collection point information, and all point information is summarized as inspection information.

[0050] Specifically, the inspection information is input into a controller embedded with a batch estimation model for batch inspection.

[0051] Embodiment 2

[0052] This embodiment is based on the batch inspection method of magnetic rings proposed by the batch inspection tool in the first embodiment. Figure 1 As shown, a method for batch testing of magnetic rings comprises the following steps:

[0053] Step 1: Use batch testing tooling to obtain the inspection information of the magnetic ring group to be tested;

[0054] Step 2: Input the inspection information into the batch estimation model, and the batch estimation model outputs the initial inspection result of the magnetic ring group to be tested;

[0055] Step 3: Select the form of cyclic re-inspection according to the initial inspection result, and obtain the qualified magnetic ring group and the unqualified magnetic ring 6 through cyclic re-inspection.

[0056] The present invention proposes to use batch inspection tooling to inspect multiple magnetic rings 6 at one time, which greatly improves the inspection efficiency compared to the traditional one-by-one testing method. Batch inspection reduces the time and labor cost required for testing each magnetic ring 6 individually, thereby reducing the overall inspection cost. The most appropriate re-inspection form is automatically selected according to the initial inspection results, and random replacement re-inspection or defective replacement re-inspection is performed, which simplifies the re-inspection process and improves efficiency.

[0057] Step 1 "obtaining inspection information of the magnetic ring group to be tested using batch inspection tooling" includes the following steps:

[0058] Step 11: Fill the tray 5 with magnetic rings 6 to form a magnetic ring group to be tested, and position the magnetic ring group to be tested on the carrier 1 so that the metal sheet 3 and the pins 61 of the magnetic ring 6 form a series path, and connect a stable current to the series path;

[0059] Step 12: using the Gauss meter probe 4 to obtain the strength and direction of the magnetic field formed by the series path, and using the position coordinates of the Gauss meter probe 4 and the acquired magnetic field strength and magnetic field direction as the inspection information of the magnetic ring group to be tested;

[0060] There are 12 magnetic rings 6 in the series path, and the 12 magnetic rings 6 are distributed in a 3×4 rectangular pattern; there are 4 Gauss meter probes 4 around each magnetic ring 6, and they are evenly cross-distributed around the magnetic ring. The magnetic field strength is B i , the magnetic field direction is θ i , i is the position number of the Gaussmeter probe 4.

[0061] This step can obtain the inspection information of multiple magnetic rings 6 at the same time through one operation, which significantly improves the detection efficiency. Using the Gauss meter probe 4 to measure the magnetic field strength and direction can more accurately evaluate the inductance value of the magnetic ring 6. The whole process can be completed by automated equipment, reducing the possibility of human error.

[0062] The batch estimation model is built based on a convolutional neural network. The batch estimation model includes a data preprocessing module, a convolutional feature extraction module, a fully connected layer module, and an output and decision module. In step 2, "inputting the inspection information into the batch estimation model, and the batch estimation model outputting the initial inspection results of the magnetic ring group to be tested" includes the following steps:

[0063] Step 21: The data preprocessing module performs Z-Score standardization on the input inspection information to obtain preprocessed data; preprocessing is performed on the magnetic field strength and direction around each magnetic ring 6, and the data can be converted to a unified scale, which helps to eliminate the dimensional differences between different sensors, making the data more comparable and consistent, thereby accelerating the training process of the model, reducing oscillations during the gradient descent process, and improving the convergence speed of the model. Preprocessing can also help avoid numerical instability problems caused by data values ​​that are too large or too small.

[0064] Specifically, Wherein, (x, y) is the position of the magnetic ring 6, i is the i-th Gaussmeter probe 4 of the magnetic ring 6, and the Gaussmeter probes 4 outside the magnetic ring 6 are numbered counterclockwise, μ B and σ B are the mean and standard deviation of all measured magnetic field strengths; μ θ and σ θ is the mean and standard deviation of all measured angles.

