Motor slot wedge assembly detection method based on magnetic field distribution characteristics

Abstract

The invention discloses a motor slot wedge assembly detection method based on magnetic field distribution characteristics, and particularly relates to the technical field of motor structure detection. Obtaining a standard magnetic field distribution map of the motor stator in a winding-free state and an actual magnetic field distribution map after the winding is installed, and obtaining a difference magnetic field map through image registration analysis; extracting magnetic flux density offset characteristics of the slot region based on the difference magnetic field map, and constructing a magnetic field disturbance characteristic vector; inputting the feature vectors into a trained multi-classification support vector machine model, and outputting the assembly state of each slot position; carrying out abnormal marking on the slot position in the loose, skew or missing state, and generating a visual detection result and an alarm prompt; the method realizes non-contact and high-precision identification of slot wedge assembly defects, is suitable for online quality monitoring in a motor assembly process, and has the advantages of high automation degree and good engineering applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of motor structure detection, and particularly relates to a motor slot wedge assembly detection method based on magnetic field distribution characteristics. BACKGROUND

[0002] As a core power source in modern industry and transportation fields, the performance and reliability of a motor depend largely on the structural stability and electromagnetic consistency of the stator winding. During the assembly of the motor stator core, the correct installation of the slot wedge not only affects the fixing strength of the winding, but also directly affects the magnetic flux distribution of the stator core, thereby relating to the electromagnetic performance and thermal stability of the whole machine. The traditional slot wedge assembly quality detection method relies on manual observation or mechanical contact type detection, which is not only low in efficiency, but also easily affected by human error, and cannot accurately identify the slight magnetic field disturbance caused by the slot wedge assembly, especially in an automatic production line, it is difficult to realize high-precision and high-throughput nondestructive online detection.

[0003] In practical applications, if there are hidden defects such as slight misalignment, looseness or non-tightness in the slot wedge assembly, it will cause abnormal local magnetic flux density distribution, and long-term operation may cause local winding temperature rise to increase, even breakdown, and in severe cases, it will cause the entire motor to fail. However, the magnetic field asymmetry problem caused by slight assembly error is usually difficult to identify effectively through conventional electrical performance testing or structural scanning methods, causing long-term latent fault hazards. SUMMARY

[0004] The purpose of the present application is to provide a motor slot wedge assembly detection method based on magnetic field distribution characteristics to solve the problems in the background art.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme: a motor slot wedge assembly detection method based on magnetic field distribution characteristics, comprising:

[0006] S1, obtaining a standard magnetic field distribution map M0 of a target motor stator in a winding-free state;

[0007] S2, obtaining an actual magnetic field distribution map M1 of the motor stator to be tested after the winding is installed and a low-frequency excitation current is passed through;

[0008] S3, performing image registration analysis on the standard magnetic field distribution map M0 and the actual magnetic field distribution map M1 to obtain a registration difference magnetic field map AM;

[0009] S4, extracting a slot area magnetic field strength change feature vector based on the registration difference magnetic field map AM, including a plurality of slot position magnetic flux density offset values at corresponding positions;

[0010] S5, input the slot area magnetic field strength variation feature vector into the trained assembly state discrimination model, output the assembly state label set corresponding to the slot, including normal, loose, skew or missing assembly state;

[0011] S6, according to the assembly state label set, alarm prompt is given to the slot with abnormal label.

[0012] Preferably, the S1 comprises:

[0013] S11, collect the radial and axial magnetic flux density distribution data of the motor stator under constant excitation condition;

[0014] S12, the collected magnetic flux density data is processed by space coordinate mapping and normalization, and a two-dimensional space distribution map is constructed;

[0015] S13, the two-dimensional space distribution map is processed by image enhancement and filtering, and finally a standard magnetic field distribution map M0 is obtained.

