A machining quality monitoring system and method for complex structure precision parts

Abstract

The application provides a processing quality monitoring system and method for complex structure precision parts. A fusion feature map is generated through a multispectral image sequence of a target precision part after processing. The spatial distribution characteristics of suspected processing defect areas are identified by combining the texture characteristics and spectral reflection characteristics distribution in the fusion feature map with a preset processing defect knowledge graph. Based on the spatial distribution characteristics, a three-dimensional line scanning module is controlled to perform three-dimensional point cloud reconstruction on the suspected processing defect areas, and three-dimensional topographic data is generated. According to the three-dimensional topographic data and the reference CAD model of the target precision part, the topographic deviation field of the suspected processing defect areas is extracted. Based on the local surface instability characteristics in the topographic deviation field and the multispectral feature vectors in the fusion feature map, the processing quality of the target precision part is classified. The technical scheme provided by the application can solve the problem of monitoring efficiency decline caused by full-field scanning point cloud data in precision part quality monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of processing quality monitoring, and more particularly to a processing quality monitoring system and method for complex structure precision parts. BACKGROUND

[0002] The current processing manufacturing industry is upgrading towards high precision and intelligentization, and processing quality monitoring has become a core link to ensure product performance and production efficiency. Traditional manual sampling detection has problems such as low efficiency, limited coverage, and being easily affected by subjective factors, and it is difficult to meet the real-time and accuracy requirements of modern production lines. With the popularization of Industry 4.0 technology, machine vision, sensors, big data analysis and other technologies are gradually integrated into the monitoring link, providing stable and reliable quality assurance for high-precision and mass production.

[0003] In the existing processing quality monitoring, the processing quality monitoring is mainly based on the core principles of data acquisition, analysis and judgment, and result feedback. Firstly, physical signal acquisition is performed to capture key data in the processing process in real time using sensors. Then, the collected key data is processed and analyzed to identify processing abnormalities. Finally, the processing abnormality results are fed back to realize processing quality monitoring. However, in the processing quality monitoring of complex structure precision parts, although three-dimensional full-field scanning can obtain complete topographic information of precision parts, when facing precision parts with complex morphological structure (i.e. having complex curved surfaces and micro features) and extremely high precision requirements, hundreds of millions or even tens of billions of point cloud data will be generated, which will cause the subsequent point cloud registration and quality defect identification to be unable to meet the efficiency requirements of quality monitoring. Therefore, how to solve the problem of monitoring efficiency decline caused by full-field scanning point cloud data in precision part quality monitoring has become a difficult problem in the industry. SUMMARY

[0004] The present application provides a processing quality monitoring system and method for complex structure precision parts, which can solve the problem of monitoring efficiency decline caused by full-field scanning point cloud data in precision part quality monitoring.

[0005] In a first aspect, the present application provides a processing quality monitoring method for complex structure precision parts, comprising the following steps:

[0006] Obtaining a multi-spectral image sequence of a target precision part after processing, and performing feature fusion on the multi-spectral image sequence to generate a fusion feature map representing the surface features of the target precision part;

[0007] Extracting texture features and spectral reflectance characteristics distribution in the fusion feature map, and identifying the spatial distribution features of suspected processing defect areas by combining the texture features and the spectral reflectance characteristics distribution with a preset processing defect knowledge graph;

[0008] Based on the spatial distribution characteristics of the suspected machining defect area, a three-dimensional line scanning module is controlled to perform local three-dimensional point cloud reconstruction on the suspected machining defect area, and sub-pixel level three-dimensional topography data is generated;

[0009] The three-dimensional topography data is non-rigidly registered with a reference CAD model of the target precision part, and a three-dimensional spatial deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model is determined, and a topography deviation field of the suspected machining defect area is generated;

[0010] Based on the local surface instability characteristics of the suspected machining defect area in the topography deviation field and the multi-spectral feature vectors in the fusion feature map, the target precision part is classified in terms of machining quality.

[0011] In some embodiments, the feature fusion of the multi-spectral image sequence generates a fusion feature map representing the surface features of the target precision part, specifically including:

[0012] The spatial registration is performed on each waveband monitoring image in the multi-spectral image sequence to obtain a registered image sequence;

[0013] The principal component analysis algorithm is used to perform feature dimension reduction on each registered image in the registered image sequence to obtain a dimension-reduced feature matrix corresponding to each registered image;

[0014] Based on all the dimension-reduced feature matrices, a fusion feature map representing the surface features of the target precision part is generated.

[0015] In some embodiments, the texture feature and the spectral reflectance characteristic distribution in the fusion feature map are extracted, specifically including:

[0016] The multi-directional filtering processing is performed on the fusion feature map to generate a multi-directional filtering response map;

[0017] The texture feature of the fusion feature map is extracted based on the multi-directional filtering response map;

[0018] The spectral channel separation is performed on the fusion feature map to obtain a gray value matrix of each spectral channel;

[0019] The spectral reflectance vector of the corresponding pixel point is calculated according to the gray value matrix of each spectral channel, and the spectral reflectance characteristic distribution of the fusion feature map is constructed according to all the spectral reflectance vectors.

