Double-sided visual detection method and system for motor stator with wiring terminal

Abstract

The invention discloses a double-sided visual detection method for a motor stator with a wiring terminal, and belongs to the technical field of motor stator detection equipment. The system has the advantages that a cooperative robot is integrated, all material transferring tasks including feeding grabbing, carrying, overturning and discharging are completed, the occupied area of equipment is effectively reduced, the hardware cost is reduced, the system structure is simplified, and the complexity of cooperative control of multiple robots is avoided. In addition, a rotatable bearing table and a plurality of cameras which are arranged in a surrounding mode are matched to collect images of the first end of the stator, and one camera which shoots the outer circumferential face of the stator is provided with a position sensor. The camera is controlled to move relative to the stator based on a signal, detected by the position sensor, of the wiring terminal so as to adjust the focal length to clearly image the wiring terminal, so that two imaging focal lengths are compatible on the same imaging station, and compared with a traditional mode that two sets of camera stations need to be arranged to correspond to different focal length requirements for respective imaging, the imaging efficiency is improved. And the hardware cost of a set of camera shooting stations is saved.

Description

Technical Field

[0001] This invention relates to the field of motor stator testing equipment, and in particular to a method and system for double-sided visual inspection of motor stators with terminals. Background Technology

[0002] Existing methods for inspecting motor stator quality mainly fall into two categories: electrical quantity testing and electromagnetic testing. Electrical quantity testing assesses motor quality by observing electrical parameters such as resistance and withstand voltage; electromagnetic testing uses electromagnetic methods such as eddy current and electromagnetic probes to identify motor defects. However, due to the insensitivity of electrical quantity testing to minor defects and the susceptibility of electromagnetic testing to interference from enameled wire, companies currently rely entirely on manual inspection of the motor stator's appearance for defect detection. This is especially true for minor defects, insulation components, winding conditions, and the appearance of the core, which is entirely manual and lacks effective automated methods.

[0003] For example, the stator of an automotive air conditioning compressor motor, or simply automotive air compressor motor stator, needs to be installed in the engine compartment or near the electric drive system of new energy vehicles. Compared with ordinary industrial motor stators, it has characteristics such as compact size, high slot fill factor, and complex and varied enameled wire winding methods. In addition, its high power density, high temperature of the iron core and windings under operating conditions, difficult heat dissipation, and frequent impacts mean that even a minor appearance defect can lead to a fatal failure. However, due to the small space, complex surface curvature, and numerous complex component structures of automotive air compressor motor stators, automated inspection is difficult. It usually relies on multiple detailed manual inspections, using visual observation to detect defects, which has problems such as high missed detection rate and low inspection efficiency.

[0004] Therefore, there is an urgent need for a device that can seamlessly integrate with existing production lines and achieve comprehensive, efficient, and non-destructive testing of stators with protruding terminals using a streamlined layout of a single workstation and a single robot. Summary of the Invention

[0005] This invention provides a double-sided visual inspection method for motor stators with terminals. It addresses the urgent need in the prior art for a device that seamlessly integrates with existing production lines, enabling comprehensive, efficient, and non-destructive inspection of stators with protruding terminals using a streamlined single-station, single-robot layout. The technical solution is as follows: On one hand, a double-sided visual inspection method for motor stators with terminals is provided, in which a single collaborative robot performs stator loading, flipping, and unloading at a fixed inspection station; the inspection station is equipped with a rotating platform, on which a support platform is mounted, and the upper surface of the support platform is provided with a guide slope inclined toward its central axis; the method includes: The stator is placed on the support platform, and the placement deviation of the stator is automatically corrected by the guide slope. The rotating stage is driven to rotate the stator, and multiple cameras arranged around it are controlled to simultaneously acquire images of the first end of the stator; wherein, a camera for capturing images of the outer circumferential surface of the stator is equipped with a position sensor and is configured to: in response to the sensing signal of the position sensor on the terminal block, when the terminal block enters its field of view, automatically drive the camera to retract radially along the stator to achieve focusing. The stator is grasped by the single collaborative robot and rotated 180° in the air before being placed back on the support platform; The rotating platform is driven to rotate again, and the multiple cameras are controlled to capture images of the second end after the stator is flipped. After the defect is determined based on the acquired double-sided images, the stator is removed and sorted by the single collaborative robot.

[0006] Optionally, before performing the step of placing the stator on the support platform, the method further includes the step of: The orientation of the wiring terminals of the incoming stator is automatically and mechanically calibrated.

[0007] Optionally, the defect determination based on the acquired double-sided images includes: For the double-sided image, a region segmentation operation based on color features is performed to separate the stator target in the image into multiple independent region images corresponding to its copper coil, binding wire, insulating paper and iron core components respectively; The multiple independent region images are input into a set of pre-trained neural network models for defect analysis; wherein each model is specifically trained for the defect features of one of the copper coil, binding wire, insulating paper or iron core components; If all neural network models analyze and determine that their corresponding independent region images are flawless, then the stator is determined to be a good product; otherwise, the good product is determined to be a defective product.

[0008] Optionally, performing color feature-based region segmentation on the double-sided image further includes: Extract the edge features of each independent region image to obtain the corresponding edge feature image; In the defect analysis step, the independent region image and its corresponding edge feature image are input together into the corresponding specialized neural network model for analysis.

[0009] On the other hand, a double-sided visual inspection system for motor stators is provided for implementing the above-described method. This system inspects motor stators with terminals and employs a single-station, single-robot architecture. The system includes: A single collaborative robot with a force-controlled adaptive gripper mounted at its end effector; A fixed testing station, which includes: a rotating stage; The support platform is installed on the rotating platform; the support platform has a hollow structure and its upper surface is provided with a guide slope inclined toward its central axis; Multiple cameras are arranged in a three-dimensional manner around the detection station; The variable-focus imaging unit includes a fifth camera for capturing images of the outer circumferential surface of the stator, a linear drive mechanism for driving the fifth camera to reciprocate along the radial direction of the stator, and a position sensor for sensing the position of the terminal block; the control terminal of the linear drive mechanism is connected to the signal output terminal of the position sensor. The controller is communicatively connected to the single collaborative robot, the rotating platform, each of the cameras and the variable-focus imaging unit, and is configured to perform a two-sided visual inspection method for motor stators with terminal blocks.

[0010] Optionally, the plurality of cameras include: a first camera located directly above the support platform, a second camera facing into the terminal hole on the stator, a third camera facing the side of the terminal along the radial direction of the stator, a fourth camera facing the binding position of the stator enameled wire winding, a fifth camera constituting the variable zoom imaging unit, and a sixth camera for photographing the inner diameter surface of the stator through the hollow structure of the support platform.

[0011] Optionally, the system further includes a straightening mechanism for automatically and mechanically straightening the orientation of the wiring terminals of the incoming stator.

[0012] Optionally, the straightening mechanism includes a guide forming a tapered channel and a lever located on the side of the channel entrance; the lever is used to contact and actuate the wiring terminals of the incoming stator.

[0013] Optionally, the support platform is a bracket with a Y-shaped hollow structure and a central positioning ring; the guide slope is the surface of multiple elastic guide members mounted on the bracket facing the axis of the central positioning ring.

[0014] Optionally, the controller consists of an industrial computer and a PLC; the industrial computer is configured to run a control program for implementing the above-mentioned double-sided visual inspection method for motor stators with terminals; the PLC is used to control the rotary table and the linear drive mechanism.