[0016] Preferably, the S2 comprises:

[0017] S21, under the action of excitation current, the magnetic flux density values at a plurality of measuring points on the outer periphery of the motor stator are collected, and the measuring points cover the full circumferential and axial area of the stator;

[0018] S22, the collected magnetic flux density data is processed by coordinate mapping and normalization, and a two-dimensional magnetic flux density map with the same space structure as the standard magnetic field distribution map M0 is constructed;

[0019] S23, the two-dimensional magnetic flux density map is processed by image enhancement and filtering to obtain the actual magnetic field distribution map M1 of the motor stator under the assembly state.

[0020] Preferably, the S3 comprises:

[0021] S31, the obtained standard magnetic field distribution map M0 and actual magnetic field distribution map M1 to be registered are respectively processed by gray scale equalization;

[0022] S32, a feature point set P0 is extracted from the magnetic field edge area of the standard magnetic field distribution map M0, and a candidate feature point set P1 corresponding to P0 is searched in the actual magnetic field distribution map M1, and the feature points are constructed according to the local magnetic flux density change slope;

[0023] S33, according to the feature point sets P0 and P1, a deformation matrix T between the two maps is calculated, which is composed of translation parameters, rotation parameters and scale parameters;

[0024] S34, apply the deformation matrix T to the actual magnetic field distribution map M1 to obtain a registration image M1' that is completely aligned with the standard magnetic field distribution map M0 in spatial position, and calculate the pixel-level difference value of M0 and M1', forming a registration difference magnetic field map AM.

[0025] Preferably, the S4 comprises:

[0026] S41, convert the pixel gray value in the registration difference magnetic field map AM into a magnetic flux density difference value matrix to establish a magnetic flux density offset mapping table;

[0027] S42, according to the motor stator structure parameters, extract the spatial region corresponding to each slot position in the registration difference magnetic field map AM to form a slot position mask template;

[0028] S43, apply the slot position mask template to the magnetic flux density offset mapping table, and extract the magnetic flux density offset value in each slot position corresponding region according to the slot position sequence;

[0029] S44, arrange the magnetic flux density offset values of all slot positions in turn to form a feature vector reflecting the slot region magnetic field disturbance.

[0030] Preferably, the S5 comprises:

[0031] S51, input the obtained slot region magnetic field disturbance feature vector as an input feature into a pre-trained assembly state discrimination model, and the discrimination model is a classification model based on a multi-class support vector machine;

[0032] S52, in the model training stage, the motor assembly samples manually labeled are used as the training set, the feature dimension corresponds to the slot magnetic flux density offset value, and the output label includes four states of normal, loose, skew and missing;

[0033] S53, after the slot region magnetic field intensity change feature vector is processed by the discrimination model, the discrimination result of each slot position corresponding state is calculated, and the final discrimination category is determined according to the maximum probability principle of classification score;

[0034] S54, the discrimination result of each slot position is composed into an assembly state marking set, and the assembly state marking set S is a one-dimensional label array, and the number of elements is consistent with the number of slot positions.

[0035] Preferably, the S6 comprises:

[0036] S61, establish a mapping relationship between the assembly state marking set S and the slot position number corresponding to the motor stator structure parameters to generate a slot position state reference table;

[0037] S62, according to the slot position state reference table, filter the slot positions whose state labels are "loose", "skew" or "missing", mark them as abnormal slot positions, and enter S63;

[0038] S63, encapsulate the number corresponding to each abnormal slot, the assembly state type and the slot space position coordinates as abnormal information, and output an alarm prompt, and enter S64;

[0039] S64, visualize all abnormal information and generate a structure detection report.

[0040] In the above technical solution, the present application provides technical effects and advantages:

[0041] 1. The present application realizes high-precision, non-contact identification of the motor slot wedge assembly state by combining slot area magnetic flux density offset feature extraction and multi-classification support vector machine model based on image registration and difference analysis method of magnetic field distribution map, breaking through the technical bottleneck that traditional manual or contact detection method is difficult to find small assembly defects.