[0020] In some embodiments, the spatial distribution characteristics of the suspected machining defect area are identified by combining the texture feature and the spectral reflectance characteristic distribution with a pre-set machining defect knowledge graph, specifically including:

[0021] perform feature normalization on the texture features and the spectral reflectance characteristic distribution to generate a normalized feature vector;

[0022] perform similarity matching on the normalized feature vector and defect feature vectors of various types of machining defects in a preset machining defect knowledge graph to obtain a feature similarity set;

[0023] extract a suspected machining defect region of the target precision part according to the feature similarity set;

[0024] perform morphological filtering and connected component analysis on the suspected machining defect region to determine spatial distribution features of the suspected machining defect region.

[0025] In some embodiments, based on the spatial distribution features of the suspected machining defect region, a three-dimensional line scanning module is controlled to perform local three-dimensional point cloud reconstruction on the suspected machining defect region to generate sub-pixel level three-dimensional topography data, specifically including:

[0026] According to the spatial distribution features of the suspected machining defect region, the scanning path and parameter configuration of the three-dimensional line scanning module are calculated;

[0027] Based on the scanning path and the parameter configuration, the three-dimensional line scanning module is controlled to perform layered scanning on the suspected machining defect region to obtain a sub-pixel level depth image;

[0028] The depth image is subjected to distortion correction and coordinate conversion to generate initial three-dimensional point cloud data;

[0029] The initial three-dimensional point cloud data is subjected to denoising processing to generate sub-pixel level three-dimensional topography data.

[0030] In some embodiments, the three-dimensional topography data is non-rigidly registered with a reference CAD model of the target precision part, specifically including:

[0031] An initial corresponding point pair is constructed according to the three-dimensional topography data and the reference CAD model;

[0032] An initial registration transformation matrix is obtained by performing rigid registration on the initial corresponding point pair using an iterative closest point algorithm;

[0033] A non-rigid transformation model is constructed according to the initial registration transformation matrix;

[0034] Based on the non-rigid transformation model, the three-dimensional topography data is iteratively optimized with the goal of minimizing point cloud distance error to complete non-rigid registration of the three-dimensional topography data and the reference CAD model.

[0035] In some embodiments, a multi-spectral camera is used to acquire a multi-spectral image sequence of the target precision part after machining.

[0036] In a second aspect, the application provides a processing quality monitoring system for complex structure precision parts, which is used to execute the processing quality monitoring method for complex structure precision parts. The system comprises:

[0037] An acquisition module is configured to acquire a multi-spectrum image sequence of a target precision part after processing, and perform feature fusion on the multi-spectrum image sequence to generate a fusion feature map representing surface features of the target precision part.

[0038] A processing module is configured to extract texture features and spectral reflection characteristic distributions in the fusion feature map, and identify spatial distribution features of a suspected processing defect area by combining the texture features and the spectral reflection characteristic distributions with a preset processing defect knowledge graph.

[0039] The processing module is further configured to control a three-dimensional line scanning module to perform local three-dimensional point cloud reconstruction on the suspected processing defect area based on the spatial distribution features of the suspected processing defect area, and generate sub-pixel level three-dimensional topographic data.

[0040] The processing module is further configured to perform non-rigid registration on the three-dimensional topographic data and a reference CAD model of the target precision part, determine a three-dimensional space deviation vector of each point cloud data point in the three-dimensional topographic data relative to the reference CAD model, and further generate a topographic deviation field of the suspected processing defect area.

[0041] An execution module is configured to perform processing quality classification on the target precision part based on local surface instability features of the suspected processing defect area in the topographic deviation field and multi-spectrum feature vectors in the fusion feature map.

[0042] In a third aspect, the application provides a computer device, which comprises a memory and a processor. The memory stores a code, and the processor is configured to acquire the code and execute the processing quality monitoring method for complex structure precision parts.

[0043] In a fourth aspect, the application provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processing quality monitoring method for complex structure precision parts is realized.

[0044] The technical scheme provided by the embodiments of the application has the following beneficial effects:

[0045] The processing quality monitoring system and method for complex structure precision parts provided by the application first acquire a multi-spectral image sequence of a target precision part after processing, and perform feature fusion on the multi-spectral image sequence to generate a fusion feature map representing the surface features of the target precision part. Secondly, the texture features and spectral reflection characteristic distribution in the fusion feature map are extracted, and the spatial distribution features of the suspected processing defect area are identified by combining the texture features, the spectral reflection characteristic distribution and a preset processing defect knowledge graph. Further, based on the spatial distribution features of the suspected processing defect area, a three-dimensional line scanning module is controlled to perform local three-dimensional point cloud reconstruction on the suspected processing defect area to generate sub-pixel level three-dimensional topography data. Then, the three-dimensional topography data is non-rigidly registered with a reference CAD model of the target precision part, and a three-dimensional space deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model is determined, and then a topography deviation field of the suspected processing defect area is generated. Finally, the target precision part is classified in processing quality based on the local surface instability features of the suspected processing defect area in the topography deviation field and the multi-spectral feature vectors in the fusion feature map.

