Product testing methods and devices

Abstract

This disclosure provides a product testing method and apparatus, relating to the field of product testing technology. The product testing method includes: acquiring first multimodal data of a device under test; processing the first multimodal data according to the processing method corresponding to the material of the device under test to obtain second multimodal data; and comparing the second multimodal data with a digital standard model of the device under test to obtain the testing result of the device under test.

Description

Technical Field

[0001] This disclosure relates to the field of product testing technology, specifically to a product testing method and apparatus. Background Technology

[0002] Silicon carbide and graphite materials are widely used in semiconductors, new energy, and other fields due to their high temperature resistance, corrosion resistance, and excellent electrical and thermal conductivity. The appearance defects (such as scratches, cracks, and missing corners) and three-dimensional dimensions (such as length, pore size, and depth) of silicon carbide and graphite products directly affect their performance.

[0003] Currently, conventional testing methods suffer from problems such as limited testing dimensions, low automation, and poor material compatibility, making it impossible to simultaneously meet the appearance and size testing needs of products made of different materials. Summary of the Invention

[0004] In view of this, the present disclosure provides a product testing method and apparatus to achieve high-precision and high-reliability automated testing of products made of different materials.

[0005] In a first aspect, one embodiment of this disclosure provides a product testing method, comprising: acquiring first multimodal data of a device to be tested; processing the first multimodal data according to the processing method corresponding to the material of the device to be tested to obtain second multimodal data; and comparing the second multimodal data with a digital standard model of the device to be tested to obtain a test result of the device to be tested.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the material of the device under test includes silicon carbide, the first multimodal data includes first image data and first point cloud data, and the second multimodal data includes second image data and second point cloud data; wherein, according to the processing method corresponding to the material of the device under test, the first multimodal data is processed to obtain the second multimodal data, including: performing adaptive illumination compensation on the first image data and enhancing the defect features of the first image data to obtain the second image data; correcting the offset points in the first point cloud data and using an interpolation algorithm to fill the point cloud holes caused by material reflection to obtain the second point cloud data.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, adaptive illumination compensation is performed on the first image data, and the defect features of the first image data are enhanced to obtain the second image data, including: dividing the first image data into multiple local regions; adjusting the exposure parameters of each of the multiple local regions, and performing defect enhancement processing on the adjusted multiple local regions to obtain the second image data.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the material of the device to be tested includes graphite, the first multimodal data includes first image data, and the second multimodal data includes second image data; wherein, according to the processing method corresponding to the material of the device to be tested, the first multimodal data is processed to obtain the second multimodal data, including: performing image opening and closing operations on the first image data to filter out surface powder of the device to be tested, thereby obtaining the second image data.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the digital standard model includes an appearance feature library and a size standard library, and the second multimodal data includes second image data and second point cloud data; wherein, comparing the second multimodal data with the digital standard model of the device under test to obtain the test result of the device under test includes: comparing the second image data with the appearance feature library to identify appearance defects; establishing a three-dimensional model of the device under test based on the second point cloud data, and comparing the three-dimensional model with the size standard library to calculate the size deviation; and determining the test result based on the appearance defects and the size deviation.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the material of the device to be tested includes graphite, and the appearance feature library includes normal texture features of graphite material; comparing the second image data with the appearance feature library to identify appearance defects includes: performing feature extraction operations on the second image data to obtain image features of texture regions in the second image data; calculating the similarity between the image features and normal texture features, and determining that the texture region includes appearance defects if the similarity is less than a similarity threshold; wherein, the similarity threshold is dynamically adjusted based on the surface roughness of the device to be tested.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, the material of the device to be tested includes graphite, and the product testing method further includes: determining the surface texture features of the three-dimensional model, the surface texture features including surface roughness and surface texture depth; based on the surface texture features, determining the suspected crack region in the three-dimensional model containing suspected cracks; extracting the three-dimensional cross-sectional curve of the suspected crack region, and determining the depth value and length value of the suspected crack based on the three-dimensional cross-sectional curve; if the depth value is greater than a first threshold and the length value is greater than a second threshold, determining that the suspected crack region includes appearance defects.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the material of the device under test includes silicon carbide, and the first multimodal data includes first image data and / or first point cloud data; wherein, acquiring the first multimodal data of the device under test includes: determining the material hardness of the device under test; using a mapping relationship model between silicon carbide hardness and a set of scanning parameters, determining a target scanning parameter set corresponding to the device under test based on the material hardness, wherein the target scanning parameter set includes at least one of the laser power, scanning frequency, and point cloud sampling density of the scanning device; acquiring the first point cloud data based on the target scanning parameter set; and / or determining the target polarization angle based on the surface reflection angle of the device under test; adjusting the polarization filter of the image acquisition device based on the target polarization angle, and acquiring the first image data.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, the product testing method also includes: obtaining the design drawings of the device to be tested; extracting dimensional annotation parameters and appearance requirement parameters from the design drawings, and constructing a digital standard model.

[0014] Secondly, one embodiment of this disclosure provides a product testing apparatus, comprising: a workbench for supporting and fixing the device to be tested; a data acquisition module for acquiring multimodal data of the device to be tested; and a control module for executing the product testing method of the first aspect.

[0015] This disclosed product testing method, by collecting multimodal data of the device under test, can simultaneously identify appearance defects and geometric dimensional deviations, overcoming the limitations of single testing methods. It adaptively selects appropriate data processing strategies based on the product's material characteristics, specifically eliminating special noise and interference introduced by material properties, and completes comparison and judgment based on a digital standard model. No manual intervention is required, achieving full-process automation and significantly improving testing accuracy, efficiency, and result consistency. Attached Figure Description

[0016] Figure 1 The diagram shown is a structural schematic of a product testing device provided in an embodiment of this disclosure.

[0017] Figure 2 The diagram shown is a flowchart illustrating a product testing method provided in an embodiment of this disclosure.