[0042] 2. The present application forms a complete defect detection technical chain by constructing a standardized data processing process, from magnetic field map preprocessing, feature vector construction, to classification discrimination and visual output. The detection result can alarm in real time and generate a structured maintenance report, providing scientific basis and decision support for quality control and subsequent maintenance in the assembly process, especially suitable for online quality monitoring in automatic motor assembly production line. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art based on these drawings.

[0044] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0046] Embodiment, please refer to Figure 1 The motor slot wedge assembly detection method based on magnetic field distribution characteristics described in the present embodiment includes:

[0047] S1, obtain the standard magnetic field distribution map M0 of the target motor stator in the no-winding state.

[0048] Step S11: Fix the motor stator to be tested on a support made of non-magnetic material, ensuring its stable position and no obstruction to the measurement area. In the state of the stator without winding, apply a constant excitation condition by an external excitation device, which adopts a constant voltage driving mode to make the excitation coil generate a sinusoidal excitation voltage with a frequency of 50 Hz and an amplitude of 12 V, forming a constant magnetic flux condition.

[0049] During the excitation process, data acquisition is performed using a magnetic flux density acquisition device arranged on the outer periphery of the motor stator, which is composed of an array of Hall effect magnetic sensors. A measurement point is arranged every 10 degrees along the circumferential direction of the stator, and a total of 36 measurement points are arranged. Three layers of measurement point planes are arranged along the axial direction of the stator, and the radial and axial magnetic flux density values at each measurement point are recorded in turn. The measured magnetic flux density is expressed in terms of magnetic flux per unit area, with the unit being Tesla.

[0050] Step S12: After data acquisition, the magnetic flux density data of each measurement point is mapped according to its physical position in space to construct a magnetic flux density spatial coordinate matrix. Assuming that the stator circumference is θ direction and the axial direction is z direction, the magnetic flux density data can be represented as B(θi, zj), where θi is the angular position of the i-th circumferential measurement point, zj is the axial height position of the j-th layer, and B represents the magnetic flux density of the corresponding measurement point. To eliminate the deviation caused by excitation intensity under different test conditions, the B(θi, zj) data is normalized. The normalization method adopts the maximum-minimum value normalization formula, which is specifically:

[0051] The maximum value of all measured magnetic flux density data is set as B_max, and the minimum value is set as B_min. The normalized value B_norm(θi, zj) of each measurement point is calculated as follows: normalized magnetic flux density = (original magnetic flux density - B_min) / (B_max - B_min); the normalized result ranges from 0 to 1, forming a two-dimensional matrix B_norm, which is used for subsequent image processing and analysis.

[0052] Step S13: Convert the normalized two-dimensional magnetic flux density matrix B_norm into a magnetic field distribution spectrum in the form of a gray-scale image, with a linear relationship between the gray value and the magnetic flux density. To improve image quality and data recognition accuracy, further image enhancement processing is performed on the spectrum, including:

[0053] Contrast stretching processing: expand the image gray scale dynamic range through linear transformation to enhance the contrast of the magnetic flux density boundary region;

[0054] Edge-preserving filtering processing: noise reduction of the spectrum, considering both spatial distance and gray scale difference, while retaining the edge features of the magnetic flux density and effectively suppressing high-frequency noise;

[0055] Pseudo-color mapping processing: apply the processed gray map to the Jet pseudo-color conversion scheme, so that the magnetic flux density change is more intuitive and visualized, where the low density area is displayed as blue and the high density area is displayed as red.

[0056] The final obtained pseudo-color image is the standard magnetic field distribution map M0 of the motor stator in the no winding state.

[0057] S2, obtain the actual magnetic field distribution map M1 of the motor stator after the winding installation is completed when the low-frequency excitation current is passed.