[0015] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) The double-sided visual inspection system for motor stators integrates a collaborative robot to complete all material transfer tasks from loading, picking, handling, flipping and unloading. This effectively reduces the equipment footprint, lowers hardware costs, simplifies the system structure, and avoids the complexity of multi-robot collaborative control. In addition, a rotatable support platform and multiple cameras arranged around it are used to collect images of the first end of the stator. One of the cameras that captures the outer circumference of the stator is equipped with a position sensor. Based on the signal of the terminal detected by the position sensor, the camera moves relative to the stator to adjust the focal length and clearly image the terminal. This allows for two imaging focal lengths to be compatible on the same imaging station, enabling adaptive and clear imaging of different outer diameter areas of the stator. Compared with the traditional method of setting up two sets of camera stations to correspond to different focal length requirements for imaging, this saves the hardware cost of one camera station and simplifies the overall structure of the visual inspection device for motor stators. A single collaborative robot grasps the stator and performs a 180-degree flip in mid-air. The stator is then placed back on the support platform, and the rotating platform is driven to rotate again. Multiple cameras are controlled to capture images of the second side of the stator after the flip, ultimately achieving double-sided imaging of the stator at a single inspection station. In summary, the cooperation of a single collaborative robot, a rotating platform, and multiple cameras can complete imaging of all parts of the stator. This optimizes the inspection process, shortens the transit time between different stations, and reduces the cumulative error caused by multiple positioning steps, thus improving inspection cycle time and accuracy.

[0016] (2) The orientation of the terminals D1 of the incoming stator D is automatically and mechanically calibrated. In this case, when a stator with protruding terminals enters the system along the stator conveyor, if the terminals of the incoming stator are not in the preset orientation, the stator is dynamically and continuously adjusted until the terminals rotate to a stable position, i.e., the terminals of the stator are in the preset orientation. Through this passive, purely mechanical method, the orientation of all incoming stators is dynamically and continuously corrected from a disordered orientation to a uniform orientation, thereby ensuring a high success rate of robot grasping, that is, ensuring that the collaborative robot can grasp from a known and safe angle.

[0017] (3) After acquiring the double-sided image of the motor stator, by performing a region segmentation operation based on color features, the stator target in the image can be separated into multiple independent region images corresponding to its copper coil, binding wire, insulating paper, and iron core components, respectively. This facilitates the determination of multiple matching neural network models based on the attribute information corresponding to each of the multiple independent region images. Then, the multiple independent region images are input into a set of pre-trained neural network models for defect analysis; wherein, each model is specifically trained for the defect features of one of the copper coil, binding wire, insulating paper, or iron core components; in this way, by inputting each independent region image into its matching neural network model, the defect detection of each independent region image can be performed by multiple neural network models, thereby improving the accuracy of the detection results of each independent region image. Finally, if all neural network models analyze and determine that their corresponding independent region images are free of defects, the stator is determined to be a good product; otherwise, it is determined to be a defective product. In this way, by combining the analysis results of multiple neural network models, the stator is determined to be a good product only when all neural network models analyze and determine that their corresponding independent region images are free of defects. This can improve the accuracy of defect detection of motor stators, thereby ensuring a high detection rate and low false alarm rate for various known and unknown minor defects, and thus improving the control of product quality.

[0018] (4) According to each independent region image, start the matching neural network model for each independent region image, so as to perform more targeted defect detection and improve the accuracy of defect detection. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a double-sided visual inspection method for a motor stator with wiring terminals provided in an embodiment of the present invention; Figure 2 This is a flowchart of another method for double-sided visual inspection of a motor stator with terminals provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the implementation process of a motor stator defect detection method based on region segmentation provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of another implementation process of a motor stator defect detection method based on region segmentation provided in an embodiment of the present invention; Figure 5This is a schematic diagram of the structure of a detection system provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a gripper provided in an embodiment of this application; Figure 7 This is a partial structural schematic diagram of a detection system provided in an embodiment of this application; Figure 8 yes Figure 7 A structural schematic diagram from another perspective is shown; Figure 9 This is a schematic diagram of the structure of the regularization mechanism provided in the embodiments of this application; Figure 10 This is a schematic diagram of the structure of the support platform provided in the embodiments of this application; Figure 11 This is a schematic diagram of the structure of a computer block device provided in an embodiment of the present invention.

[0021] The accompanying drawings illustrate specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0024] Please refer to Figures 1 to 11 , Figure 1 This is a flowchart of a double-sided visual inspection method for a motor stator with terminals provided in an embodiment of the present invention. Figure 2 This is a flowchart of another method for double-sided visual inspection of a motor stator with terminals provided in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the implementation process of a motor stator defect detection method based on region segmentation provided in an embodiment of the present invention. Figure 4This is a schematic diagram of another implementation process of a motor stator defect detection method based on region segmentation provided in an embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of a detection system provided in an embodiment of the present invention. Figure 6 This is a schematic diagram of the structure of a gripper provided in an embodiment of this application. Figure 7 This is a partial structural diagram of a detection system provided in an embodiment of this application. Figure 8 yes Figure 7 The diagram shows a structural design from another perspective. Figure 9 This is a schematic diagram of the structure of the regulating mechanism provided in the embodiments of this application. Figure 10 This is a schematic diagram of the structure of the support platform provided in the embodiment of this application. Figure 11 This is a schematic diagram of the structure of a computer block device provided in an embodiment of the present invention.

[0025] This invention provides a double-sided visual inspection method for motor stators with terminals. The stator can be a motor stator for an automotive air conditioning compressor or an industrial drive motor. A single collaborative robot 300 performs the loading, flipping, and unloading of the stator at a fixed inspection station. The inspection station is equipped with a rotating platform 202, on which a support platform 201 is mounted. The upper surface of the support platform 201 has a guide slope inclined towards its central axis. This double-sided visual inspection method for motor stators with terminals may include: Step 1: Place the stator on the support platform and use the guide ramp to automatically correct the placement deviation of the stator D; Step 2: Drive the rotating stage 202 to rotate the stator D, and control multiple cameras 203 arranged around it to simultaneously acquire images of the first end of the stator D. Among them, a camera X5 used to capture images of the outer circumferential surface of the stator is equipped with a position sensor Q1 and is configured to: respond to the sensing signal of the position sensor Q1 on the terminal D1, automatically drive the camera X5 to retract radially along the stator D to achieve focusing when the terminal D1 enters its field of view; Step 3: After the stator is picked up in the air by a single collaborative robot 300 and rotated 180°, it is placed back on the support platform 201. Step 4: Drive the rotating stage 202 to rotate again, and control multiple cameras 203 to acquire images of the second end after the stator flips. Step 5: After the defect judgment is completed based on the acquired double-sided image of stator D, a single collaborative robot 300 removes the stator and sorts it.

[0026] For example, in step 5 above, the method includes: detecting whether the sample stator is a good product by collecting image information; if the sample stator is detected to be a good product, controlling the collaborative robot 300 to cooperate with the stator clamping component to clamp the stator and place it on the good product conveyor belt. If a sample stator is detected to be defective, the collaborative robot 300 is controlled to work with the stator clamping device to clamp the stator and place it on the defective product conveyor belt.