[0018] Figure 3 The diagram shown is a flowchart illustrating the steps of comparing second multimodal data with a digital standard model of the device under test to obtain the detection result of the device under test, according to an embodiment of this disclosure.

[0019] Figure 4 The diagram shown is a flowchart illustrating the steps for acquiring the first multimodal data of a device under test according to an embodiment of this disclosure.

[0020] Figure 5 The diagram shown is a flowchart illustrating the steps of processing first multimodal data to obtain second multimodal data according to the processing method corresponding to the material of the device under test, as provided in an embodiment of this disclosure.

[0021] Figure 6 The diagram shown is a flowchart illustrating the steps of performing adaptive illumination compensation on first image data and enhancing the defect features of the first image data to obtain second image data, according to an embodiment of this disclosure.

[0022] Figure 7 The diagram shown is a flowchart illustrating the steps of comparing second image data with an appearance feature library to identify appearance defects according to an embodiment of this disclosure.

[0023] Figure 8 The diagram shown is a flowchart illustrating a product testing method provided in another embodiment of this disclosure. Detailed Implementation

[0024] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0025] Silicon carbide and graphite products are widely used in the semiconductor and new energy fields, such as silicon carbide substrates, silicon carbide structural components, wafers, graphite electrodes, and graphite boats. The appearance quality (e.g., surface scratches, cracks, chipped edges, missing corners, and chipped pieces) and dimensional accuracy (e.g., length, hole diameter, depth, flatness, and step height) of these products directly determine their lifespan and operational reliability. During the production stage, a comprehensive inspection of the product's appearance and dimensions is necessary.

[0026] Currently, the inspection of silicon carbide and graphite products mainly relies on manual inspection: inspectors visually examine the product's appearance for defects and use tools such as calipers and micrometers to measure key dimensions to identify dimensional defects. This method depends on the inspector's experience, is inefficient, and is prone to misjudgment due to fatigue. It cannot meet the industrial product inspection requirements for automated, high-precision, and high-efficiency appearance defect identification and dimensional inspection.

[0027] To achieve these goals, relevant technologies typically employ visual inspection or 3D scanning inspection. Visual inspection relies on industrial cameras to capture two-dimensional images of the product and uses image processing algorithms to identify surface defects such as scratches and cracks. However, visual inspection cannot obtain the product's three-dimensional dimensional information, resulting in missing dimensions such as depth and height.

[0028] 3D scanning inspection relies on a 3D scanner to collect point cloud data and compares the reconstructed 3D model with a standard model to detect dimensional defects. However, 3D scanning inspection lacks sensitivity in recognizing two-dimensional appearance features such as fine surface scratches and color differences.

[0029] Therefore, there is an urgent need for a fully automated inspection solution capable of simultaneously and accurately identifying product appearance defects and measuring 3D dimensions. This is especially true for materials with unique physical properties such as silicon carbide and graphite, whose inherent material characteristics introduce significant noise and interference into image or point cloud data, further exacerbating the risks of missed or misjudged appearance defects and inaccurate dimensional measurements. As the application of materials like silicon carbide and graphite in high-end manufacturing deepens, the demand for automated inspection technologies that combine comprehensiveness, high precision, and material adaptability is becoming increasingly urgent.

[0030] To address the above problems, this disclosure provides a product testing method and apparatus. The product testing method includes: acquiring first multimodal data of a device under test; processing the first multimodal data according to the processing method corresponding to the material of the device under test to obtain second multimodal data; and comparing the second multimodal data with a digital standard model of the device under test to obtain the test result of the device under test. When processing products made of special materials, personalized testing schemes are adopted for different materials, achieving high-precision and high-reliability automated testing of products made of different materials.

[0031] It is understood that the product testing methods and devices provided in this disclosure are not limited to silicon carbide or graphite products, but can also be applied to the appearance and size testing of other hard and brittle materials or highly reflective materials such as cemented carbide and glass, or other materials with loose and easily powdery characteristics such as carbon composite materials, talc, and mica. The processing method can be adjusted according to the characteristics of the target material, and will not be elaborated here.

[0032] Figure 1 The diagram shown is a structural schematic of a product testing device provided in an embodiment of this disclosure. Figure 1 As shown, the product inspection device includes a workbench 110, a data acquisition module 120, and a control module 130. The product inspection device can be deployed at the end of the product production line or at a separate inspection station for automated inspection of the appearance and dimensions of the devices to be inspected.

[0033] The worktable 110 is used to support and fix the device under test. The worktable 110 typically includes a highly flat table surface and a rotatable fixture platform. Optionally, the highly flat table surface may be made of marble, with a flatness of less than or equal to 0.002 mm / m, ensuring the stability of the device under test. The fixture platform has a rotation function and is equipped with adjustable soft positioning fixtures to accommodate devices of different sizes and shapes, ensuring the device's position is fixed during acquisition and avoiding measurement errors caused by movement. Optionally, the fixtures are made of soft rubber to avoid damaging the device surface.

[0034] The acquisition module 120 is used to acquire multimodal data of the device under test. Optionally, the acquisition module 120 integrates an image acquisition device and a 3D scanning device, responsible for simultaneously acquiring two-dimensional images and three-dimensional point cloud data of the device under test. The image acquisition device can be a high-definition industrial camera equipped with a ring light source, which uniformly illuminates the surface of the device through diffuse reflection illumination, reducing shadow interference. The 3D scanning device can be a structured light 3D scanner. The image acquisition device and the 3D scanning device can be fixed at the end of the same robotic arm, maintaining a certain distance (e.g., 100mm). The relative positions of the two acquisition devices are precisely calibrated to ensure that they acquire data from the same or similar perspectives, which is beneficial for the spatial coordinate alignment of the two-dimensional image and the three-dimensional point cloud data.