[0058] Step S21: install the excitation coil on the motor stator after the winding installation is completed, and apply a low-frequency sinusoidal excitation current to the excitation coil through an external excitation controller, the excitation current frequency is set to 50 Hz, and the voltage amplitude is 12 volts.

[0059] Under the action of the excitation current, the magnetic field of the motor stator outer periphery is collected. The collection device includes a Hall effect magnetic sensor array, a total of 108 measuring points, one measuring point is arranged every 10 degrees in the circumferential direction, and three layers of equidistant measuring point planes are arranged in the axial direction, so as to cover the entire stator circumferential and axial space region. Each measuring point records the magnetic flux density value at the corresponding position, and the unit is Tesla.

[0060] Step S22: map the collected magnetic flux density values according to the geometric positions of the measuring points in two-dimensional space coordinates, establish a polar coordinate system, where the angle coordinate corresponds to the circumferential direction position, and the axial coordinate corresponds to the sensor plane position, to form a magnetic flux density space distribution matrix B_m(θi,zj).

[0061] In order to eliminate the influence of fluctuations in the electrical excitation process on the absolute value of the magnetic flux density, the magnetic flux density values of all measuring points are normalized. The normalization adopts a linear maximum and minimum value normalization method. The normalized two-dimensional matrix B_norm(θi,zj) constitutes an initial actual magnetic field distribution map consistent with the spatial distribution structure of the standard magnetic field distribution map M0, which is used for subsequent image processing operations.

[0062] Step S23: In order to improve the boundary definition and local anomaly response capability of the magnetic flux density map, the above normalized map B_norm(θi,zj) is processed, mainly including image enhancement processing and image filtering processing.

[0063] The processed image is converted into a color gradient by pseudo-color mapping (using the Jet color mapping scheme) to form the final actual magnetic field distribution map M1.

[0064] S3, image registration analysis is performed on the standard magnetic field distribution map M0 and the actual magnetic field distribution map M1 to obtain the registration difference magnetic field map ΔM.

[0065] Step S31: The obtained standard magnetic field distribution map M0 and the actual magnetic field distribution map M1 are both two-dimensional pseudo-color images, which need to be converted into grayscale image form for image processing. The conversion method is as follows: for each pixel, the red, green and blue channels are weighted and averaged, and the specific calculation formula is: gray value = 0.3*red channel value + 0.59*green channel value + 0.11*blue channel value.

[0066] After the grayscale image is generated, the histogram equalization processing is performed on M0 and M1 respectively to enhance the image contrast and unify the brightness distribution. The specific processing is as follows: the pixel grayscale histogram of each image is counted, the cumulative distribution function is calculated, and the pixel grayscale value is nonlinearly mapped using the cumulative distribution function to expand the grayscale dynamic range of the original image and enhance the image detail performance, providing a stable grayscale basis for subsequent feature point extraction.

[0067] Step S32: For the standard magnetic field distribution map M0 after equalization processing, a consistent edge detection method based on gradient direction is used to identify the magnetic field strength boundary region, and the main steps are as follows:

[0068] The local grayscale gradient direction and gradient amplitude of each pixel in the image are calculated, and the Sobel operator is used for horizontal and vertical gradient calculation;

[0069] The pixel points with gradient amplitude greater than 15% of the maximum grayscale value are taken as candidate edge points;

[0070] The 3*3 neighborhood around the candidate edge points is fitted with the grayscale change rate, and the points with obvious slope change and consistent change direction are taken as feature points P0.

[0071] Subsequently, in the actual magnetic field distribution map M1, according to the position and local slope characteristics of each point in P0, a candidate feature point set P1 is searched in M1 image, and the matching criteria include slope direction similarity, grayscale gradient amplitude similarity and spatial position proximity. The point pair with a matching error lower than the preset threshold value εp is confirmed as an effective matching point pair for subsequent registration matrix solving.