[0065] That is, B i,(x,y) represents the magnetic field strength obtained by the i-th Gaussmeter probe 4 located outside the (x, y) magnetic ring 6, B i ' ,(x,y) Is relative to B i,(x,y)The standardized data of i,(x,y) represents the magnetic field direction obtained by the i-th Gaussmeter probe 4 located outside the (x, y) magnetic ring 6, θ i ' ,(x,y) is the relative θ i,(x,y) of standardized data.

[0066] Step 22: The convolution feature extraction module extracts magnetic field features from the input preprocessed data through a convolution layer to obtain a magnetic field feature map.

[0067] The convolutional layer formula is O j =f(W j *V+b j ), where O j is the magnetic field characteristic map generated by the jth convolution kernel; W j is the weight of j convolution kernels; V is the magnetic field matrix, V = [B ix , B iy ], B ix is the component of the magnetic field strength in the x direction measured by the i-th Gaussmeter probe 4, B iy is the component of the magnetic field intensity in the y direction measured by the i-th Gaussmeter probe 4; b j is a bias term; * indicates convolution operation; f(·) is an activation function, and the activation function f(x)=max(0,x) is selected. The convolution layer can automatically learn the characteristics of the magnetic ring inductance value from the magnetic field, and then help the subsequent fully connected layer to estimate and judge the inductance value.

[0068] Step 23: The fully connected layer module flattens the features of the magnetic field feature map through the fully connected layer, and integrates the features through the fully connected layer to obtain fused features;

[0069] The formula for the fully connected layer is H = W fc F+b fc , where H is the fusion feature, F is the expansion vector of the magnetic field feature map output by the last convolutional layer, and W fc is the weight matrix of the fully connected layer, b fc is the bias vector of the fully connected layer.

[0070] Step 24 "the output and decision module obtains the predicted inductance value of each magnetic ring according to the fusion features, determines whether each magnetic ring meets the standard, and outputs the initial inspection result of the magnetic ring group to be tested" includes the following steps:

[0071] Step 241: Process the fused features through the output layer to obtain the predicted inductance value; the formula of the output layer is L = W out H+b out , where L is the predicted inductance value, W out is the weight matrix of the output layer, b outis the bias vector of the output layer;

[0072] Step 242: Set the range of qualified inductance values ​​to (L min , L max ), use the formula to judge whether the predicted inductance value of the magnetic ring 6 is qualified. The judgment formula is: Q is the prediction result of a single magnetic ring 6, 1 indicates qualified, and 0 indicates unqualified.

[0073] Step 243: If the predicted inductance values ​​of all magnetic rings 6 in the magnetic ring group to be tested are qualified, the initial inspection result is that all are qualified, otherwise the initial inspection result is that some are qualified, and the position information of the unqualified magnetic ring 6 is output.

[0074] In order to improve the accuracy of the model, the batch estimation model is optimized using the loss function. The formula of the loss function is S = αS reg +(1-α)S cls ,in,

[0075] α is a hyperparameter 0<α<1;

[0076] S reg is the regression loss, N is the number of magnetic rings 6, L i is the real inductance of the ith magnetic ring, is the predicted inductance value obtained by the model;

[0077] S cls is the classification loss, zi is the label of whether the i-th magnetic ring 6 is qualified, 1 means qualified, 0 means unqualified, is the pass rate predicted by the model.

[0078] The batch estimation model is trained with the training set collected by batch inspection tooling until the loss function converges; wherein, the position information of the magnetic ring 6, the magnetic field strength, the magnetic field direction, the measured inductance value and the measured results are integrated into a data set, and the data set is divided into a training set, a validation set and a test set.