[0027] In summary, this invention provides a method for double-sided visual inspection of motor stators with terminals. The double-sided visual inspection system integrates a collaborative robot to complete all material handling tasks, from loading and gripping, transporting, flipping, and unloading. This effectively reduces equipment footprint, lowers hardware costs, simplifies system structure, and avoids the complexity of multi-robot collaborative control. Furthermore, a rotatable platform and multiple cameras arranged around the stator are used to acquire images of the first end of the stator. One camera, which captures images of the outer circumference of the stator, is equipped with a position sensor. Based on signals detected by the position sensor regarding the terminals, the camera moves relative to the stator to adjust its focal length for clear imaging of the terminals. This allows for compatible imaging at two focal lengths within the same imaging station, enabling adaptive and clear imaging of different outer diameter areas of the stator. Compared to the traditional method requiring two sets of camera stations to meet different focal length requirements, this saves the hardware cost of one camera station and simplifies the overall structure of the visual inspection device for motor stators. A single collaborative robot grasps the stator and performs a 180-degree flip in mid-air. The stator is then placed back on the support platform, and the rotating platform is driven to rotate again. Multiple cameras are controlled to capture images of the second side of the stator after the flip, ultimately achieving double-sided imaging of the stator at a single inspection station. In summary, the cooperation of a single collaborative robot, a rotating platform, and multiple cameras can complete imaging of all parts of the stator. This optimizes the inspection process, shortens the transit time between different stations, and reduces the cumulative error caused by multiple positioning steps, thus improving inspection cycle time and accuracy.

[0028] Optionally, before placing the stator D on the support platform 201, the double-sided visual inspection method for motor stators with terminals D1 further includes step S1: automatically and mechanically adjusting the orientation of the terminals D1 of the incoming stator D. In this case, when a stator with protruding terminals enters the system along the stator conveyor line, if the terminals of the incoming stator are not in a preset orientation, the stator is dynamically and continuously adjusted until the terminals rotate to a stable position, i.e., the terminals of the stator are in the preset orientation. Through this passive, purely mechanical method, the orientation of all incoming stators is dynamically and continuously adjusted from disordered to a uniform orientation, thereby ensuring a high success rate of robot grasping, that is, ensuring that the collaborative robot can grasp from a known and safe angle.

[0029] In some embodiments, it can be achieved through Figure 3 The steps shown achieve "defect determination based on the acquired double-sided images": 301. For the double-sided image, perform a region segmentation operation based on color features to separate the stator target in the image into multiple independent region images corresponding to its copper coil, binding wire, insulating paper and iron core components respectively; Here, the double-sided images include: stator images from multiple angles of the motor stator and terminal images.

[0030] Optionally, the stator and terminal blocks are imaged using multiple image acquisition components in the stator appearance defect visual inspection system, resulting in stator and terminal block images from multiple perspectives. These multiple image acquisition components include multiple cameras positioned at different locations. By using multiple cameras positioned at different locations on the motor stator to acquire stator and corresponding terminal block images, richer double-sided images can be obtained. For example, a first camera acquires a front image of the motor stator; a second camera acquires an image of the interior of the holes in the corresponding terminal blocks; a third camera acquires an image of the surface of the terminal blocks; a fourth camera acquires an image of the binding position of the enameled wire windings of the motor stator; a fifth camera acquires an image of the outer circumference of the motor stator; and a sixth camera acquires an image of the inner diameter surface of the motor stator. The double-sided images include: a front image, a binding position image, an outer circumference image, and an inner diameter surface image; the terminal block images include: an image of the interior of the holes and a surface image. This allows for the acquisition of stator and terminal block images from various perspectives, thereby improving the accuracy of defect detection.

[0031] Optionally, the system-pre-made background image can be subtracted from the acquired frontal image, binding position image, circumferential outer side image, inner diameter surface image, hole interior image, and surface image to remove the background structure of the captured image, such as the residual part of the 612-stator stage not blocked by the stator, thereby obtaining a double-sided image.

[0032] Optionally, step 301 above can be implemented through the following process: First, extract the edge features of each independent region image to obtain the corresponding edge feature image; then, in the defect analysis step, input the independent region image and its corresponding edge feature image into the corresponding specialized neural network model for analysis; in this way, different neural network models are used for analysis of different independent region images, which can improve the accuracy of defect detection for each independent region image.

[0033] Optionally, semantic detection can be performed on the two-sided image first, and then region segmentation can be performed on the two-sided image to accurately achieve region segmentation of the two-sided image and obtain multiple independent region images.

[0034] Optionally, based on the color threshold in the double-sided image, the double-sided image is divided into a copper coil region, a binding wire region, an insulating paper region, and an iron core region; and edge lines are cut into the copper coil region, binding wire region, insulating paper region, and iron core region respectively to obtain copper coil edge lines, binding wire edge lines, insulating paper edge lines, and iron core edge lines; wherein, the multiple independent region images include: copper coil region, binding wire region, insulating paper region, iron core region, copper coil edge lines, binding wire edge lines, insulating paper edge lines, and iron core edge lines.

[0035] For example, for the imaging result image (i.e., double-sided image) of each camera station, the stator image is divided into copper coil area, binding wire area, insulating paper area and iron core area according to three color thresholds: yellow, white, gray-white and black-gray. Then, for the imaging result image (i.e., double-sided image) of each camera station, the edge lines of the divided stator copper coil area, binding wire area, insulating paper area and iron core area are cut and extracted as copper coil edge line, binding wire edge line, insulating paper edge line and iron core edge line, respectively.

[0036] 302. Input the multiple independent region images into a set of pre-trained neural network models for defect analysis.

[0037] Each model is specifically trained for a defect feature of one of the copper coils, binding wires, insulating paper, or iron core components.

[0038] Here, a matching neural network model is activated for each independent region image, which enables more targeted defect detection and improves the accuracy of defect detection.

[0039] Optionally, the metal material, surface structure information, texture information, and color information of multiple independent region images are determined; wherein the attribute information includes: metal material, surface structure information, texture information, and color information; and then, based on the metal material, surface structure information, texture information, and color information, a neural network model matching each independent region image is determined.

[0040] Here, surface structure information includes: frequency characteristics, defect scale, and deformation properties of individual region images. That is, by determining the frequency characteristics, defect scale, and deformation properties of multiple independent region images, this surface structure information is obtained, which facilitates the selection of a suitable network model for each independent region image. Texture information includes: the intensity of texture details, texture smoothness, and texture distribution. Color information includes: color uniformity and color values.

[0041] Optionally, by using information on metal material, surface structure, texture, and color, the required network structure, feature extraction method, semantic hierarchical modeling parameters, and computational speed for each independent region image are determined. Then, based on the network structure, feature extraction method, semantic hierarchical modeling parameters, and computational speed, a neural network model matching each independent region image is determined. Thus, by analyzing the metal material, surface structure, texture, and color information of each segmented region and its edge lines, the neural network model matching that region and edge line can be further analyzed. For example, based on the network structure, feature extraction method, semantic hierarchical modeling parameters, and computational speed, a residual network model matching the iron core region, a residual-feature pyramid network model matching the iron core edge line, a deep convolutional network model matching the copper coil region, a dual deep convolutional network model matching the copper coil edge line, a high-resolution network model matching the binding wire region, a dense edge detection network model matching the binding wire edge line, a segmentation transformer network model matching the insulation paper region, and a residual-feature pyramid network model matching the insulation paper edge line can be determined.

[0042] Here, the metal material, texture details, surface scratches, bumps or cracks in the iron core area are local fine-grained structures. A deep ResNet-101 deep convolutional neural network (i.e., residual network model) with strong feature extraction capabilities is adopted, and the network is optimized by drawing on the efficient network (EfficientNet-B3) to achieve lightweight and reduce processing time. The residual-feature pyramid network model (ResNet-FPN) is used in the edge line sub-region.

[0043] The copper coil area has a uniform color, and surface defects are mostly minor scratches, coil miswinding, overlap, distortion, and other structural issues. A deep convolutional network model (U-Net) with strong pixel-level structured representation recognition capabilities is used, and multi-scale dilated convolution is introduced to modify it. A dual deep convolutional network model (U²-Net) is used for the edge line sub-region. The insulating paper area has a smooth texture and uniform color, but defects are mostly damage, curling edges, and foreign matter attachment. A segmentation transformer network model (SegFormer-B2) sensitive to shape contours and regional consistency is used, and ResNet-FPN is used for the edge line sub-region. The binding wire area has extremely fine texture and a high-frequency regular structure. Defects mainly include looseness, fuzzing, and misaligned lines. A modified high-resolution network model (HRNet-W18) is used, and a dense edge detection network model (DexiNed) is used for the edge line sub-region.