[0072] Step S33: The matching point pairs in the feature point sets P0 and P1 obtained in step S32 are used to construct the point pair coordinate mapping relationship. The point coordinates in P0 are set as (xi, yi), and the corresponding point coordinates in P1 are set as (xi', yi'), then a two-dimensional affine transformation matrix T is found to make: ; wherein the parameters a and d are scale factors, the parameters b and c are used to describe rotation, and tx and ty are translation offsets.

[0073] ​The least square method is used to globally optimize fitting of multiple matching point pairs to solve parameters of the deformation matrix T, so as to ensure that the M1 image is aligned with M0 as much as possible after affine transformation.

[0074] Step S34: The deformation matrix T is applied to the actual magnetic field image M1, and the gray resampling of non-integer pixel positions is performed through a bilinear interpolation algorithm to obtain a registered image M1', which is completely aligned with M0 in terms of spatial structure. Then, a difference matrix between M0 and M1' is calculated in a pixel-by-pixel difference manner, and the difference calculation formula is as follows: difference value D(θi, zi) = |M0(θi, zi) - M1'(θi, zi)|; a two-dimensional matrix composed of all difference values is a registration difference magnetic field map ΔM, which reflects the magnetic field abnormal area caused by winding assembly and can be used for subsequent magnetic flux disturbance feature extraction and assembly defect discrimination processing.

[0075] S4, extracting a slot area magnetic field strength change feature vector based on the registration difference magnetic field map ΔM, including magnetic flux density offset values of multiple slots at corresponding positions.

[0076] Step S41: Since ΔM is in the form of a gray image, in order to effectively analyze electromagnetic physical quantities, it is necessary to convert the pixel gray value thereof into a magnetic flux density offset. The conversion relationship is set according to the original normalized magnetic flux density dimension and image mapping parameters, and a linear mapping manner is adopted:

[0077] It is assumed that the gray value range in ΔM is 0 to 255, the original magnetic flux density normalized value range is 0 to 1, and the actual magnetic flux density change range is set to 0 to ΔB_max, and the unit is Tesla;

[0078] The calculation formula of the magnetic flux density difference value B_d corresponding to the gray value g of each pixel is: magnetic flux density difference value B_d = (g - 0) / (255 - 0) * (ΔB_max - 0) = g * ΔB_max / 255. All pixel points of the magnetic flux density difference value are mapped to a two-dimensional matrix according to their spatial coordinates to form a magnetic flux density offset mapping table B_d(θi, zj), wherein θi represents the circumferential position, and zj represents the axial position.

[0079] Step S42: Based on the structural design drawing of the motor stator, the total number of slot positions, the circumferential arrangement manner and the slot width size and other parameters are extracted to construct a slot position mask template Mask_s for accurately positioning the area range corresponding to each slot position in the magnetic flux density offset mapping table. The specific construction manner is as follows:

[0080] It is assumed that there are N_s slot positions on the stator, which are distributed at equal intervals along the 360-degree circumferential direction;

[0081] Each slot position corresponds to a sector area in the atlas, and the angle range thereof is 360 degrees divided by the number of slot positions;

[0082] Each sector slot region extends in the axial direction, covering a range corresponding to the sampling region zj;

[0083] Each slot region is defined as a binary mask region, and the pixels with a mask value of 1 represent belonging to the slot region, and the pixels with a value of 0 are non-slot regions.

[0084] The finally formed Mask_s includes N_s independent slot regions, numbered in the order of S_1 to S_Ns, for subsequent feature extraction.

[0085] Step S43: Apply each slot mask region Sk (k = 1, 2, …, N_s) to the magnetic flux density offset mapping table B_d(θi, zj) in turn, and extract all magnetic flux density difference values within the corresponding slot region. Calculate the arithmetic mean of all non-zero pixel values in each region to obtain the magnetic flux density offset value B_k of the slot, and the calculation formula is as follows: B_k = sum of all pixel magnetic flux density difference values in the mask region ÷ pixel number; B_k obtained represents the average offset degree of the magnetic field disturbance corresponding to the slot, which can reflect the overall magnetic field asymmetry.