[0035] It is understandable that, in addition to the image acquisition devices and 3D scanning devices listed above, the acquisition module 120 can also be configured with other acquisition devices according to actual needs, and this disclosure does not impose specific restrictions on this.

[0036] The robotic arm can move the acquisition module 120 within the workbench space at multiple angles and positions to achieve comprehensive, blind-spot-free acquisition of the device under test. The robotic arm is typically a multi-axis industrial robotic arm, with a working range covering the entire workbench area. The end effector of the robotic arm is equipped with a quick-change interface for rapid replacement of the acquisition module 120, facilitating maintenance and upgrades.

[0037] The robotic arm receives the acquisition path from the control module 130 and moves along the acquisition path, driving the acquisition module 120 at the end to achieve multi-angle acquisition of the device under test. The acquisition path is a motion trajectory planned by the control module 130 based on the digital model of the product under test, covering all key viewing angles (such as front, back, left, right, top, and bottom) and several transitional viewing angles of the device under test. The robotic arm works in conjunction with the aforementioned rotatable fixture to achieve comprehensive acquisition of the device under test without blind spots.

[0038] The control module 130 is the core processing unit of the device, typically consisting of an industrial computer and corresponding software systems. The control module 130 connects to the workbench 110, the acquisition module 120, and the robotic arm via wired or wireless communication. It is responsible for coordinating the timing of the actions of each hardware component, such as planning the acquisition path of the robotic arm, controlling the rotation of the gripper, and triggering the acquisition module 120 to scan. Simultaneously, the acquired multimodal data is transmitted to the control module 130 in real time. The control module 130 executes the product testing method of this disclosure to test the device under test and generate test results. The entire process, from data acquisition to test report generation, is automatically completed under the coordination of the control module, requiring no manual intervention in any processing step.

[0039] After briefly introducing the exemplary product testing device of Embodiment 1 of this disclosure, the following is combined with... Figures 2 to 8 The product testing method provided in the embodiments of this disclosure is described in detail.

[0040] Figure 2 The diagram shown is a schematic flowchart of a product testing method provided in an embodiment of this disclosure. Figure 2 As shown in the embodiments of this disclosure, the product testing method includes the following steps.

[0041] S210, acquire the first multimodal data of the device under test.

[0042] The first multimodal data is a set of raw inspection data acquired from the device under test using at least one data acquisition device during product inspection, including at least two different physical properties or dimensional information. Optionally, the first multimodal data includes, but is not limited to: two-dimensional image data reflecting the optical properties of the surface of the device under test, point cloud data reflecting the three-dimensional spatial structure of the device under test, laser scattering data obtained by illuminating the surface of the device under test with a laser source at a specific angle, and pixel-by-pixel depth map data directly output by a binocular vision or time-of-flight camera.

[0043] The first multimodal data can simultaneously characterize the surface appearance information and three-dimensional geometric dimension information of the device under test. Optionally, the two-dimensional image data includes grayscale or color images of the device under test from multiple preset viewpoints. The two-dimensional image data can be used to determine surface appearance information such as texture, color, scratches, cracks, and missing corners. The point cloud data contains a three-dimensional point set representing the shape of the device under test, composed of a large number of spatial coordinate points. The point cloud data can be used to determine the three-dimensional geometric dimension information of the device under test, such as its outline, length, aperture, depth, and step height. Laser scattering data can provide surface appearance information such as microcracks, minor scratches, or roughness variations on the surface of the device under test. The depth map data includes a pixel-by-pixel depth value array of the device under test, and similar to the point cloud data, it can provide three-dimensional geometric dimension information of the device under test.

[0044] Understandably, the data from the above multiple modalities can be selectively combined based on actual detection accuracy requirements and cost considerations, and this disclosure does not impose specific restrictions on them as the first multimodal data.

[0045] S220, based on the processing method corresponding to the material of the device under test, the first multimodal data is processed to obtain the second multimodal data.

[0046] Different processing methods are used to process the first multimodal data for different material properties, so as to suppress or eliminate specific noise, distortion or interference introduced into the first multimodal data due to the inherent physical properties of the material of the device under test.

[0047] The processing methods for different materials are based on a set of optimized algorithms pre-set for the characteristics of each material. For example, silicon carbide, with its high reflectivity and hardness, will result in large overexposed areas in its images, masking minor scratches and cracks. Furthermore, its high hardness makes it prone to specular reflection, causing voids or shifts in the point cloud data. Graphite, on the other hand, is porous and prone to shedding powder, resulting in a large amount of loose powder particles mixed into its images. These particles are misidentified as scratches or cracks, and the natural porous texture of graphite is difficult to distinguish from real cracks in the 3D point cloud. If the first multimodal data of silicon carbide and graphite devices are processed using the same method, it is impossible to specifically eliminate the different noises and interferences introduced by different materials, significantly reducing the accuracy of subsequent detection processes.

[0048] To this end, this disclosure provides different sets of algorithms for different materials. When processing the first multimodal data, the material category of the device under test is first identified. Then, based on the material category, a corresponding processing strategy is matched from multiple preset algorithm sets, and the second multimodal data is generated based on this processing strategy. Through this material-adaptive processing method, the interference of different material characteristics on the detection process can be effectively eliminated, preserving and enhancing the true defect features, thereby providing accurate and reliable multimodal data for subsequent comparative analysis.

[0049] S230, compare the second multimodal data with the digital standard model of the device under test to obtain the test result of the device under test.

[0050] A digital standard model is a reference model or database stored in digital form and used to characterize product conformity standards.

[0051] As an optional implementation, the product testing method provided in this disclosure also includes: obtaining the design drawings of the device to be tested; extracting dimensional parameters and appearance requirement parameters from the design drawings; and constructing a digital standard model.

[0052] For example, the design drawings of the device under test can be in CAD format (such as DWG, STEP), PDF format, or any image format. The digital standard model includes a library of dimensional standards and a library of appearance features.