[0044] Different segmented regions exhibit significant differences in texture distribution, frequency characteristics, defect scale, deformation properties, and image noise types. Therefore, differentiated strategies need to be adopted in network structure design, feature extraction methods, semantic hierarchical modeling capabilities, and computational efficiency.

[0045] The iron core region exhibits strong metallic material and texture details. Surface scratches, dents, or cracks are localized fine-grained structures. While the ResNet-101 network, with its large depth and wide receptive field, is suitable for extracting high-frequency local details from the metal surface, its computational cost is high. Therefore, EfficientNet-B3 is used for lightweight optimization to improve inference speed. The iron core edge lines are significantly affected by metal reflection and light source angle, making single-scale models prone to missed detections. Therefore, a feature pyramid network model capable of multi-layer feature fusion must be used.

[0046] Copper coils have a uniform color, strong surface reflectivity, and complex curved surfaces. Surface defects are mostly structural issues such as minor scratches, coil miswinding, overlap, and distortion. U-Net typically focuses on pixel-level structural representation; by incorporating multi-scale dilated convolutions, it can better capture cross-scale structural defects and is more sensitive to changes in the curved surface structure and topological relationships of coils. Copper wires have a tortuous structure and high density, with edge line complexity far exceeding that of hard metal edges. Therefore, a model like U²-Net, with multi-scale recovery and a two-level encoding-decoding structure, is needed to handle them.

[0047] Insulating paper has a smooth texture and uniform color. Defects typically manifest as shape damage, curled edges, missing corners, or foreign matter covering the surface. It requires the highest level of consistency in its overall outline. The SegFormer semantic segmentation model, belonging to the Transformer semantic segmentation model, does not rely on convolution and excels at capturing global consistency, thus better identifying these deformable and large-area structural changes. The edge features of insulating paper are macro-outlines, requiring less detailed local representation. Imperfections are mostly meso-scale deformations, making it suitable for multi-scale modeling using feature pyramid network models.

[0048] The binding lines have extremely fine textures and high-frequency regular structures. Defects are usually very small (such as slack points, fuzz, and broken lines), making them extremely sensitive to resolution. HRNet can sustainably maintain high-resolution features at multiple scales, preserving both high resolution and semantic features, making it sensitive to detail-oriented defects. The binding lines have extremely small widths and high texture frequencies. On high-frequency fine textures, DexiNed outputs multiple layers of features simultaneously, capturing very subtle edge perturbations.

[0049] In this embodiment of the invention, based on the significant differences in material properties, texture frequency, illumination response, geometric structure, and typical defect morphology among different regions, and the inability of a single model to simultaneously represent multiple features such as high-frequency metallic textures, curve topological changes, blocky deformations, and micro-linear structures, different neural network models are selected for different regions such as the iron core, copper coil, insulating paper, and binding wire. The surface of the iron core region exhibits high-frequency metallic textures, and its fine scratches and cracks are often only a few pixels wide. Therefore, a deep ResNet-101 (combined with the lightweight structure of EfficientNet) is used to enhance the ability to extract subtle details. The twisting, overlapping, and bending of the copper coil are structural topological changes rather than local texture changes. Therefore, U-Net, which has pixel-level structural reconstruction capabilities and enhances the multi-scale receptive field through dilated convolution, is used. Defects in insulating paper are mainly manifested as abrupt changes in macroscopic shape, such as blocky tears, missing corners, and curled edges. The SegFormer-B2 architecture of the Transformer can more accurately model contour consistency through a global self-attention mechanism. The binding wires are extremely fine and have high texture frequency. Minor breaks and slack need to maintain high-resolution features. Therefore, HRNet-W18 and DexiNed, which is specifically designed for the edges of fine lines, are used. Theoretically, the features of different regions have drastically different frequency distributions. High-frequency metallic fine lines, low-frequency paper structures, large-scale deformations, and sub-pixel fine lines cannot be optimally represented by the same model. Field experiments also verified that compared with the initial attempt to use a unified network for recognition (the false negative rate reached 11%-12%, and the false negative rate reached 25%), the region-customized model significantly reduced the false negative rate and improved the robustness to feature weakening, illumination changes, and deformation defects compared with a single general network. The false negative rate is close to 0, and the false negative rate is about 7%.

[0050] 303. If all neural network models analyze and determine that their corresponding independent region images are flawless, then the stator is determined to be a good product; otherwise, the stator is determined to be a defective product.

[0051] Here, a neural network model matching each independent region image is used to detect defects in that independent region image, which can yield multiple highly accurate detection results.

[0052] Optionally, the artificial intelligence automatic judgment model is activated, and a preset neural network model is loaded. The model judges the eight regions cut out from the imaging result image of each camera station: copper coil region, binding wire region, insulation paper region, iron core region, copper coil edge line, binding wire edge line, insulation paper edge line, and iron core edge line. If any one region is judged as defective by the neural network, the motor is judged as defective. The pre-built neural network model here comes from previous training. The training process is as follows: Based on a large batch of normal stators and a small batch of defective stators obtained in advance as training sets, the stators in the training set are photographed and imaged by a robot workstation system for detecting stator defects in automotive air compressor motors. Following the above steps, the image results obtained from each camera station are divided into eight subsets: copper coil area, binding wire area, insulation paper area, iron core area, copper coil edge line, binding wire edge line, insulation paper edge line, and iron core edge line. These subsets are then input into a pre-designed neural network artificial intelligence algorithm for training. The algorithm iterates repeatedly until convergence to obtain a neural network model used for discrimination.

[0053] Among the multiple detection results output by each neural network model, if any one of the detection result representation areas is judged as defective by the neural network, the motor is determined to be a defective motor, and the target detection result is output to the central control software.

[0054] In this embodiment of the invention, after acquiring a double-sided image of the motor stator, a region segmentation operation based on color features is performed to separate the stator target in the image into multiple independent region images corresponding to its copper coils, binding wires, insulating paper, and iron core components. This facilitates the determination of multiple matching neural network models based on the attribute information corresponding to each of the multiple independent region images. Then, the multiple independent region images are input into a set of pre-trained neural network models for defect analysis; each model is specifically trained for defect features of one of the copper coils, binding wires, insulating paper, or iron core components. Thus, by inputting each independent region image into its corresponding matching neural network model, multiple neural network models can be used to detect defects in each independent region image, thereby improving the accuracy of the detection results for each independent region image. Finally, if all neural network models analyze and determine that their corresponding independent region images are free of defects, the stator is determined to be a good product; otherwise, it is determined to be a defective product. In this way, by combining the analysis results of multiple neural network models, the stator is determined to be a good product only when all neural network models analyze and determine that their corresponding independent region images are free of defects. This can improve the accuracy of defect detection of motor stators, thereby ensuring a high detection rate and low false alarm rate for various known and unknown minor defects, and thus improving the control of product quality.

[0055] In some embodiments, it can also be achieved through Figure 4 The steps shown demonstrate how to detect defects in a double-sided image: 401. Use artificial intelligence algorithms to determine whether the motor stator is functioning properly.

[0056] This can be achieved through the following steps 411 to 415 (not shown in the diagram): 411. Subtract the system-preset background image from the acquired image to remove the background structure of the captured image.