[0046] Therefore, the application can solve the problem of monitoring efficiency decline caused by full-field scanning point cloud data in precision part quality monitoring. Firstly, a multi-spectral image sequence of a target precision part after processing is acquired and feature fusion is performed to generate a fusion feature map, which can effectively integrate surface information of different spectral bands and provide a more comprehensive surface feature basis for subsequent defect identification. Secondly, the texture features and spectral reflection characteristic distribution in the fusion feature map are extracted and combined with the processing defect knowledge graph to identify the spatial distribution features of the suspected processing defect area, which can accurately locate the area where defects may exist from the two dimensions of texture structure and material reflection characteristics, reducing missed detection and false detection. Further, based on the spatial distribution features of the suspected area, a three-dimensional line scanning module is controlled to perform local three-dimensional point cloud reconstruction to generate sub-pixel level three-dimensional topography data, which can reduce redundant scanning while ensuring high precision, improve three-dimensional reconstruction efficiency, and avoid the problem that three-dimensional full-field scanning causes point cloud registration and quality defect identification to fail to meet the efficiency requirements of quality monitoring. Then, the three-dimensional topography data is non-rigidly registered with the reference CAD model to generate a topography deviation field, which can quantitatively represent the geometric deviation between the suspected area and the design model and provide accurate three-dimensional quantitative basis for defect evaluation. Finally, the processing quality is classified based on the local surface instability features of the topography deviation field and the multi-spectral feature vectors of the fusion feature map, which can comprehensively implement accurate grading of the processing part quality from the geometric morphology and material characteristics, ensuring the comprehensiveness and accuracy of quality evaluation. In summary, the technical solution provided by the application can solve the problem of monitoring efficiency decline caused by full-field scanning point cloud data in precision part quality monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1is an exemplary flowchart of a processing quality monitoring method for a complex-structure precision part according to some embodiments of the present application;

[0048] Figure 2 is an exemplary flowchart of determining a spatial distribution feature according to some embodiments of the present application;

[0049] Figure 3 is a structural schematic diagram of a processing quality monitoring system for a complex-structure precision part according to some embodiments of the present application;

[0050] Figure 4 is a structural schematic diagram of a computer device for implementing a processing quality monitoring method for a complex-structure precision part according to some embodiments of the present application. DETAILED DESCRIPTION

[0051] In order to better understand the technical solutions of the present application, the technical solutions of the present application will be described in detail below in combination with the drawings of the specification and specific embodiments.

[0052] Reference Figure 1 The figure is an exemplary flowchart of a processing quality monitoring method for a complex-structure precision part according to some embodiments of the present application, which mainly includes the following steps:

[0053] In step S101, a multi-spectral image sequence after processing of a target precision part is obtained, and feature fusion is performed on the multi-spectral image sequence to generate a fusion feature map representing the surface features of the target precision part.

[0054] In specific implementation, the multi-spectral image sequence after processing of the target precision part can be obtained by a multi-spectral camera, and in other embodiments, the multi-spectral image sequence after processing of the target precision part can also be obtained by other spectral sensors, which is not limited here. The multi-spectral image sequence is obtained by combining a plurality of waveband monitoring images, and the waveband monitoring image refers to a visualized image that presents precision part processing quality related monitoring information under different spectral wavebands, which is the core basis for subsequent quality analysis and judgment.

[0055] In some embodiments, the feature fusion of the multi-spectral image sequence to generate the fusion feature map representing the surface features of the target precision part is implemented by the following steps:

[0056] The spatial registration is performed on each waveband monitoring image in the multi-spectral image sequence to obtain a registered image sequence;

[0057] The principal component analysis algorithm is used to perform feature dimension reduction on each registered image in the registered image sequence to obtain a dimension-reduced feature matrix corresponding to each registered image;

[0058] The fusion feature map representing the surface features of the target precision part is generated based on all the dimension-reduced feature matrices.

[0059] In a specific implementation, first, a band image with the highest signal-to-noise ratio in the multispectral image sequence is selected as a reference image, a 8*8 pixel template window is set for each of the remaining band monitoring images, the cross-correlation coefficients of the remaining band monitoring images and the reference image in the template window region are calculated, the spatial offset corresponding to the maximum cross-correlation coefficient is determined, the band monitoring images are translated and corrected according to the spatial offset (i.e., the spatial position of each pixel in the band monitoring image is added to the spatial offset), and then all the registration images are obtained. The registration images are combined in the original order to obtain a registration image sequence. Here, the registration image refers to the band monitoring image after registration, and the spatial registration refers to a process of correcting the spatial position deviation between different images so that the pixels corresponding to the same target region in the images are at the same coordinate position. Second, the covariance matrix of each registration image in the registration image sequence is calculated, the eigenvalues and the corresponding eigenvectors are obtained by performing eigenvalue decomposition on the covariance matrix through the Jacobi iteration method, the first three eigenvectors (i.e., more than 95% of the original data information is retained) are selected as the principal component directions after the eigenvalues are sorted in descending order, the two-dimensional data matrix corresponding to the registration image is multiplied by the three selected eigenvectors, and then the dimension-reduced feature matrix corresponding to each registration image is obtained. The two-dimensional data matrix refers to the pixel matrix of the registration image expanded by rows, and the dimension-reduced feature matrix refers to the registration image after dimension reduction. Finally, the maximum pixel value and the minimum pixel value at the same spatial position are extracted from the dimension-reduced feature matrices corresponding to all the registration images, the original pixel value is mapped to the gray interval of 0-255 through the maximum pixel value and the minimum pixel value combined with a linear stretching formula, and the pixel value at the original spatial position is replaced to obtain the fusion feature map representing the surface features of the target precision part. The linear stretching formula is: normalized gray value = (original element value-minimum pixel value) / (maximum pixel value-minimum pixel value)*255.