[0053] The control module can automatically identify dimension lines, values, and tolerance symbols (such as "Φ10±0.01") in drawings using optical character recognition (OCR) and graphic element parsing technologies, converting them into structured dimensioning parameters and storing them in the dimension standard library. Alternatively, it can convert design drawings into standard 3D models and store them in the dimension standard library.

[0054] At the same time, the text descriptions in the technical requirements column of the drawings (such as "no visible scratches on the surface" and "crack length ≤ 0.1mm") are identified, parsed and quantified into specific appearance requirement parameters (such as scratch grayscale threshold and crack length threshold), and stored in the appearance feature library.

[0055] In this embodiment, standard information is extracted directly from the product design source, eliminating the tedious steps of manually building models, avoiding human error, ensuring the consistency of detection criteria, and forming the basis of the automated detection process.

[0056] After determining the digital standard model, the surface appearance information and three-dimensional geometric dimension information of the device under test are determined based on the second multimodal data, and compared with the digital standard model to determine whether the device under test has appearance defects or dimensional defects.

[0057] Specifically, the surface appearance information of the device under test is compared with the appearance feature library in the digital standard model. The appearance feature library contains judgment thresholds for various defects (such as scratch length threshold, crack width threshold, and defect area threshold). If the geometric parameters of the detected defects in the surface appearance information exceed the corresponding judgment threshold, the device under test is determined to have an appearance defect; conversely, if no appearance defect is detected or the geometric parameters of all appearance defects do not exceed the threshold, the appearance is deemed acceptable.

[0058] The three-dimensional geometric dimensions of the device under test are compared with a dimensional standard library in a digital standard model. The dimensional standard library contains the labeled values ​​of each key dimension and their allowable tolerance ranges (such as length, aperture, depth, etc.). The deviation between the actual measured value and the labeled value of each dimension in the three-dimensional geometric dimension information is calculated. If the deviation exceeds the corresponding tolerance range, the device under test is determined to have a dimensional defect; otherwise, if no dimensional deviation is detected or all dimensional deviations are within the tolerance range, the dimensions are deemed acceptable.

[0059] Based on the above comparison of appearance and size, if the device under test has neither appearance defects nor size defects, it is ultimately determined to be a qualified product; otherwise, it is determined to be a non-qualified product, and the specific information of the non-qualified items can be further output for subsequent analysis or repair reference.

[0060] In this embodiment, by collecting multimodal data of the device under test, both appearance defects and geometric deviations can be identified simultaneously, overcoming the limitations of a single detection method. Appropriate data processing strategies are adaptively selected based on the product's material characteristics, specifically eliminating noise and interference introduced by these material properties. Comparison and judgment are completed based on a digital standard model, requiring no manual intervention and achieving full-process automation, significantly improving detection accuracy, efficiency, and result consistency.

[0061] The following section introduces a specific method for determining the test results.

[0062] Figure 3 The diagram shown is a flowchart illustrating the steps of comparing second multimodal data with a digital standard model of the device under test to obtain the detection result of the device under test, according to an embodiment of this disclosure.

[0063] like Figure 3 As shown in this embodiment, the step of comparing the second multimodal data with the digital standard model of the device under test to obtain the detection result of the device under test includes the following steps.

[0064] S310, compare the second image data with the appearance feature library to identify appearance defects.

[0065] In this embodiment of the disclosure, the multimodal data includes two-dimensional image data and point cloud data. Hereinafter, the two-dimensional image data and point cloud data in the first multimodal data will be referred to as first image data and first point cloud data, respectively; and the two-dimensional image data and point cloud data in the second multimodal data will be referred to as second image data and second point cloud data, respectively.

[0066] The appearance feature library pre-stores various defect type judgment thresholds, such as scratch length threshold, crack width threshold, and defect area threshold. The control module extracts potential defect areas from the second image data using image processing algorithms, calculates their geometric parameters (such as length, width, and area), and then compares these geometric parameters with the corresponding judgment thresholds in the appearance feature library.

[0067] For example, edge detection algorithms can be used to identify the outline of scratches or cracks, grayscale contrast algorithms can be used to identify defective areas such as missing corners or chipped edges, and morphological analysis can be used to calculate the size of defects, thereby locating and identifying appearance defects.

[0068] S320, based on the second point cloud data, establishes a three-dimensional model of the device under test, and compares the three-dimensional model with the size standard library to calculate the size deviation.

[0069] The dimensional standard library contains the annotation values ​​for each key dimension and their allowable tolerance range.

[0070] The discrete points in the second point cloud data are reconstructed into a continuous three-dimensional model through operations such as registration and meshing. By calculating the key dimensions (including length, width, depth, curvature, etc.) on the three-dimensional model and comparing them with the labeled values ​​in the dimension standard library, the actual dimension deviation of each dimension is obtained.

[0071] For example, the height or diameter at a specific location can be extracted using cross-sectional analysis, and the curvature can be calculated using least-squares fitting.

[0072] S330 determines the inspection results based on appearance defects and dimensional deviations.

[0073] If the geometric parameters of the defective area exceed the corresponding judgment threshold, the device under test is determined to have an appearance defect; otherwise, if no defective area is detected or the geometric parameters of all defective areas do not exceed the judgment threshold, the appearance is deemed acceptable.

[0074] If the dimensional deviation exceeds the corresponding allowable tolerance range, the device under test is determined to have a dimensional defect; conversely, if no dimensional deviation is detected or all dimensional deviations are within the allowable tolerance range, the dimensions are deemed acceptable.

[0075] Based on the above judgment results, a final inspection conclusion is generated according to the preset judgment rules. Specifically, if no appearance defects are detected (i.e., the geometric parameters of all defect areas do not exceed the judgment threshold) and no dimensional defects are detected (i.e., all dimensional deviations are within the allowable tolerance range), the device under inspection is judged to be a qualified product; otherwise, it is judged to be a non-qualified product.