[0057] 412. For the imaging results of each camera station, the stator image is divided into copper coil area, binding wire area, insulation paper area and iron core area according to three color thresholds: yellow, white, gray-white and black-gray.

[0058] 413. For the imaging results of each camera station, the edge lines of the split stator copper coil area, binding wire area, insulating paper area, and iron core area are cut and extracted into copper coil edge lines, binding wire edge lines, insulating paper edge lines, and iron core edge lines, respectively.

[0059] 414. By loading a preset neural network, the eight regions cut out from the imaging result image of each camera station are judged separately. If any one region is judged as defective by the neural network, the motor is judged as defective.

[0060] 415, output the result of whether the motor is faulty to the central control software.

[0061] 402, the robot will clamp the stator out of the detection module.

[0062] 403. Determine if the stator is defective.

[0063] If the stator is defective, proceed to step 404; if the stator is normal, proceed to step 405.

[0064] 404. If the stator is defective, the robot will place it on the defective product conveyor belt.

[0065] 405. If the stator is a normal product, the robot will place the stator on the normal product conveyor belt.

[0066] 406, determine if the user clicked "Stop Detection".

[0067] If the user clicks "Stop Detection", proceed to step 407; otherwise, the stator reaches and triggers the photoelectric sensor in a preset posture via the guide comb mechanism.

[0068] 407. If the user clicks "Stop Detection" to end the operation, the robot will return to the grasping posture.

[0069] In this embodiment of the invention, based on the rich image dataset obtained from multimodal, multi-camera, multi-angle redundant imaging, the stator is segmented into eight specialized subsets with surface and line features according to color characteristics. Neural network training is then performed on these subsets, improving the detection accuracy of the neural network algorithm for automotive air conditioning compressor stators—objects with unique structural composition, complex defect features, and three-dimensional surface structures. This specific customized training process and neural network training structure ensure a high detection rate for defects in automotive air conditioning compressor stators by the AI ​​algorithm, and achieve non-transferability and uniqueness for other types of objects, thus improving the algorithm's specificity. The targeted neural network training process ensures sensitivity to defects in automotive air conditioning compressor stators, and the customized process ensures that the AI ​​algorithm overcomes the accuracy decline problem caused by high transferability in general AI models, improving the long-term stability of defect detection accuracy. The entire system is software-defined; when changing product models, it can quickly adapt by mainly adjusting the robot program and vision algorithm, without large-scale hardware modifications. The equipment can autonomously complete tasks such as posture correction and automatic alignment, demonstrating a high level of intelligence and perfectly meeting the urgent needs of modern manufacturing for flexible and intelligent production.

[0070] like Figures 5 to 8 As shown in the figure, this application embodiment also provides a double-sided visual inspection system for motor stators to implement the above-mentioned double-sided visual inspection method for motor stators with terminals. This double-sided visual inspection system for motor stators is used to inspect motor stators with terminals and adopts a single-station, single-robot architecture. It may include: a single collaborative robot 300, whose end effector is equipped with a force-controlled adaptive gripper 400 for gripping and placing the stator D.

[0071] A fixed testing station may include: a rotating stage 202; The support platform 201 installed on the rotating platform 202 has a hollow structure and its upper surface is provided with a guide slope m inclined towards its central axis. Here, after the stator D contacts the guide slope m on the support platform 201, it can achieve self-centering positioning. In a purely mechanical and passive way, it cleverly solves the two major problems of positioning deviation and workpiece impact that are common in automated loading and unloading, and significantly improves the reliability, adaptability and protection capability of the equipment for the workpiece.

[0072] Multiple cameras (203) are arranged in a three-dimensional layout around the inspection station; The variable zoom imaging unit includes a fifth camera X5 for capturing images of the outer circumferential surface of the stator, a linear drive mechanism Q2 for driving the fifth camera X5 to reciprocate along the radial direction of the stator, and a position sensor Q1 for sensing the position of the terminal D1. The control end of the linear drive mechanism Q2 is connected to the signal output end of the position sensor Q1. The controller is communicatively connected to the individual collaborative robot 300, the rotating stage 202, each camera 203, and the variable-focus imaging unit, and is configured to execute the aforementioned double-sided visual inspection method for motor stators with terminal blocks. In this embodiment, the controller consists of an industrial computer and a PLC. The industrial computer is configured to run a control program for implementing the aforementioned double-sided visual inspection method for motor stators with terminal blocks. The PLC controls the rotating stage and the linear drive mechanism.

[0073] For example, a force-controlled adaptive gripper 400 is used to sense whether the cylindrical stator is being gripped by the robot in the correct posture. This gripper can be configured with different end grippers to adapt to different stator models. Force control feedback is used to judge the force after gripping. If the force exceeds the effective range, the robot is alerted that there is an abnormal gripping of the motor; if the force is below the effective range, the collaborative robot is alerted that the gripper is not gripping properly. Simultaneously, the collaborative robot's end gripper innovatively wraps easily replaceable medical-grade self-adhesive gauze S2, eliminating the need for chemical adhesives to avoid contaminating the motor stator. The medical-grade self-adhesive gauze S2 also increases the friction between the robot and the stator, reducing the risk of deflection during stator gripping. Furthermore, the medical-grade self-adhesive gauze S2 can absorb oil contamination brought in from upstream processes on the stator surface, objectively reducing the probability of oil contamination on the stator core's outer periphery.

[0074] In this embodiment, a collaborative robot 300 is integrated into the double-sided visual inspection system for motor stators. This robot completes all material handling tasks, from loading and gripping to transporting, flipping, and unloading. This effectively reduces the equipment footprint, lowers hardware costs, simplifies the system structure, and avoids the complexity of multi-robot collaborative control. Furthermore, a rotatable platform 201 and multiple cameras 203 arranged around the stator are used to acquire images of the first end of the stator D. One fifth camera X5, which captures images of the outer circumference of the stator, is equipped with a position sensor Q1. The controller, based on the signal from the terminal detected by the position sensor Q1, controls the fifth camera X5 to move relative to the stator via a linear drive mechanism Q2 to adjust the focal length for clear imaging of the terminal. This allows for the compatibility of two imaging focal lengths at the same imaging station, enabling adaptive and clear imaging of different outer diameter areas of the stator's outer circumference. Compared to the traditional method of requiring two sets of camera stations to meet different focal length requirements, this saves the hardware cost of one camera station and simplifies the overall structure of the visual inspection device for motor stators. A single collaborative robot 300 grasps the stator and performs a 180-degree flip in mid-air. The stator is then placed back on the support platform 201, and the rotating platform 202 is driven to rotate again. Multiple cameras are controlled to capture images of the second end of the stator after the flip, ultimately achieving double-sided imaging of the stator at a single inspection station. In summary, the cooperation of a single collaborative robot 300, the rotating platform 202, and multiple cameras 203 can complete imaging of all parts of the stator. This optimizes the inspection process, shortens the transit time between different inspection stations, and reduces the cumulative error caused by multiple positioning steps, thus improving inspection cycle time and accuracy.