[0060] It should be noted that the fusion feature map in the present application refers to a single image obtained by fusing the effective information of each band of the multispectral image. Through the determination of the fusion feature map, the problem of high analysis complexity caused by the dispersion of multi-band information is solved, and the key features of the part surface are made clearer and more identifiable through feature enhancement, thereby providing a unified, efficient and accurate visual and data basis for subsequent core analysis links such as precision part surface defect recognition, texture feature extraction and processing quality evaluation.

[0061] In step S102, the texture features and the spectral reflection characteristic distribution in the fusion feature map are extracted, and the spatial distribution features of the suspected processing defect region are identified by combining the texture features and the spectral reflection characteristic distribution with a preset processing defect knowledge graph.

[0062] In some embodiments, the texture feature and the spectral reflectance distribution in the fusion feature map are extracted by the following steps:

[0063] The fusion feature map is subjected to multi-directional filtering processing to generate a multi-directional filtering response map;

[0064] The texture feature of the fusion feature map is extracted based on the multi-directional filtering response map;

[0065] The fusion feature map is subjected to spectral channel separation to obtain a gray value matrix of each spectral channel;

[0066] The spectral reflectance vector of each pixel point is calculated according to the gray value matrix of each spectral channel, and the spectral reflectance distribution of the fusion feature map is constructed according to all the spectral reflectance vectors.

[0067] In a specific implementation, first, when the fusion feature map is subjected to multi-directional filtering processing, a Gabor filter set (containing filters in 0°, 45°, 90°, and 135° directions) is selected, and each direction of the Gabor filter is subjected to convolution operation with the fusion feature map to obtain a multi-directional filtering response map, which refers to a map filtered by filters in different directions; second, when the texture feature is extracted based on the multi-directional filtering response map, the gray level co-occurrence matrix method is used to calculate the angular second moment, contrast, and entropy statistics in different directions and distances, and these statistics are combined as the texture feature of the fusion feature map, which refers to a feature quantity used to represent the distribution rule and combination mode of the pixel gray value in space; third, when the fusion feature map is subjected to spectral channel separation, the pixel gray values corresponding to different spectral bands in the fusion feature map are separated according to the original spectral band division of the multi-spectral imaging system to form a two-dimensional matrix (i.e., the gray value matrix of each spectral channel) corresponding to each original spectral band, which refers to a two-dimensional data matrix containing all pixel gray values in a single spectral band; and finally, when the spectral reflectance vector of each pixel point is calculated according to the gray value matrix of each spectral channel, the reflectance conversion coefficients of each spectral channel are obtained through existing standard whiteboard calibration, and the gray value of each pixel point in each spectral channel is multiplied by the conversion coefficient of the corresponding spectral channel to obtain the reflectance value of each pixel point in each band, and the vector formed by combining these reflectance values is obtained to obtain the spectral reflectance vector of each pixel point, and the spectral reflectance vectors of all pixel points are arranged according to the corresponding spatial positions to form the spectral reflectance distribution of the fusion feature map.

[0068] It should be noted that the spectral reflectance distribution in this application refers to a dataset describing the reflectance variation of each pixel in the fused feature map under different spectral bands. The spectral reflectance distribution can reflect the differences in material composition in different areas of the part surface (such as the different spectral responses of different metal platings, oxide layers, or contaminants), and can be used to identify whether there are problems such as uneven material, local oxidation, or oil residue on the part surface. On the other hand, texture features focus on the microstructure of the part surface (such as the direction, density, and uniformity of the processing lines, as well as texture anomalies caused by defects such as micro-scratches and dents), and can accurately capture information such as whether the surface processing quality meets the standards and whether there is physical damage. The combination of the two avoids the limitations of relying solely on spectral data to judge structural defects or relying solely on texture features to distinguish material problems, and can form a comprehensive characterization of the part surface state, providing more reliable decision support for subsequent practical applications such as part surface quality grading, defect type determination, and processing technology optimization.

[0069] In some embodiments, reference Figure 2 As shown in the figure, this is an exemplary flowchart of determining spatial distribution features according to some embodiments of this application. In this embodiment, the spatial distribution features of suspected processing defect areas can be identified by combining the texture features and the spectral reflectance distribution with a preset processing defect knowledge graph through the following steps:

[0070] In step S1021, the texture features and the spectral reflectance distribution are normalized to generate a normalized feature vector.

[0071] In step S1022, the normalized feature vector is matched with the defect feature vectors of various processing defect types in the preset processing defect knowledge graph to obtain a feature similarity set;

[0072] In step S1023, suspected machining defect areas of the target precision part are extracted based on the feature similarity set;

[0073] In step S1024, morphological filtering and connected component analysis are performed on the suspected processing defect region to determine the spatial distribution characteristics of the suspected processing defect region.