[0076] If the item is deemed unqualified, further details about the unqualified item can be output, such as "Surface scratch length 0.15mm (threshold 0.1mm)" or "Aperture deviation +0.03mm (tolerance ±0.02mm)", for subsequent analysis or rework reference.

[0077] In this embodiment, based on processed high-quality images and point cloud data, defects exceeding the threshold are accurately identified and dimensional deviations are quantified, avoiding misjudgments based on a single dimension and achieving a comprehensive assessment of product conformity. Simultaneously, it can flexibly adapt to various defect detection algorithms and size comparison strategies, improving the adaptability and scalability of the judgment rules. The entire detection process is completed automatically without manual intervention, ensuring the consistency and reliability of the detection results.

[0078] The characteristics and corresponding processing methods of silicon carbide and graphite devices under test in the multimodal data acquisition, processing, and comparison process are described below. It is understood that silicon carbide and graphite are merely exemplary materials used in the embodiments of this disclosure and are not intended to limit the scope of protection of this disclosure. For other materials with similar high reflectivity and high hardness characteristics to silicon carbide, or other materials with similar porous and powdery characteristics to graphite, the above-described targeted optimization processing strategies can be referenced, and the preprocessing algorithm parameters or modules can be adjusted accordingly based on the specific material characteristics to achieve similar detection effects.

[0079] In some embodiments, the material of the device under test includes silicon carbide. The following describes a specific implementation of a multimodal data acquisition process for a silicon carbide device under test.

[0080] Figure 4 The diagram shown is a flowchart illustrating the steps for acquiring the first multimodal data of a device under test according to an embodiment of this disclosure.

[0081] like Figure 4 As shown in this embodiment, the step of acquiring the first multimodal data of the device under test includes the following steps.

[0082] S410, determine the material hardness of the device to be tested.

[0083] Material hardness can be expressed using Mohs hardness or Vickers hardness, which can be determined by checking product batch information or process databases, or by using a hardness tester for rapid measurement.

[0084] S420 utilizes a mapping model between silicon carbide hardness and scanning parameter set to determine the target scanning parameter set corresponding to the device under test based on the material hardness.

[0085] The mapping model between silicon carbide hardness and the set of scanning parameters is a mathematical model or lookup table, pre-calibrated experimentally, that reflects the correspondence between silicon carbide hardness and optimal scanning parameters. Higher silicon carbide hardness typically results in higher surface reflectivity. Therefore, during the acquisition of the first point cloud data, it is necessary to appropriately reduce the scanning power to avoid signal saturation and increase the scanning frequency to increase the sampling point density to compensate for reflection loss.

[0086] The target scanning parameter set includes at least one of the following: laser power, scanning frequency, and point cloud sampling density of the scanning device. For example, the material hardness of the device under test is converted into a corresponding laser power (e.g., 5mW-20mW range), scanning frequency (e.g., 100 Hz-500 Hz), and point cloud sampling density (e.g., 100 points / mm²-500 points / mm²) using a mapping model. For instance, if the material hardness of the device under test is 9.5H, the scanning parameter set is determined to be a laser power of 15 mW, a scanning frequency of 300 Hz, and a point cloud sampling density of 300 points / mm².

[0087] The S430 collects first point cloud data based on the target scanning parameter set.

[0088] The target scanning parameters are configured for the 3D scanning device used to acquire point cloud data. Then, data is acquired from the device to be inspected according to the preset scanning path to obtain the first point cloud data. During the acquisition process, the parameters can be fine-tuned in real time according to the point cloud quality. For example, the laser power can be temporarily reduced when encountering areas with excessive reflection.

[0089] S440 determines the target polarization angle based on the surface reflection angle of the device under test.

[0090] The surface reflection angle refers to the angle of direction of the reflected light relative to the surface normal when the light source is reflected from the surface of the device under test.

[0091] The high reflectivity of silicon carbide surfaces leads to large-area overexposure during image acquisition, easily causing defects to be confused with reflective areas and resulting in a high rate of missed detections. To avoid this problem, this embodiment of the present disclosure adds a rotatable polarizing filter to the front end of the image acquisition device and adjusts the polarization direction of the filter to be perpendicular to the polarization direction of the reflected light. This effectively filters out more than 90% of the specular reflection light while retaining the diffuse reflection light from the defect area.

[0092] Therefore, in this embodiment, the dominant polarization direction of the reflected light at each acquisition angle is pre-calculated based on the surface geometry of the device under test and the incident direction of the light source, or the optimal polarization angle is automatically searched through real-time feedback during image acquisition to obtain the target polarization angle. At the target polarization angle, the polarizing filter can suppress specular reflection light to the greatest extent.

[0093] S450 adjusts the polarization filter of the image acquisition device based on the target polarization angle and acquires the first image data.

[0094] Based on the target polarization angle, the angle of the polarizing filter is automatically adjusted by a motor to align its transmission axis with the target angle; then, the first image data is acquired under preset lighting conditions. During the acquisition process, the angle of the polarizing filter can be dynamically adjusted according to changes in the reflection angle at different viewing angles. For example, when the robotic arm moves the image acquisition device to each acquisition position, the angle of the polarizing filter is adjusted based on the target polarization angle corresponding to that acquisition position before the acquisition is performed.

[0095] In this embodiment, taking advantage of the high hardness and high reflectivity of silicon carbide material, the point cloud scanning parameters are adjusted adaptively by hardness to reduce point cloud holes from the source and ensure the integrity of the first point cloud data. At the same time, the polarization filter of the image acquisition device is dynamically adjusted based on the surface reflection angle to effectively suppress the interference of surface reflection on image acquisition, improve data quality during the acquisition stage, and enhance the robustness and accuracy of the detection process.

[0096] Next, we will introduce a data processing method applicable to silicon carbide-based devices under test.