[0075] In summary, this application provides a double-sided visual inspection system for motor stators, which may include a single collaborative robot and a fixed inspection station. The system integrates a single collaborative robot to complete all material handling tasks, from loading and gripping, transporting, flipping, and unloading. This effectively reduces equipment footprint, lowers hardware costs, simplifies system structure, and avoids the complexity of multi-robot collaborative control. Furthermore, a rotatable platform and multiple cameras arranged around the stator are used to acquire images of the first end of the stator. A fifth camera, which captures images of the outer circumference of the stator, is equipped with a position sensor. The controller, based on signals from the terminal blocks detected by the position sensor, controls the camera's movement relative to the stator via a linear drive mechanism to adjust the focal length for clear imaging of the terminal blocks. This allows for compatible imaging focal lengths at the same imaging station, enabling adaptive and clear imaging of different outer diameter areas of the stator's outer circumference. Compared to the traditional method requiring two sets of camera stations to meet different focal length requirements, this saves the hardware cost of one camera station and simplifies the overall structure of the visual inspection device for motor stators. A single collaborative robot grasps the stator and performs a 180-degree flip in mid-air. The stator is then placed back on the support platform, and the rotating platform is driven to rotate again. Multiple cameras are controlled to capture images of the second side of the stator after the flip, ultimately achieving double-sided imaging of the stator at a single inspection station. In summary, the cooperation of a single collaborative robot, a rotating platform, and multiple cameras can complete imaging of all parts of the stator. This optimizes the inspection process, shortens the transit time between different stations, and reduces the cumulative error caused by multiple positioning steps, thus improving inspection cycle time and accuracy.

[0076] For example, position sensor Q1 is used to detect the distance between the stator and the camera position of camera X5 during rotation. When an external protrusion (such as the lead terminal D1 housing) on ​​the outer circumference of the stator rotates to a position in front of camera X5, the distance measurement result of position sensor Q1 changes. The controller determines the amount of motion displacement required for linear drive mechanism Q2 to drive camera X5 based on the first sensing position signal returned by position sensor 400 and by looking up the built-in mapping table. The control component controls linear drive mechanism 300 to drive camera X5 quickly away from stator D to a position where clear imaging is possible, thereby rapidly shifting the focus of camera X5 backward and achieving clear imaging of the external protrusion (such as the lead terminal D1 housing). When the external protrusion on the outer circumference of the stator rotates away from the position in front of camera X5, position sensor Q1 detects that the distance has returned to normal and sends a second signal. The controller controls linear drive mechanism Q2 to drive camera X5 back to its original position based on the second signal. This allows for the compatibility of two imaging focal lengths on the same imaging station, enabling adaptive and clear imaging of different outer diameter regions on the stator's outer perimeter. Compared to the traditional method that requires two sets of camera stations to meet different focal length requirements for separate imaging, this saves the hardware cost of one camera station and simplifies the overall structure of the visual inspection device used for motor stators.

[0077] Optional, such as Figure 7 and Figure 8 As shown, the multiple cameras 203 in the detection system include: a first camera X1 located directly above the support platform 201; a second camera X2 facing the terminal hole on the stator; a third camera X3 located on the right side of the stator and radially towards the side of the terminal; a fourth camera X4 located on the right side of the stator and facing the binding position of the stator enameled wire winding; a fifth camera X5 located on the left side of the stator and forming a variable zoom imaging unit; and a sixth camera X6 used to photograph the inner diameter surface of the stator through the hollow structure of the support platform 201, located on the left side of the stator and facing the inner diameter surface of the stator.

[0078] Here, the fifth camera X5 can be positioned behind the stator D. The visual inspection device for the motor stator includes a light source component Y1 positioned between the fifth camera X5 and the sample stator D. The position sensor Q1 can be mounted on the side of the light source component Y1. For example, the light source component Y1 is typically a strip light source or a coaxial light source, used in conjunction with the camera X5. It illuminates the outer surface of the stator D at a specific angle, effectively highlighting scratches, pits, and other defects on the metal surface, providing optimal illumination conditions for outer surface inspection.

[0079] Here, the first camera X1 is mainly used to capture images of the upper surface of stator D to detect the regularity of the winding arrangement, scratches, bumps, dirt on the core end face, and the condition of the insulating paper; the second camera X2 is specifically used to detect whether there are foreign objects, burrs, or processing defects inside the terminal holes, which is a key detail that is difficult to observe from a normal angle; the third camera X3 is used to capture images of the side and surface of the terminals to detect whether their appearance is damaged, deformed, or contaminated; the fourth camera X4 is used to detect whether the binding wire is firm, whether it is broken or loose, and whether the windings in the binding area are neat; the sixth camera X6 uses a shooting method similar to an industrial endoscope or a large-angle overhead shot to clearly capture the inner diameter surface of the stator. During the rotation of stator D, the sixth camera X6 acquires 360° images of the inner diameter to detect whether the inner hole insulation coating is intact and whether there are scratches or foreign objects.

[0080] In this invention, the visual inspection device for a motor stator includes: a first light source component Y2 mounted above a support platform 201 and a second light source component Y3 mounted below the stator (i.e., the stator support position) on the support platform 201. The first light source component Y2, the second light source component Y3, and the light source component Y1 cooperate to provide illumination for each camera in a plurality of image acquisition components. For example, the first light source component Y2 is typically a ring light or a dome light source, providing uniform, shadowless illumination for the first camera X1 directly above it, and also providing sufficient and appropriately angled supplementary lighting for the second camera X2 and the sixth camera X6 shooting from an angled upward, to highlight the features of the target area. It should be noted that the first light source component Y2 has a cutout area for the first camera X1, the second camera X2, and the sixth camera X6 to capture images. The second light source component Y3 is typically a flat backlight or an upward-illuminating ring light source. The second light source component Y3 mainly serves as bottom supplementary lighting and contour illumination, providing a uniform background light for the entire imaging area, helping to eliminate bottom shadows, and can cooperate with other cameras to detect contour dimensions or burrs.

[0081] Here, multiple cameras and three light source components form six camera stations. The first camera station consists of the first camera X1, the first light source component Y2, and the second light source component Y3; the second camera station consists of the second camera X2 and the first light source component Y2; the third camera station consists of the third camera X3, the first light source component Y2, and the second light source component Y3; the fourth camera station consists of the fourth camera X4, the first light source component Y2, and the second light source component Y3; the fifth camera station consists of the fifth camera X5, the first light source component Y2, and the light source component Y1; and the sixth camera station consists of the sixth camera X6, the first light source component Y2, and the second light source component Y3. For example, the first camera station can be turned off after taking one image when the platform 201 starts rotating; the second camera station takes one image every 51.42° from the start to the end of the rotation of the platform 201, for a total of 7 images per revolution, and is turned off when the platform 201 finishes rotating. The third camera station takes one image every 120° from the start to the end of the rotation of the support platform 201, for a total of 3 images per rotation. It is turned off when the support platform 201 finishes rotating. The fourth camera station takes one image every 45° from the start to the end of the rotation of the support platform 201, for a total of 8 images per rotation. It is turned off when the support platform 201 finishes rotating. The fifth camera station takes one image every 180° from the start to the end of the rotation of the support platform 201. An additional image is taken after the photoelectric position sensor is triggered by a protruding part of the stator (such as a lead terminal housing), causing the linear drive mechanism to move. A total of 3 images are taken per rotation. It is turned off when the support platform 201 finishes rotating. The sixth camera station takes one image every 51.42° from the start to the end of the rotation of the support platform 201, for a total of 7 images per rotation. It is turned off when the support platform 201 finishes rotating.