[0086] Step S44: Arrange all slot magnetic flux density offset values obtained in step S43 in the order of slot number from S_1 to S_Ns to form a one-dimensional vector F = [B1, B2, …, B_Ns] as the feature vector of the slot field disturbance. The feature vector has the following technical features:

[0087] The number dimension is equal to the number of motor stator slots;

[0088] Each element B_k has a physical quantity meaning, and the unit is Tesla;

[0089] The vector can be directly used as input data for subsequent classification and discrimination models to identify abnormal slots and judge the assembly state. The feature vector accurately quantifies the magnetic field disturbance degree caused by the motor winding assembly in each slot under the premise of ensuring the spatial position correspondence.

[0090] S5, input the slot magnetic field intensity change feature vector into the trained assembly state discrimination model, and output the assembly state label set of the corresponding slot, including normal, loose, skewed or missing assembly states.

[0091] Step S51: Form the slot magnetic field intensity change feature vector as an input variable and input it into the pre-trained assembly state discrimination model. The model uses a support vector machine classification algorithm (Support Vector Machine, abbreviated as SVM), and is extended to a multi-classification structure, using a one-versus-all strategy to construct multiple binary classification sub-models to complete the four-classification task.

[0092] The input of the model is a one-dimensional slot area magnetic field strength variation feature vector, the dimension of which is equal to the number of stator slot positions, and each element represents a slot position magnetic flux density offset value; the output of the model is a set of classification labels, indicating the assembly state of each slot position.

[0093] Step S52: In the model training stage, a supervised learning structure is constructed, a large number of actual motor assembly samples are collected, and each slot position assembly state is manually labeled by an experienced quality inspection engineer. The labeled labels are divided into four categories: normal: slot wedge fastening, accurate positioning; loose: slot wedge is not pressed tightly, there is a gap; skew: slot wedge is installed obliquely, the edge is not flat; missing: slot wedge is not installed or seriously falls off.

[0094] Each training sample contains:

[0095] Input: Feature vector F_train, dimension N (number of slots), each element is a magnetic flux density offset value;

[0096] Output: Label vector L_train, dimension N, each element is the state label (such as “normal” or “missing”) of the corresponding slot position.

[0097] The support vector machine is used to build a classification boundary with a radial basis function as the kernel function, the penalty factor C and the kernel width σ are adjusted, and the optimal hyperparameter combination is selected through cross-validation to ensure the generalization ability of the model.

[0098] Step S53: The slot area magnetic field disturbance feature vector F_input of the motor to be tested is input into the trained support vector machine classification model, and each slot position corresponding feature value is individually judged. The model outputs the confidence scores corresponding to the four types of assembly states through multiple support vector machine sub-models, and the maximum score principle is used to determine the final classification result:

[0099] Let the four state scores of slot position k be D1(k), D2(k), D3(k), and D4(k), respectively, then the final discrimination class C_k is the class number corresponding to the maximum score. C_k=argmax{D1(k),D2(k),D3(k),D4(k)}; this method can effectively improve the discrimination accuracy of the model for multi-class boundaries, and is especially suitable for recognizing state changes caused by small magnetic flux differences in the motor assembly process.

[0100] Step S54: The discrimination results Ck of all slot positions k∈{1,2,...,N} are summarized and combined into a one-dimensional assembly state label set S=[C1,C2,...,CN] in order of slot position number.

[0101] The assembly state label set S satisfies the following characteristics:

[0102] The number of elements is equal to the total number of motor stator slot positions;

[0103] Each element is an enumerated label value, corresponding to one of the four assembly states;

[0104] The state label set S is directly used as output result for slot wedge defect identification, automatic alarm or visual feedback.

[0105] The generation of the label set S completes the mapping from the magnetic flux disturbance features to the assembly quality state, providing a basis for automatic detection and online verification in the motor production line.