[0097] Figure 5 The diagram shown is a flowchart illustrating the steps of processing first multimodal data to obtain second multimodal data according to the processing method corresponding to the material of the device under test, as provided in an embodiment of this disclosure.

[0098] like Figure 5 As shown in this embodiment, the step of processing the first multimodal data to obtain the second multimodal data according to the processing method corresponding to the material of the device under test includes the following steps.

[0099] S510, adaptive illumination compensation is performed on the first image data, and the defect features of the first image data are enhanced to obtain the second image data.

[0100] Adaptive illumination compensation refers to dynamically adjusting exposure or grayscale mapping based on the brightness distribution of local areas of an image in order to suppress overexposed areas caused by high reflectivity and enhance details in dark areas.

[0101] Due to the highly reflective nature of silicon carbide surfaces, overexposed areas with strong reflections inevitably exist in images. These overexposed areas mask subtle defects, leading to missed detections. By adaptively adjusting exposure or grayscale transformation parameters based on local image content, the exposure intensity of overexposed areas is reduced, while the exposure intensity of underexposed areas is increased, thereby improving the contrast between defects and the background.

[0102] Furthermore, defect features in the image can be enhanced to highlight the defect outline. Defect features refer to the visually salient attributes of surface anomalies such as scratches, cracks, and chipped edges in the image, such as edges and contrast.

[0103] Alternatively, the Laplacian operator or the Sobel operator can be used to perform convolution operations on the image to enhance the gradient value of the defect edges. Deep learning methods can also be used to achieve end-to-end image enhancement mapping. Frequency domain filtering methods can also be used to retain high-frequency defect information while suppressing low-frequency uneven illumination components through a high-pass filter.

[0104] S520 corrects the offset points in the first point cloud data and uses an interpolation algorithm to fill the point cloud holes caused by material reflection, thus obtaining the second point cloud data.

[0105] Offset points refer to anomalous points where the measured spatial coordinates deviate from their actual positions due to specular reflection of laser or structured light by the high hardness of silicon carbide surfaces. Point cloud voids refer to blank areas in the point cloud that are not recorded by the scanner due to missing reflected light signals. These defects are the result of the combined effect of the high hardness and high reflectivity of silicon carbide.

[0106] For offset points, edge feature points (such as chamfers and hole edges) on the surface of the device to be tested are extracted as reference points. Rigid transformation or non-rigid registration algorithms are used to correct the overall or local offset of the point cloud, so as to correct the offset points in real time and reduce the point cloud registration error.

[0107] To address the void problem, an interpolation method based on neighborhood point clouds is adopted. By utilizing the normal vector and curvature information of the effective point cloud around the void, a new point cloud that is continuous with the original surface is generated through interpolation to fill the void in the point cloud caused by reflection, ensuring the integrity of the point cloud and avoiding deviations during dimensional measurement.

[0108] In this embodiment, adaptive illumination compensation effectively suppresses interference from high material reflectivity and enhances the contrast of subtle defects. At the same time, it corrects point cloud offset and fills in holes, restoring the true shape of the device under test. It eliminates measurement errors caused by material characteristics at the data level, improves the quality of images and point clouds, and provides a reliable data foundation for subsequent appearance recognition and size comparison, thereby significantly improving the accuracy and robustness of silicon carbide device detection.

[0109] Figure 6 The diagram shown is a flowchart illustrating the steps of performing adaptive illumination compensation on first image data and enhancing the defect features of the first image data to obtain second image data, according to an embodiment of this disclosure.

[0110] like Figure 6 As shown in the embodiment of this disclosure, the step of performing adaptive illumination compensation on the first image data and enhancing the defect features of the first image data to obtain the second image data includes the following steps.

[0111] S610, the first image data is divided into multiple local regions.

[0112] The high reflectivity of silicon carbide surfaces often manifests as localized overexposure rather than uniform overexposure across the entire image. Applying uniform compensation parameters to the entire image will not yield good processing results.

[0113] Therefore, in this embodiment of the present disclosure, the first image data is divided into several non-overlapping image sub-blocks according to spatial location, which are regarded as multiple local regions.

[0114] Alternatively, a regular grid partitioning method can be used, such as dividing the image into 16×16 pixel rectangular sub-blocks; or an adaptive partitioning strategy can be used to dynamically adjust the size and shape of the blocks according to the image content (such as edge density or grayscale variance).

[0115] S620: Adjust the exposure parameters of multiple local regions and perform defect enhancement processing on the adjusted local regions to obtain the second image data.

[0116] The exposure parameters are adjusted individually for the brightness distribution of each local area, thereby increasing or decreasing the exposure intensity of that local area and achieving fine-grained illumination compensation.

[0117] Optionally, based on the grayscale distribution statistics (such as mean, variance, minimum, and maximum) of each local area, the optimal exposure adjustment amount for that area is dynamically calculated, so that the exposure intensity of overexposed areas is reduced and the exposure intensity of underexposed areas is increased, thereby improving the contrast between defects and background across the entire image.

[0118] Subsequently, edge enhancement processing is applied to each local area to further highlight the defect outline.

[0119] In this embodiment, by dividing the image into multiple local regions and adjusting the exposure parameters of each region, refined processing of highly reflective images is achieved: reducing the exposure intensity of overexposed areas and increasing the exposure intensity of underexposed areas to avoid loss of local details; combined with subsequent defect enhancement processing, the contrast between minor defects such as scratches and cracks and the background is significantly improved. This method flexibly adapts to the lighting differences in different regions, effectively improving the accuracy of appearance defect identification.

[0120] In some embodiments, the material of the device under test includes graphite. The following describes a data processing method applicable to a graphite-based device under test.

[0121] In some embodiments, the step of processing the first multimodal data to obtain the second multimodal data according to the processing method corresponding to the material of the device under test includes: performing image opening and closing operations on the first image data to filter out surface powder of the device under test and obtain the second image data.