[0082] In the embodiments of this application, such as Figure 9As shown, the detection system may further include a straightening mechanism 500 for automatically and mechanically straightening the orientation of the terminals of the incoming stator. For example, the straightening mechanism 500 may include a guide 501 forming a tapered channel T and a lever 502 located on the side of the channel T inlet k1. The lever 502 is used to contact and actuate the terminals D1 of the incoming stator D. That is, the position of the lever 502 is matched with the contour of the inner wall m1 of the channel T and is set such that when the terminals D1 of the incoming stator D are in a non-preset orientation, they contact the lever 502 to force the stator D to rotate under the frictional force of the stator conveyor line 100, until the terminals D1 rotate to a preset orientation that avoids interference under the limit of the inner wall m1 of the channel T. For example, the channel T is also provided with an outlet k2 opposite to the inlet k1. In this scenario, when a stator with protruding terminals enters the mechanism via the stator conveyor, if the terminals are not in a preset orientation, they will rotate under the forward friction force provided by the conveyor belt after contacting the toggle block 502. This ensures that the terminals do not interfere with the sidewall of the guide channel. As the stator continues to be conveyed, it is dynamically and continuously toggled by the inner wall of the channel until the terminals rotate to a stable position, i.e., the terminals are in the preset orientation. Through this passive, purely mechanical method, the orientation of all incoming stators is dynamically and continuously corrected from a disordered state to a uniform orientation, thereby ensuring a high success rate for robot grasping, and ensuring that the collaborative robot can grasp from a known and safe angle.

[0083] Here, the guide member 501 in the stator terminal orientation adjustment mechanism includes a pair of opposing side plates, the gap between which forms a gradually narrowing channel T. One of the side plates can be placed vertically, while the other is placed at an angle, forming a channel whose width gradually decreases from the direction away from the feed inlet k1. Additionally, the guide member 501 may also include a guide side plate arranged parallel to the vertically placed side plate, forming a guide channel between the guide side plate and the side plate, which is clearance-fitted with the outer diameter of the stator.

[0084] For example, such as Figure 5 As shown, the straightening mechanism 500 can be installed on the conveyor line 100, and a stator detection sensor S1 electrically connected to the controller is provided at the conveyor line 100. This control component can be configured to: upon receiving the position signal of the stator D detected by the stator detection sensor S1, control the assisting robot 300 to cooperate with the stator clamping component 400 to clamp the stator D and transfer it to the stator bearing position. In this way, through the cooperation of the stator detection sensor S1 and the collaborative robot 300, the efficiency and accuracy of stator D transfer are improved, and the degree of automation of stator D transfer is enhanced.

[0085] Optional, such as Figure 5 and Figure 10 As shown, the support platform 201 is a bracket with a Y-shaped hollow structure L and a central positioning ring 201a. The guide slope m is the surface of multiple elastic guide members 201b mounted on the bracket facing the axis of the central positioning ring 201a. It should be noted that the Y-shaped hollow structure L can provide an illumination channel for the light source component at the bottom. For example, the bracket may include: a frame 201c, three connecting arms 201d, and a central positioning ring 201a. The three connecting arms 201d are arranged in a Y-shape. One end of each connecting arm 201d is securely connected to the outer circumferential surface of the central positioning ring 201a, and the other end of the connecting arm 201d is securely connected to the inner ring side of the frame 201c. There are three elastic guide members 201b, which are respectively fixed to the ends of the three connecting arms 201d that connect to the outer circumferential surface of the central positioning ring 201a. Among them, two of the three connecting arms 201d can be symmetrically arranged about the extension line of the other connecting arm 201d, and the area enclosed by the three connecting arms 201d and the frame 201c is a Y-shaped hollow area L.

[0086] Before using the stator appearance defect detection system to inspect the stator, the system typically needs to perform the following steps: S10: After the motor stator defect detection workstation is powered on, an integrated electronic control and algorithm host system is started. This host system is an industrial control computer host responsible for the motion control of the collaborative robot and the intelligent detection of stator defects. It is accompanied by a PLC controller that communicates with it and performs electronic control functions. After the industrial control computer host completes its self-test, the automatic motion control software completes the communication self-test with the collaborative robot, the force-controlled adaptive gripper, and various photoelectric sensors; the PLC controller completes the communication self-test with each drive mechanism. If a communication abnormality is found after the self-test, the indicator light on the top of the system's equipment body flashes red and an alarm sound is emitted to prompt the user to manually check the machine for faults and troubleshoot communication wiring problems; if there is no communication abnormality, the indicator light on the top of the system's equipment body remains on yellow, and the reset button embedded in the outer shell of the system's equipment body flashes yellow, indicating that the machine has entered the resettable state.

[0087] S20: When the user presses and holds the reset button embedded in the equipment casing, the reset button's flashing yellow light changes to a solid yellow light. The PLC controller sends a command to control the rotary drive component to rotate >1 revolution and ≤2 revolutions until it reaches the zero position; it also controls the linear drive mechanism's linear motor to return to the zero position near the stator placement position. If a problem occurs during the reset, the indicator light on top of the equipment body flashes red, prompting the user to manually check for machine malfunctions and troubleshoot motor problems. If the reset is successful, after completion, the indicator light on top of the equipment body remains solid green, the yellow light on the reset button embedded in the equipment casing goes out, and the start button embedded in the equipment casing flashes green, prompting the user to start the machine and enter the reset operation in the central control software.

[0088] S30: When the user presses the start button embedded in the device's casing, the button's flashing green light changes to a solid green light. On the main control software interface of the industrial control computer, all buttons except "Start Detection" and "Stop Detection" unlock, and the reset robot button changes from disabled to enabled. The user then clicks the "Reset Robot" button on the touchscreen. The robot automatically starts resetting to the reset posture to prevent it from being in a faulty or other uncontrollable position during the last shutdown, ensuring a controllable starting point for movement. After the robot reaches the reset posture, the "Reset Robot" button automatically updates to the "Initialize Gripper" button. If the robot cannot reach the reset posture due to obstacles, the robot servo controller reports an error, and the industrial control computer displays a robot error message, prompting the user to manually check for robot faults and troubleshoot any obstacles.

[0089] S40: When the user clicks the "Initialize Gripper" button on the touchscreen, the force-controlled adaptive electric gripper module starts a self-test program, clears any fault information that may have existed in the previous run, updates the status, and then begins to rotate one full turn to return to the 0° position. During the rotation, it opens once, closes once, and then opens again to update the closed position and the maximum open position, and updates the force control feedback value when closing to adapt to any newly replaced gripper fingers and the thickness of the self-adhesive bandage gauze. After completing the above actions, the force-controlled adaptive electric gripper module enters the standby state, and the "Start Detection" button on the main control software interface of the industrial control computer changes from the disabled state to the enabled state. The "Initialize Gripper" button is automatically updated and displayed as the "Reset Robot" button again. S50: When the user clicks "Start Detection" on the touchscreen, the "Stop Detection" button on the main control software interface of the industrial control computer changes from disabled to enabled, and the "Reset Robot" button changes from enabled to disabled. The collaborative robot moves to the gripping posture and waits for the stator from the previous production process at the material waiting station.

[0090] S60: After the stator is completed in the previous production process, it is inverted on the production line conveyor belt at the waiting station. When it reaches the stator detection sensor, it triggers the photoelectric sensing signal, prompting the collaborative robot that a stator to be measured has arrived.

[0091] S70: The collaborative robot starts from the grasping posture, moves to the position of the stator, and then activates the force-controlled adaptive electric gripper module to grasp the stator that has reached the position of the stator detection sensor on the production line conveyor belt. After the gripper closes, the industrial control computer reads and judges whether the force control feedback value of the force-controlled adaptive electric gripper module is normal. If it is not normal, it means that the grasping has failed. The main control software interface issues a prompt to remind the user to correct the motor posture. The robot pauses and waits for the user to confirm the motor posture and click the "OK" button to continue. If the force control feedback value is normal, the robot grips the stator and moves it to send the stator into the six-camera full-posture vision inspection module and place it on the platform.

[0092] The stator appearance defect visual inspection system embodiment and stator appearance defect visual inspection method embodiment provided in this invention can be referred to each other, and the embodiments of this invention will not be described again here.