[0106] S6, according to the assembly state label set, the slot position with abnormal label is alarmed.

[0107] Step S61: associate the obtained assembly state label set S with the structural design parameters of the motor stator. There are N_s slots in the stator structure, the slot number is sequentially numbered from 1 to N_s in the circumferential direction, and the assembly state label set S is a one-dimensional label array, the kth element S_k represents the assembly state label of the kth slot. Correspondence between S and slot number, generate slot state table, each record contains the following fields:

[0108] Slot number ID (1 to N_s); assembly state label (normal, loose, skew, missing); angle position θk; axial height layer zj. The table structure is used for subsequent abnormal state screening and position output.

[0109] Step S62: traverse the slot state table, and judge the assembly state label field of each record. If the state label field value is "normal", ignore the slot; if it is any of the following three types, it is determined to be an abnormal slot:

[0110] Loose: indicating that the slot wedge is not completely pressed;

[0111] Skew: indicating that the slot wedge insertion angle deviation causes skew;

[0112] Missing: indicating that the slot wedge is not installed or has fallen off.

[0113] The screening result is stored in the form of an abnormal slot number list, which is used as a basis for subsequent abnormal information generation and alarm prompt.

[0114] Step S63: for each slot number in the abnormal slot number list, extract the following information from the slot state table:

[0115] Slot number ID; assembly state label (loose, skew or missing); spatial position coordinates (θk, zj), representing the position of the slot in the stator circumferential and axial directions, respectively.

[0116] The above information is combined into a structured data format, named as an exception information structure Ek, and all the exception slot information is collected to form a list E = {E1, E3,..., Em}, where m is the number of exception slots. Subsequently, an alarm prompt logic is triggered, and each piece of exception information in E is synchronously output to a graphical interface, a text or graphical warning is popped up, and a sound prompt is generated to prompt the user that there is an assembly defect at present.

[0117] Step S64: Each record in the exception information E is marked with a position on the motor stator structure diagram, and graphical markers are used to distinguish different exception types, for example, a red dot represents a missing part, a yellow triangle represents a skew, and an orange square represents a loose part. In the graphical interface, the user is allowed to click on any exception slot marker to view detailed assembly status information, including the number, exception type, position coordinates, and confidence score. At the same time, a structure detection report is generated based on all the exception information, and the report content includes: detection time; motor number; total number of exception slots; exception type and position of each slot; detection conclusion and recommended repair operation.

[0118] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for detecting motor slot wedge assembly based on magnetic field distribution characteristics, characterized in that: include: S1. Obtain the standard magnetic field distribution map M0 of the target motor stator in the unwinding state; S2. Obtain the actual magnetic field distribution spectrum M1 of the motor under test when a low-frequency excitation current is applied to the stator after the winding is installed; S3. Perform image registration analysis on the standard magnetic field distribution map M0 and the actual magnetic field distribution map M1 to obtain the registration difference magnetic field map ΔM. S4. Based on the registration difference magnetic field map ΔM, extract the feature vector of magnetic field intensity change in the slot area, including the magnetic flux density offset values ​​of multiple slots at corresponding positions; S5. Input the feature vector of magnetic field intensity change in the slot area into the trained assembly state discrimination model, and output the assembly state label set of the corresponding slot, including normal, loose, skewed or missing assembly state. S6. Based on the assembly status mark set, issue an alarm for slots with abnormal marks.

2. The method for detecting motor slot wedge assembly based on magnetic field distribution characteristics according to claim 1, characterized in that: S1 includes: S11. Collect radial and axial magnetic flux density distribution data of the motor stator under constant excitation conditions; S12. Perform spatial coordinate mapping and normalization on the collected magnetic flux density data to construct a two-dimensional spatial distribution map; S13. The two-dimensional spatial distribution map is processed by image enhancement and filtering to finally obtain the standard magnetic field distribution map M0.