[0122] Powder refers to particles that naturally fall off or adhere to the surface of graphite materials due to their loose texture. In images, they appear as isolated noise points with small area and high contrast, and are easily misjudged as scratches or defects.

[0123] To avoid this problem, this embodiment employs morphological filtering on the image to eliminate floating powder while preserving scratches. Specifically, opening and closing operations are performed on the first image data. The opening operation involves first performing erosion on the pixel region of the image and then dilation, used to remove isolated, small noise points or floating powder particles. The closing operation involves first performing dilation on the pixel region of the image and then erosion, used to bridge small breaks or gaps in real defects, maintaining their continuity.

[0124] The embodiments of this disclosure, through the above-described method, can effectively filter out false defects caused by floating powder and retain the continuity of real scratches, presenting them as complete continuous lines. The two operations are typically used in combination to achieve a balance between noise reduction and fidelity.

[0125] The following section introduces a comparison method applicable to a graphite-based device under test.

[0126] Figure 7 The diagram shown is a flowchart illustrating the steps of comparing second image data with an appearance feature library to identify appearance defects according to an embodiment of this disclosure.

[0127] like Figure 7 As shown, the step of comparing the second image data with the appearance feature library to identify appearance defects includes the following steps.

[0128] S710, perform feature extraction on the second image data to obtain the image features of the texture region in the second image data.

[0129] Image features of textured regions refer to parameters that can quantitatively describe the visual attributes of textured regions, such as the contrast, energy, entropy, and other statistical quantities of the gray-level co-occurrence matrix of textured regions.

[0130] Because the normal surface of graphite has specific texture characteristics (such as uniform graininess and a certain degree of roughness), while appearance defects such as cracks can disrupt this regularity and form abnormal local patterns, these areas are considered texture regions. By extracting the image features of texture regions, the difference between normal and texture regions can be transformed into a distance or similarity metric in the feature space.

[0131] S720 calculates the similarity between image features and normal texture features, and determines that the texture region includes appearance defects if the similarity is less than the similarity threshold.

[0132] The normal texture features of graphite materials are extracted in advance from the surface of qualified graphite products, representing the texture features under defect-free conditions. Normal texture features can be pre-stored in an appearance feature library.

[0133] Normal texture features exhibit clustered distribution in the feature space, while texture features in defective regions deviate from these cluster centers. By calculating the similarity between the feature vector of the texture region to be detected and the normal feature library, and comparing it with a similarity threshold, it is possible to determine whether an anomaly exists. If the similarity is less than the similarity threshold, it indicates that the texture to be detected differs significantly from normal textures, and is judged to have an appearance defect (such as cracks or scratches); otherwise, it is judged to be a normal texture or an acceptable loose structure.

[0134] The similarity threshold is dynamically adjusted based on the surface roughness of the device under test. Specifically, the higher the surface roughness, the greater the variation in normal texture itself, and the lower the similarity threshold is accordingly, so as to avoid misjudging normal rough textures as defects; conversely, the lower the roughness, the higher the similarity threshold is accordingly, to improve the sensitivity to subtle defects.

[0135] In this embodiment, by extracting image features from the second image data and comparing them with normal texture features, quantitative discrimination of appearance defects is achieved. Simultaneously, the similarity threshold is dynamically adjusted based on the surface roughness of the device under test, enabling the judgment rules to adapt to the texture fluctuations of the material itself. This effectively distinguishes normal loose textures from actual cracks, scratches, and other defects, significantly reducing the false judgment rate in graphite detection and improving the accuracy and robustness of appearance inspection.

[0136] In addition, surface defects can be detected based on the three-dimensional model of the device under test, which will be described in detail below.

[0137] Figure 8 The diagram shown is a flowchart illustrating a product testing method provided in another embodiment of this disclosure.

[0138] like Figure 8 As shown in the embodiments of this disclosure, the product testing method further includes the following steps.

[0139] S810 determines the surface texture features of the 3D model.

[0140] Graphite surfaces naturally exhibit porous textures, with normal surface textures typically falling within a certain range, while true cracks manifest as deeper grooves. Based on this, it is possible to distinguish between normal and abnormal areas on a graphite surface.

[0141] Surface texture features include surface roughness and surface texture depth. Optionally, height data of local areas can be extracted from the 3D model, a reference plane can be fitted using the least squares method, and then the arithmetic mean of the absolute values ​​of the height deviations of each point in the local area from the reference plane can be calculated as the roughness. The difference between the maximum positive deviation and the maximum negative deviation can be taken as the texture depth. By quantitatively calculating surface roughness and surface texture depth, numerical basis can be provided for distinguishing normal textures from cracks.

[0142] S820, based on surface texture features, identifies suspected crack regions in a 3D model that contain suspected cracks.

[0143] The normal ranges for surface roughness and surface texture depth can be determined by experimentally collecting and statistically analyzing normal porous texture data of different graphite surfaces. For example, the roughness of a normal surface should be less than 0.5 μm, and the texture depth of a normal surface should be less than 0.01 mm. If, in a 3D model, the surface roughness of a local area is less than 0.5 μm, and the surface texture depth is less than 0.01 mm, then this local area is determined not to contain any suspected cracks. However, if the surface roughness or surface texture depth of a local area exceeds the normal range, then this local area is initially marked as a suspected crack area containing suspected cracks.

[0144] S830 extracts the three-dimensional cross-sectional curve of the suspected crack region and determines the depth and length of the suspected crack based on the three-dimensional cross-sectional curve.

[0145] The initially screened suspected crack areas may contain local depth anomalies or local undulations in normal loose texture, while real cracks typically have a geometric shape extending in a certain direction. To distinguish real cracks from suspected crack areas, a three-dimensional cross-sectional curve of the suspected crack area can be extracted first. A three-dimensional cross-sectional curve is a two-dimensional contour line obtained by intersecting a plane with the three-dimensional model of the suspected crack area. It can intuitively reflect the height changes of the area, thus facilitating the accurate quantification of its depth and length values.