[0079] Step 3 "Select the form of re-inspection based on the initial inspection results" includes the following steps:

[0080] If the initial inspection result is that all products are qualified, a random replacement re-inspection will be carried out; if the initial inspection result is that some products are qualified, a defective replacement re-inspection will be carried out;

[0081] The random replacement re-inspection is to randomly replace 1 to 2 magnetic rings 6 in the magnetic ring group to be tested with magnetic rings that have not been measured, and re-execute steps 1 and 2. If the initial inspection result is all qualified again, a qualified magnetic ring group is obtained. If the initial inspection result changes from all qualified to partially qualified, a defective replacement re-inspection is performed;

[0082] The defective replacement re-inspection is to replace the unqualified magnetic ring 6 in the magnetic ring group to be tested with a qualified magnetic ring 6, and re-execute steps 1 and 2. If the initial inspection result changes from partially qualified to fully qualified, a qualified magnetic ring group is obtained. If the initial inspection result is partially qualified again, the defective replacement re-inspection is re-executed.

[0083] Different re-inspection strategies are adopted for different initial inspection results. When all magnetic rings are qualified, random replacement re-inspection is used to further verify the stability of the results; when some magnetic rings are found to be unqualified, defective replacement re-inspection is used to determine the reasons for the unqualified and eliminate unqualified products. Such a strategy ensures the effectiveness of quality control while taking into account efficiency. Through multiple cycles of re-inspection, it can be ensured that each magnetic ring is fully inspected, thereby ensuring the quality of the final product and reducing the risk of rework or return due to product quality problems.

[0084] Those skilled in the art will appreciate that the above-mentioned embodiments are specific examples for implementing the present invention, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for batch inspection of magnetic rings, characterized in that: The following steps are involved: Step 1: Use batch inspection tooling to obtain inspection information of the magnetic ring group to be tested; Step 2: Input the inspection information into the batch estimation model, and the batch estimation model outputs the initial inspection results of the magnetic ring group to be tested; Step 3: Select the cyclic re-inspection form according to the initial inspection result, and obtain the qualified magnetic ring group and the unqualified magnetic ring through the cyclic re-inspection (6).

2. A method for batch inspection of magnetic rings according to claim 1, characterized in that: The batch inspection tooling comprises: a carrier plate (1), a Gaussmeter probe (4), a metal sheet (3) and a tray (5); the tray (5) is used to load a magnetic ring (6); the tray (5) is placed on the carrier plate (1) via a limit block (2); the metal sheet (3) is embedded in the carrier plate (1) and is used to connect the magnetic ring (6) in series; a plurality of Gaussmeter probes (4) are evenly distributed on the carrier plate (1) and are used to obtain inspection information.

3. A method for batch inspection of magnetic rings according to claim 2, characterized in that: The batch estimation model is trained by a training set collected by batch detection tooling until the loss function converges; wherein the position information of the magnetic ring (6), the magnetic field strength, the magnetic field direction, the measured inductance value and the measured results are integrated into a data set, and the data set is divided into a training set, a verification set and a test set.

4. A method for batch inspection of magnetic rings according to claim 2, characterized in that: The method of obtaining the inspection information of the magnetic ring group to be tested by using the batch inspection tool comprises the following steps: Step 11: Fill the tray (5) with magnetic rings (6) to form a magnetic ring group to be tested, and position the magnetic ring group to be tested on the carrier (1), so that the metal sheet (3) and the pins (61) of the magnetic ring (6) form a series path, and connect a stable current to the series path; Step 12: using the Gauss meter probe (4) to obtain the strength and direction of the magnetic field formed by the series path, and using the position coordinates of the Gauss meter probe (4) and the acquired magnetic field strength and magnetic field direction as the inspection information of the magnetic ring group to be tested; The number of magnetic rings (6) in the series path is N, and the N magnetic rings (6) are distributed in a rectangular shape, N=n x ×n y , n x and n y Respectively represent the number of magnetic rings (6) on the x-axis and y-axis; each magnetic ring (6) is surrounded by M Gaussmeter probes (4), which are evenly cross-distributed around the magnetic ring; the magnetic field strength is B i , the direction of the magnetic field is θ i , i is the position number of the Gaussmeter probe (4).