[0074] In a specific implementation, first, when performing feature normalization on the texture feature and the spectral reflectance characteristic distribution, a min-max normalization method is used to respectively map each statistic (such as an angular second moment, a contrast, and an entropy) in the texture feature and each band reflectance value in the spectral reflectance characteristic distribution to the interval [0, 1] to obtain dimensionless normalized features, and all the normalized features are combined to form a normalized feature vector, which is a vector composed of the normalized texture feature and the spectral reflectance characteristic distribution; second, a Euclidean distance algorithm is used to calculate the feature similarity between the normalized feature vector and the defect feature vectors of various types of machining defects in a preset machining defect knowledge graph, to obtain a feature similarity set, the various types of defects include but are not limited to scratches, depressions, and uneven materials, and the defect feature vectors of the various types of defects in the machining defect knowledge graph are specifically set according to historical normalized feature vectors simulated by machine learning, which is not limited here, the feature similarity set contains feature similarities corresponding to different machining defect types, and the feature similarity refers to the similarity between the defect feature detected by the current precision part and each type of defect feature; then, a suspected machining defect area of the target precision part is extracted according to the feature similarity set, that is, the machining defect type corresponding to the maximum feature similarity in the feature similarity set is extracted, an identification pixel interval corresponding to the machining defect type is called, the pixel value of each pixel point in the foregoing fusion feature map is compared with the identification pixel interval, the pixel points falling into the identification pixel interval are taken as suspected machining defect points, and then the area formed by all the suspected machining defect points is taken as the suspected machining defect area of the target precision part, the identification pixel interval is specifically set according to the corresponding machining defect type by expert knowledge or simulation experiment, which is not limited, and the candidate machining defect area refers to a pixel area set in which it is preliminarily determined that machining defects may exist; finally, when performing morphological filtering and connected domain analysis on the candidate machining defect area, morphological opening operation is first used to remove noise points in the candidate machining defect area, and then an 8-neighbor connected domain labeling algorithm is used to identify the connected area corresponding to the candidate machining defect area, and the spatial distribution feature of the suspected machining defect area is formed by the center position and the shape factor (which can be represented by the area of the connected area) of the connected area.

[0075] It should be noted that the spatial distribution characteristics of the suspected machining defect area in the present application refer to the distribution information of the suspected machining defect area in the image space. By determining the spatial distribution characteristics of the suspected machining defect area, the specific area where the defect may exist on the surface of the precision part can be accurately locked, thereby avoiding the waste of efficiency caused by subsequent detection of the whole part, and providing a clear target range for subsequent local three-dimensional point cloud reconstruction. The whole surface of the part does not need to be scanned redundantly, and only the suspected area pointed by the spatial distribution characteristics needs to be focused, thereby greatly improving the efficiency and accuracy of three-dimensional reconstruction, and ensuring the coherence and efficiency of the whole machining quality monitoring process from feature recognition to accurate positioning to depth analysis.

[0076] In step S103, based on the spatial distribution characteristics of the suspected machining defect area, a three-dimensional line scanning module is controlled to perform local three-dimensional point cloud reconstruction on the suspected machining defect area, and sub-pixel level three-dimensional topographic data is generated.

[0077] In some embodiments, based on the spatial distribution characteristics of the suspected machining defect area, a three-dimensional line scanning module is controlled to perform local three-dimensional point cloud reconstruction on the suspected machining defect area, and sub-pixel level three-dimensional topographic data is generated. The specific implementation is as follows:

[0078] According to the spatial distribution characteristics of the suspected machining defect area, the scanning path and parameter configuration of the three-dimensional line scanning module are calculated;

[0079] Based on the scanning path and the parameter configuration, the three-dimensional line scanning module is controlled to perform layered scanning on the suspected machining defect area, and sub-pixel level depth images are obtained;

[0080] The depth images are subjected to distortion correction and coordinate conversion to generate initial three-dimensional point cloud data;

[0081] The initial three-dimensional point cloud data is subjected to denoising processing to generate sub-pixel level three-dimensional topographic data.