[0146] S840, if the depth value is greater than the first threshold and the length value is greater than the second threshold, it is determined that the suspected crack area includes appearance defects.

[0147] A true crack possesses both sufficient depth and a certain length, while loose textures or localized depth anomalies on the graphite surface are often shallower or shorter. Therefore, a first threshold can be set to measure the depth value, and a second threshold can be set to measure the length value. If the depth value of a suspected crack is greater than the first threshold and the length value is greater than the second threshold, then the suspected crack is determined to be a true crack, meaning that the suspected crack area includes visual defects.

[0148] In this embodiment, the crack identification method based on a 3D model initially screens suspected crack areas by extracting surface roughness and texture depth, then quantifies the depth and length values ​​of the suspected cracks, and finally makes a final judgment based on a threshold, so as to effectively distinguish between normal loose textures and real cracks in graphite materials. Simultaneously, cross-validation of the image and the 3D model improves the accuracy and reliability of the detection.

[0149] The above text combined Figures 1 to 8 The present disclosure describes in detail embodiments of the methods and apparatus. In addition, embodiments of the present disclosure can also be computer program products, comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the product detection methods according to various embodiments of the present disclosure described above.

[0150] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0151] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the product testing methods according to various embodiments of this disclosure described above.

[0152] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0153] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the specific details described above.

[0154] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0155] It should also be noted that in the apparatus, devices, and methods of this disclosure, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions to this disclosure.

[0156] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0157] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A product testing method, characterized in that, include: Acquire the first multimodal data of the device under test; Based on the processing method corresponding to the material of the device under test, the first multimodal data is processed to obtain the second multimodal data; The second multimodal data is compared with the digital standard model of the device under test to obtain the detection result of the device under test.

2. The method according to claim 1, characterized in that, The material of the device under test includes silicon carbide, the first multimodal data includes first image data and first point cloud data, and the second multimodal data includes second image data and second point cloud data; The step of processing the first multimodal data according to the processing method corresponding to the material of the device under test to obtain the second multimodal data includes: Adaptive illumination compensation is applied to the first image data, and the defect features of the first image data are enhanced to obtain the second image data; The offset points in the first point cloud data are corrected, and the point cloud holes caused by material reflection are filled by an interpolation algorithm to obtain the second point cloud data.

3. The method according to claim 2, characterized in that, The step of performing adaptive illumination compensation on the first image data and enhancing the defect features of the first image data to obtain the second image data includes: The first image data is divided into multiple local regions; The exposure parameters of each of the multiple local regions are adjusted, and defect enhancement processing is performed on the adjusted multiple local regions to obtain the second image data.

4. The method according to claim 1, characterized in that, The material of the device under test includes graphite, the first multimodal data includes first image data, and the second multimodal data includes second image data; The step of processing the first multimodal data according to the processing method corresponding to the material of the device under test to obtain the second multimodal data includes: The first image data is processed by opening and closing operations to filter out surface powder from the device under test, thereby obtaining the second image data.

5. The method according to claim 1, characterized in that, The digital standard model includes an appearance feature library and a size standard library, and the second multimodal data includes second image data and second point cloud data; The step of comparing the second multimodal data with the digital standard model of the device under test to obtain the detection result of the device under test includes: The second image data is compared with the appearance feature library to identify appearance defects; Based on the second point cloud data, a three-dimensional model of the device to be tested is established, and the three-dimensional model is compared with the size standard library to calculate the size deviation; The test results are determined based on the appearance defects and the dimensional deviations.

6. The method according to claim 5, characterized in that, The material of the device to be tested includes graphite, and the appearance feature library includes normal texture features of graphite material. The step of comparing the second image data with the appearance feature database to identify appearance defects includes: Perform feature extraction on the second image data to obtain the image features of the texture region in the second image data; Calculate the similarity between the image features and the normal texture features, and if the similarity is less than a similarity threshold, determine that the texture region includes the appearance defect; The similarity threshold is dynamically adjusted based on the surface roughness of the device to be detected.

7. The method according to claim 5, characterized in that, The material of the device to be tested includes graphite, and the method further includes: Determine the surface texture features of the three-dimensional model, wherein the surface texture features include surface roughness and surface texture depth; Based on the surface texture features, the suspected crack regions containing suspected cracks in the three-dimensional model are determined. Extract the three-dimensional cross-sectional curve of the suspected crack region, and determine the depth and length of the suspected crack based on the three-dimensional cross-sectional curve; If the depth value is greater than a first threshold and the length value is greater than a second threshold, the suspected crack region is determined to include the appearance defect.

8. The method according to claim 1, characterized in that, The material of the device under test includes silicon carbide, and the first multimodal data includes first image data and / or first point cloud data; The acquisition of the first multimodal data of the device under test includes: Determine the material hardness of the device under test; Using a mapping model between silicon carbide hardness and scanning parameter set, the target scanning parameter set corresponding to the device under test is determined based on the material hardness. The target scanning parameter set includes at least one of the following: laser power of the scanning device, scanning frequency, and point cloud sampling density. Based on the target scanning parameter set, the first point cloud data is collected; and / or The target polarization angle is determined based on the surface reflection angle of the device under test; Adjust the polarization filter of the image acquisition device based on the target polarization angle, and acquire the first image data.

9. The method according to claim 1, characterized in that, Also includes: Obtain the design drawings of the device under test; The dimensional annotation parameters and appearance requirement parameters are extracted from the design drawings to construct the digital standard model.

10. A product testing device, characterized in that, include: The worktable is used to support and fix the device to be tested; The acquisition module is used to acquire multimodal data of the device under test; A control module for executing the product testing method according to any one of claims 1 to 9.

Images