5. A method for batch inspection of magnetic rings according to claim 4, characterized in that: The batch estimation model is constructed based on a convolutional neural network and includes a data preprocessing module, a convolutional feature extraction module, a fully connected layer module, and an output and decision module. Inputting the inspection information into the batch estimation model and outputting the initial inspection results of the magnetic ring group to be tested by the batch estimation model include the following steps: Step 21: The data preprocessing module performs Z-Score normalization on the input test information to obtain preprocessed data; Step 22: The convolution feature extraction module extracts magnetic field features from the input pre-processed data through the convolution layer to obtain a magnetic field feature map; Step 23: The fully connected layer module flattens the features of the magnetic field feature map through the fully connected layer, and integrates the features through the fully connected layer to obtain fused features; Step 24: The output and decision module obtains the predicted inductance value of each magnetic ring based on the fusion features, determines whether each magnetic ring meets the standard, and outputs the initial inspection result of the magnetic ring group to be tested.

6. A method for batch inspection of magnetic rings according to claim 5, characterized in that: The convolutional layer formula is O j =f(W j *V+b j ), where O j is the magnetic field characteristic map generated by the jth convolution kernel; W j is the weight of j convolution kernels; b j is a bias term; * represents a convolution operation; f(·) is an activation function, and the activation function f(x) = max(0,x) is selected.

7. A method for batch inspection of magnetic rings according to claim 6, characterized in that: The formula of the fully connected layer is H=W fc F+b fc , where H is the fusion feature, F is the expansion vector of the magnetic field feature map output by the last convolutional layer, and W fc is the weight matrix of the fully connected layer, b fc is the bias vector of the fully connected layer.

8. The method for batch testing of magnetic rings according to claim 7, characterized in that: The output and decision module obtains the predicted inductance value of each magnetic ring based on the fusion features, determines whether each magnetic ring meets the standard, and outputs the initial inspection result of the magnetic ring group to be tested, including the following steps: Step 241: Process the fused features through the output layer to obtain the predicted inductance value; the formula of the output layer is L = W out H+b out , where L is the predicted inductance value, W out is the weight matrix of the output layer, b out is the bias vector of the output layer; Step 242: Set the qualified inductance value range to (L min , L max ), use the formula to judge whether the predicted inductance value of the magnetic ring (6) is qualified. The judgment formula is: Step 243: If the predicted inductance values ​​of all the magnetic rings (6) in the magnetic ring group to be tested are qualified, the initial inspection result is that all are qualified; otherwise, the initial inspection result is that some are qualified, and the position information of the unqualified magnetic rings (6) is output.

9. The method for batch testing of magnetic rings according to claim 8, characterized in that: The batch estimation model is optimized using a loss function, and the formula of the loss function is S = αS reg +(1-α)S cls ,in, α is a hyperparameter 0<α<1; S reg is the regression loss, N is the number of magnetic rings (6), L i is the true inductance of the i-th magnetic ring, is the predicted inductance value obtained by the model; S cls is the classification loss, zi is the label indicating whether the i-th magnetic ring (6) is qualified, 1 means qualified, 0 means unqualified, is the pass rate of model prediction.

10. The method for batch inspection of magnetic rings according to claim 9, characterized in that: The method of selecting the cyclic retest form according to the initial test results includes the following steps: If the initial inspection results show that all products are qualified, a random replacement re-inspection will be carried out; if the initial inspection results show that some products are qualified, a defective replacement re-inspection will be carried out; The random replacement re-inspection is to randomly replace 1 to 2 magnetic rings (6) in the magnetic ring group to be tested with magnetic rings that have not been measured, and re-execute steps 1 and 2. If the initial inspection result is again all qualified, a qualified magnetic ring group is obtained. If the initial inspection result changes from all qualified to partially qualified, a defective replacement re-inspection is performed; The defective replacement re-inspection is to replace the unqualified magnetic ring (6) in the magnetic ring group to be tested with a qualified magnetic ring (6), and re-execute steps 1 and 2. If the initial inspection result changes from partially qualified to fully qualified, a qualified magnetic ring group is obtained. If the initial inspection result is again partially qualified, the defective replacement re-inspection is re-executed.

Images