[0082] In a specific implementation, first, the spatial distribution characteristics of the suspected machining defect area are calculated to obtain the scanning path and parameter configuration of the three-dimensional line scanning module. The center position in the spatial distribution characteristics is extracted as the scanning center of the three-dimensional line scanning module, the machining defect point farthest from the scanning center in the suspected machining defect area is obtained, and the distance between the scanning center and the machining defect point is taken as the scanning radius of the three-dimensional line scanning module. Then, the scanning path of the three-dimensional line scanning module is obtained. According to the sub-pixel level accuracy requirement, the parameter configuration of the three-dimensional line scanning module is set. The parameter configuration includes the scanning resolution (i.e. the distance between adjacent scanning lines), the scanning speed (i.e. the acquisition frequency), and the laser power (i.e. the laser power for adapting to the reflection characteristics of the surface material of the machining part). The scanning path and parameter configuration of the three-dimensional line scanning module are formed. The scanning path refers to the path planning for guiding the three-dimensional line scanning module to complete the scanning of the target area. The parameter configuration refers to the scanning parameter set for guiding the three-dimensional line scanning module to complete the scanning of the target area. Second, based on the scanning path and parameter configuration, the three-dimensional line scanning module is controlled to perform layered scanning on the suspected machining defect area to obtain a sub-pixel level depth image. The depth image refers to an image obtained by sub-pixel level identification of the suspected machining defect area. Then, the depth image is subjected to distortion correction and coordinate conversion to generate initial three-dimensional point cloud data. First, the Zhang calibration method (by pre-shooting a calibration board image to calculate the camera intrinsic parameter) is used to correct the distortion of the depth image to eliminate the depth error caused by the camera lens distortion. Then, the two-dimensional coordinates (u, v) and the depth value z of each pixel point in the corrected depth image are combined according to the external parameter of the three-dimensional line scanning module (the positional relationship between the module and the world coordinate system) to be converted into three-dimensional coordinates (X, Y, Z) in the world coordinate system to obtain the initial three-dimensional point cloud data. The initial three-dimensional point cloud data refers to an initial point cloud set containing the three-dimensional coordinate information of the surface of the suspected machining defect area after distortion correction and coordinate conversion. Finally, an existing filter is used to perform denoising processing on the initial three-dimensional point cloud data to generate sub-pixel level three-dimensional topography data. For example, a statistical filter can be used to perform denoising processing on the initial three-dimensional point cloud data. The statistical filter first calculates the average distance of each point cloud data point and a number of adjacent points in its neighborhood. Then, a standard deviation threshold is set according to the average distance of all points. The noise points with an average distance exceeding the threshold are removed, and the effective point cloud data is retained. The denoised point cloud data is the sub-pixel level three-dimensional topography data.

[0083] It should be noted that the three-dimensional topography data in the present application refers to point cloud data reflecting the three-dimensional morphology of the suspected machining defect area. The determination of the three-dimensional topography data converts the suspected machining defect area from the planar features of the two-dimensional image to three-dimensional quantitative information with accurate spatial dimensions, which not only breaks through the limitation that the two-dimensional image cannot accurately represent the three-dimensional morphology of the defect (such as the depth of the recess, the height of the protrusion, and the cross-sectional profile of the scratch), but also provides a high-precision three-dimensional data basis for subsequent registration with the reference CAD model and deviation calculation, ensuring that the evaluation of the machining quality of precision parts is upgraded from “qualitative judgment” to “quantitative analysis”, and effectively improving the accuracy and reliability of the machining quality monitoring of precision parts.

[0084] In step S104, the three-dimensional topography data is non-rigidly registered with the reference CAD model of the target precision part, and a three-dimensional spatial deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model is determined, and a topography deviation field of the suspected machining defect area is generated.

[0085] In some embodiments, the non-rigid registration of the three-dimensional topography data with the reference CAD model of the target precision part is specifically implemented by the following steps:

[0086] An initial corresponding point pair is constructed according to the three-dimensional topography data and the reference CAD model;

[0087] An initial registration transformation matrix is obtained by rigidly registering the initial corresponding point pair using an iterative closest point algorithm;

[0088] A non-rigid transformation model is constructed according to the initial registration transformation matrix;

[0089] Based on the non-rigid transformation model, the three-dimensional topography data is iteratively optimized to minimize the point cloud distance error, and the non-rigid registration of the three-dimensional topography data with the reference CAD model is completed.

[0090] It should be noted that the reference CAD model in the present application refers to a computer-aided design (CAD) model used as a reference standard in engineering scenarios such as three-dimensional geometric shape design, part machining quality detection, or point cloud registration. It is a digital model constructed based on the design drawings, technical parameters, or theoretical geometric features of the part, which contains accurate three-dimensional geometric information of the part (such as surface profile, size parameter, feature structure, etc.), and is used for comparison with the three-dimensional topography data (such as point cloud data obtained by scanning) actually obtained, to evaluate the deviation between the actual part and the design target.

[0091] In a specific implementation, first, when constructing initial corresponding point pairs according to the three-dimensional topographic data and the reference CAD model, a uniform sampling algorithm is used to select a certain number of representative points from the three-dimensional topographic data point cloud, and the surface of the reference CAD model is meshed and sampled to obtain a corresponding number of model points. The spatial Euclidean distance between the three-dimensional topographic data representative points and the model points is calculated, and the points with a distance less than a set threshold are matched two by two to form initial corresponding point pairs. The initial corresponding point pairs refer to the point-to-point matching relationship set initially established between the three-dimensional topographic data and the reference CAD model. Second, when performing rigid registration on the initial corresponding point pairs using the iterative closest point algorithm, a rotation matrix and a translation vector that minimize the spatial position deviation of the two points are calculated according to the corresponding point pairs (the rotation matrix and the translation vector can be determined according to the singular value decomposition method, which is not described here). The three-dimensional topographic data is spatially transformed according to the rotation matrix and the translation vector, and the nearest neighbor points of the transformed three-dimensional topographic data points on the reference CAD model surface are found again to update the corresponding point pairs. When the deviation of the corresponding point pairs converges, the obtained rotation matrix and translation vector are combined into an initial registration transformation matrix. The initial registration transformation matrix refers to a mathematical matrix that preliminarily aligns the three-dimensional topographic data to the reference CAD model coordinate system through rigid transformation. Third, when constructing a non-rigid transformation model according to the initial registration transformation matrix, the three-dimensional topographic data after initial registration is used as the basis, and an interpolation method based on a radial basis function is used. The feature points (such as curvature points) in the three-dimensional topographic data are taken as control points, a spatial deformation function (specifically a radial basis function) is constructed based on the deviation of the control points and the corresponding points of the reference CAD model, and the non-control points are flexibly adjusted according to the spatial deformation function to form a non-rigid transformation model. The non-rigid transformation model refers to a mathematical model that can realize local flexible deformation of the three-dimensional topographic data relative to the reference CAD model, and can process non-rigid deformation of a part caused by machining. Finally, when iteratively optimizing the non-rigid transformation model with the goal of minimizing the point cloud distance error, the three-dimensional topographic data is first deformed by substituting it into the non-rigid transformation model. The shortest distance (i.e., the point cloud distance error) between the deformed data point cloud and the corresponding model points on the surface of the reference CAD model is calculated. Then, the weight coefficients of the radial basis function are adjusted to reduce the point cloud distance error. When the point cloud distance error reaches a minimum value, the three-dimensional topographic data and the reference CAD model achieve precise geometric fitting, and non-rigid registration is completed.