[0093] Optionally, the transmission medium can be a wired link (e.g., but not limited to, coaxial cable, optical fiber, and Digital Subscriber Line (DSL)) or a wireless link (e.g., but not limited to, Wireless Fidelity (WIFI), Bluetooth, and mobile block device networks). It should be noted that the control block device provided in the above embodiments is only an example illustrating the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer block device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the method embodiments provided in the above embodiments belong to the same concept, and their specific implementation processes are detailed in the method embodiments, and will not be repeated here.

[0094] Figure 11 This is a schematic diagram of the structure of a computer block device provided in an embodiment of the present invention. For example, as shown... Figure 11 As shown, the computer block device 1100 includes: a memory 1101, a processor 1102, and a computer program 1103 stored in the memory 1101 and running on the processor 1102. When the processor 1102 executes the computer program 1103, the computer block device can execute any of the aforementioned region segmentation-based motor stator defect detection methods.

[0095] Furthermore, embodiments of the present invention also protect a control block device, which may include a memory and a processor. The memory stores executable program code, and the processor is used to call and execute the executable program code to perform a motor stator defect detection method based on region segmentation provided by the embodiments of the present invention. Embodiments of the present invention can divide the control block device into functional modules according to the above method examples. For example, each module may correspond to a specific function, or two or more functions may be integrated into a processing module. The integrated module can be implemented in hardware. It should be noted that the module division in the embodiments of the present invention is illustrative and only represents a logical functional division; other division methods may exist in actual implementation. It should be noted that all relevant content of each step involved in the above method embodiments can be referenced to the functional description of the corresponding functional module, and will not be repeated here. It should be understood that the control block device provided by the embodiments of the present invention is used to execute the above-mentioned motor stator defect detection method based on region segmentation, and therefore can achieve the same effect as the above-described implementation method. When using integrated units, the control block device may include a processing module and a storage module. When the control block device is applied to a block device, the processing module can be used to control and manage the actions of the block device. The storage module can be used to support the block device in executing mutual program code, etc. The processing module can be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of Digital Signal Processing (DSP) and a microprocessor, etc., and the storage module can be a memory.

[0096] Furthermore, the control block device provided in the embodiments of the present invention may specifically be a chip, component, or module. The chip may include a connected processor and a memory. The memory stores instructions, and when the processor calls and executes the instructions, the chip can execute the region-segmentation-based motor stator defect detection method provided in the above embodiments. The embodiments of the present invention also provide a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the aforementioned method steps to implement the region-segmentation-based motor stator defect detection method provided in the above embodiments.

[0097] This invention also provides a computer program product. When the computer program product is run on a computer, it causes the computer to perform the aforementioned related steps to implement the motor stator defect detection method based on region segmentation provided in the above embodiments. The control block device, computer-readable storage medium, computer program product, or chip provided in this invention are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they achieve can be referred to in the beneficial effects of the corresponding methods provided above, and will not be repeated here. Through the description of the above embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the control block device can be divided into different functional modules to complete all or part of the functions described above. In the embodiments provided by this invention, it should be understood that the disclosed control block device and method can be implemented in other ways. For example, the control block device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another control block device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between control block devices or units may be electrical, mechanical, or other forms.

[0098] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multiple task processing and parallel processing are possible or may be advantageous. The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. The above content is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be covered within the protection scope of the present invention.

[0099] In this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The term "a plurality" refers to two or more unless otherwise expressly defined.

[0100] The above description is merely an optional embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for double-sided visual inspection of a motor stator with terminals, characterized in that, A single collaborative robot performs stator loading, flipping, and unloading at a fixed inspection station; the inspection station is equipped with a rotating platform, on which a support platform is mounted, the upper surface of which has a guide slope inclined toward its central axis; the method includes: The stator is placed on the support platform, and the placement deviation of the stator is automatically corrected by the guide slope. The rotating stage is driven to rotate the stator, and multiple cameras arranged around it are controlled to simultaneously acquire images of the first end of the stator; wherein, a camera for capturing images of the outer circumferential surface of the stator is equipped with a position sensor and is configured to: in response to the sensing signal of the position sensor on the terminal block, when the terminal block enters its field of view, automatically drive the camera to retract radially along the stator to achieve focusing. The stator is grasped by the single collaborative robot and rotated 180° in the air before being placed back on the support platform; The rotating platform is driven to rotate again, and the multiple cameras are controlled to capture images of the second end after the stator is flipped. After the defect is determined based on the acquired double-sided images, the stator is removed and sorted by the single collaborative robot.

2. The method according to claim 1, characterized in that, Before performing the step of placing the stator on the support platform, the method further includes the step of: The orientation of the wiring terminals of the incoming stator is automatically and mechanically calibrated.

3. The method according to claim 1, characterized in that, The defect determination based on the acquired double-sided images includes: For the double-sided image, a region segmentation operation based on color features is performed to separate the stator target in the image into multiple independent region images corresponding to its copper coil, binding wire, insulating paper and iron core components respectively; The multiple independent region images are input into a set of pre-trained neural network models for defect analysis; wherein each model is specifically trained for the defect features of one of the copper coils, binding wires, insulating paper or iron core components. If all neural network models analyze and determine that their corresponding independent region images are flawless, then the stator is determined to be a good product; otherwise, the good product is determined to be a defective product.

4. The method according to claim 3, characterized in that, The step of performing a region segmentation operation based on color features on the double-sided image further includes: Extract the edge features of each independent region image to obtain the corresponding edge feature image; In the defect analysis step, the independent region image and its corresponding edge feature image are input together into the corresponding specialized neural network model for analysis.

5. A double-sided visual inspection system for motor stators for implementing the method according to any one of claims 1-4, characterized in that, Used for inspecting motor stators with terminal blocks, employing a single-station, single-robot architecture, including: A single collaborative robot with a force-controlled adaptive gripper mounted at its end effector; A fixed testing station, which includes: a rotating stage; The support platform is installed on the rotating platform; the support platform has a hollow structure and its upper surface is provided with a guide slope inclined towards its central axis; Multiple cameras are arranged in a three-dimensional manner around the detection station; The variable-focus imaging unit includes a fifth camera for capturing images of the outer circumferential surface of the stator, a linear drive mechanism for driving the fifth camera to reciprocate along the radial direction of the stator, and a position sensor for sensing the position of the terminal block; the control terminal of the linear drive mechanism is connected to the signal output terminal of the position sensor. The controller is communicatively connected to the single collaborative robot, the rotating platform, each of the cameras, and the variable-focus imaging unit, and is configured to perform the method of claim 1.

6. The system according to claim 5, characterized in that, The plurality of cameras include: a first camera located directly above the support platform; a second camera facing into the terminal hole on the stator; a third camera facing the side of the terminal along the radial direction of the stator; a fourth camera facing the binding position of the stator enameled wire winding; a fifth camera constituting the variable zoom imaging unit; and a sixth camera for photographing the inner diameter surface of the stator through the hollow structure of the support platform.

7. The system according to claim 5, characterized in that, The system also includes a straightening mechanism for automatically and mechanically straightening the orientation of the wiring terminals of the incoming stator.

8. The system according to claim 7, characterized in that, The regulating mechanism includes a guide forming a tapered channel and a lever located on the side of the channel entrance; the lever is used to contact and move the wiring terminals of the incoming stator.

9. The system according to claim 5, characterized in that, The support platform is a bracket with a Y-shaped hollow structure and a central positioning ring; the guide slope is the surface of multiple elastic guide members installed on the bracket facing the axis of the central positioning ring.

10. The system according to claim 5, characterized in that, The controller consists of an industrial computer and a PLC; the industrial computer is configured to run a control program for implementing the method of any one of claims 1-4; the PLC is used to control the rotary table and the linear drive mechanism.

Images