3. The method for detecting motor slot wedge assembly based on magnetic field distribution characteristics according to claim 2, characterized in that: S2 includes: S21. Under the action of excitation current, collect the magnetic flux density values ​​at multiple measuring points on the outer periphery of the motor stator, with the measuring points covering the entire circumferential and axial regions of the stator. S22. Perform coordinate mapping and normalization on the collected magnetic flux density data to construct a two-dimensional magnetic flux density map consistent with the spatial structure of the standard magnetic field distribution map M0. S23. Perform image enhancement and filtering on the two-dimensional magnetic flux density map to obtain the actual magnetic field distribution map M1 of the stator of the motor under test in the assembled state.

4. The method for detecting motor slot wedge assembly based on magnetic field distribution characteristics according to claim 1, characterized in that: S3 includes: S31. Perform grayscale equalization processing on the obtained standard magnetic field distribution map M0 to be registered and the actual magnetic field distribution map M1 respectively. S32. Extract the set of feature points P0 from the edge region of the magnetic field of the standard magnetic field distribution map M0, and search for the set of candidate feature points P1 corresponding to P0 in the actual magnetic field distribution map M1. The feature points are constructed based on the slope of the local magnetic flux density change. S33. Based on the feature point sets P0 and P1, obtain the deformation matrix T between the two maps, which is composed of translation parameters, rotation parameters and scale parameters. S34. Apply the deformation matrix T to the actual magnetic field distribution map M1 to obtain a registration image M1' that is completely aligned with the standard magnetic field distribution map M0 in spatial position. Calculate the pixel-level difference between M0 and M1' to form the registration difference magnetic field map ΔM.

5. The method for detecting motor slot wedge assembly based on magnetic field distribution characteristics according to claim 1, characterized in that: S4 includes: S41. Convert the pixel grayscale values ​​in the registered differential magnetic field map ΔM into a magnetic flux density difference matrix and establish a magnetic flux density offset mapping table. S42. Based on the stator structure parameters of the motor, extract the spatial regions corresponding to each slot in the registration difference magnetic field map ΔM to form a slot mask template. S43. Apply the slot mask template to the magnetic flux density offset mapping table, and extract the magnetic flux density offset value in the corresponding area of ​​each slot in the slot order. S44. Arrange the magnetic flux density offset values ​​of all slots in sequence to form a characteristic vector reflecting the magnetic field disturbance in the slot area.

6. The method for detecting motor slot wedge assembly based on magnetic field distribution characteristics according to claim 5, characterized in that: S5 includes: S51. The obtained magnetic field disturbance feature vector of the slot area is used as the input feature and input into the pre-trained assembly state discrimination model. The discrimination model is a classification model built based on multi-class support vector machine. S52. During the model training phase, manually labeled motor assembly samples are used as the training set. The feature dimension corresponds to the slot magnetic flux density offset value. The output labels include four states: normal, loose, skewed and missing. S53. After processing the feature vector of magnetic field intensity change in the slot area through the discrimination model, calculate the discrimination result of the state corresponding to each slot, and determine the final discrimination category based on the principle of the maximum probability of classification score. S54. The judgment results of each slot are combined into an assembly status label set. The assembly status label set S is a one-dimensional label array with the same number of elements as the number of slots.

7. The method for detecting motor slot wedge assembly based on magnetic field distribution characteristics according to claim 6, characterized in that: S6 includes: S61. Establish a mapping relationship between the assembly status mark set S and the slot number corresponding to the motor stator structural parameters, and generate a slot status lookup table. S62. According to the slot status comparison table, filter the slots with status labels of "loose", "skewed" or "missing", mark them as abnormal slots, and proceed to S63. S63. Encapsulate the number, assembly status type and spatial coordinates of each abnormal slot into abnormal information, output an alarm prompt, and proceed to S64. S64. Visualize all abnormal information and generate a structural inspection report.

Images