[0092] It should be noted that the non-rigid registration of the three-dimensional topography data and the reference CAD model in the present application refers to the process of eliminating local geometric deviations caused by part processing deformation by flexibly adjusting the spatial form of the three-dimensional topography data. Through non-rigid registration, not only can accurate position correspondence be provided for subsequent geometric error analysis, but also a reliable geometric alignment basis can be provided for part quality detection, defect identification (such as local concave and convex) and model correction in reverse engineering, ensuring that subsequent analysis or processing correction work is based on accurate form matching, avoiding error misjudgment or correction deviation caused by the inability of rigid registration to handle non-rigid deformation.

[0093] In some embodiments, determining the three-dimensional spatial deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model, and then generating the topography deviation field of the suspected processing defect area is specifically implemented by the following steps:

[0094] Performing spatial position mapping on the three-dimensional topography data and the reference CAD model after non-rigid registration, and determining the corresponding nearest point of each point cloud data point in the three-dimensional topography data on the reference CAD model;

[0095] Calculating the spatial coordinate difference between each point cloud data point and its corresponding nearest point to obtain the corresponding three-dimensional spatial deviation vector of each point cloud data point;

[0096] Intercepting the three-dimensional spatial deviation vector of the suspected processing defect area, and constructing the topography deviation field of the suspected processing defect area from all the intercepted three-dimensional spatial deviation vectors.

[0097] In a specific implementation, firstly, when performing spatial position mapping on the non-rigidly registered three-dimensional topography data and the reference CAD model to determine the nearest point of each point cloud data point on the reference CAD model, a k-neighbor search algorithm is adopted, the surface of the reference CAD model is first discretized into a triangular mesh, for each point cloud data point in the three-dimensional topography data, the nearest triangular mesh in the triangular mesh of the reference CAD model to the point cloud data point is searched, and the vertical projection point of the point cloud data point to the triangular mesh is taken as the corresponding nearest point of the point cloud data point on the reference CAD model, the nearest point refers to the point on the surface of the reference CAD model that has the smallest spatial distance to the point cloud data point in the three-dimensional topography data; secondly, when calculating the spatial coordinate difference value of each point cloud data point and the corresponding nearest point to obtain a three-dimensional spatial deviation vector, the three-dimensional coordinates (X1, Y1, Z1) of the point cloud data point and the three-dimensional coordinates (X2, Y2, Z2) of the nearest point are extracted respectively, the deviation values in the X, Y and Z directions are calculated in the manner of X direction deviation ΔX = X1-X2, Y direction deviation ΔY = Y1-Y2 and Z direction deviation ΔZ = Z1-Z2, and ΔX, ΔY and ΔZ are combined into the three-dimensional spatial deviation vector corresponding to the point cloud data point, the three-dimensional spatial deviation vector refers to a vector describing the position deviation of each point in the three-dimensional topography data relative to the nearest point of the reference CAD model in the X, Y and Z coordinate axis directions; finally, the three-dimensional spatial deviation vector of the suspected machining defect area is intercepted, and a topography deviation field of the suspected machining defect area is constructed from all the intercepted three-dimensional spatial deviation vectors, that is, all the intercepted three-dimensional spatial deviation vectors are arranged according to the corresponding spatial positions to form the topography deviation field of the suspected machining defect area.

[0098] It should be noted that the topography deviation field in the present application refers to a set of deviation vectors of each position in the suspected machining defect area relative to the reference CAD model, and the topography deviation field can clearly reflect the deviation size, direction and distribution law of each position in the suspected machining defect area, which provides direct geometric data support for judging the severity of the machining defect.

[0099] In step S105, the target precision part is classified according to the machining quality based on the local surface instability features of the suspected machining defect area in the topography deviation field and the multispectral feature vectors in the fused feature map.

[0100] In some embodiments, the machining quality classification of the target precision part based on the local surface instability features of the suspected machining defect area in the topography deviation field and the multispectral feature vectors in the fused feature map is implemented by the following steps:

[0101] The local surface instability features of the suspected machining defect area are extracted from the topography deviation field, and the local surface instability features include surface roughness, curvature mutation rate and recess depth;

[0102] extracting a multispectral feature vector corresponding to the suspected machining defect area from the fusion feature map, the multispectral feature vector including a spectral reflection peak wavelength and a waveband contrast;

[0103] generating a comprehensive feature matrix from the local surface instability feature parameter and the multispectral feature vector;

[0104] inputting the comprehensive feature matrix into a pre-trained deep learning classification model, and outputting a machining quality category of the target precision part.

[0105] In a specific implementation, first, the local surface instability features (including surface roughness, curvature mutation rate and recess depth) of the suspected machining defect region are extracted from the topographic deviation field. The arithmetic mean deviation method is used to calculate the arithmetic mean of the deviation values of all point cloud data points in the suspected machining defect region as the surface roughness. When the curvature mutation rate is extracted, the local curvature of each point cloud data point in the suspected machining defect region is calculated through principal component analysis, and then the proportion of the number of times that the local curvature difference of adjacent point cloud data points exceeds a set threshold value to the total amount of the local curvature difference is taken as the curvature mutation rate. The point cloud data points with negative deviation values (indicating lower than the reference CAD model) in the suspected machining defect region are screened, and the absolute value of the minimum deviation value is taken as the recess depth. The surface roughness, curvature mutation rate and recess depth form the local surface instability features. All the local surface instability features are a set of quantitative parameters for describing the abnormal surface geometry of the suspected region. Second, the multispectral feature vector (including spectral reflection peak wavelength and band contrast) corresponding to the suspected machining defect region is extracted from the fusion feature map. The corresponding suspected machining defect region is intercepted in the fusion feature map, and the gray values of each spectral channel in the suspected machining defect region are extracted. The spectral wavelength corresponding to the spectral channel with the maximum gray value is taken as the spectral reflection peak wavelength, and the average gray value difference ratio of the spectral channel corresponding to the spectral reflection peak wavelength and the adjacent spectral channel is calculated as the band contrast (i.e., the spectral reflection peak wavelength is taken as the spectral reflection peak channel, the two spectral channels directly adjacent to the left and right of the peak channel (left adjacent spectral channel and right adjacent spectral channel) are selected, the difference between the average gray value of the waveband monitoring image under the peak channel and the average gray value of the waveband monitoring image under the left adjacent channel, and the difference between the average gray value of the waveband monitoring image under the peak channel and the average gray value of the waveband monitoring image under the right adjacent channel are calculated, then the two differences are divided by the average gray value of the peak channel to obtain two ratios, and the average of the two ratios is taken as the final band contrast. If the peak channel is located at the leftmost side of all spectral channels (without a left adjacent spectral channel), only the difference between the average gray value of the waveband monitoring image under the peak channel and the average gray value of the waveband monitoring image under the right adjacent channel is calculated and divided by the average gray value corresponding to the peak channel as the band contrast. If the peak channel is located at the rightmost side of all spectral channels (without a right adjacent channel), only the difference between the average gray value of the waveband monitoring image under the peak channel and the average gray value of the waveband monitoring image under the left adjacent channel is calculated and divided by the average gray value corresponding to the peak channel as the band contrast. The two are combined to form a multispectral feature vector. The multispectral feature vector is a vector for describing the reflection characteristics of the suspected region under different spectral bands.Then, parameters of the local surface instability feature (surface roughness, curvature mutation rate, and recess depth) and components of the multispectral feature vector (reflection peak wavelength and band contrast) are arranged in columns to generate a comprehensive feature matrix, which is a data matrix integrating the local surface instability feature and the multispectral feature; finally, when the comprehensive feature matrix is input into the pre-trained deep learning classification model to output the processing quality category, the pre-trained deep learning classification model adopts a convolutional neural network (CNN), first extracts features of the comprehensive feature matrix through a convolutional layer of the deep learning classification model, simplifies the feature dimension through a pooling layer, then maps the extracted features to preset processing quality categories (such as qualified, slight defect, and serious defect) through a fully connected layer, and finally outputs the processing quality category corresponding to the target precision part, wherein the pre-trained deep learning classification model is trained through a large number of sample data (comprehensive feature matrix of the local surface instability feature and the multispectral feature of the part, and corresponding "qualified", "slight defect", and "serious defect" category labels) labeled with the processing quality category, which is not described herein.

[0106] It should be noted that the processing quality category of the target precision part refers to the grade of the part processing quality divided according to the preset standard, such as qualified, slight defect, and serious defect.

[0107] In addition, another aspect of the present application provides a processing quality monitoring system for a precision part with a complex structure in some embodiments, which is described with reference to Figure 3 The figure is a structural schematic diagram of a processing quality monitoring system for a precision part with a complex structure according to some embodiments of the present application, which includes an acquisition module 201, a processing module 202, and an execution module 203, which are described as follows:

[0108] The acquisition module 201 is mainly used to acquire a multispectral image sequence of a target precision part after processing, and to perform feature fusion on the multispectral image sequence to generate a fusion feature map representing the surface feature of the target precision part.

[0109] The processing module 202 is mainly used to extract texture features and spectral reflection characteristic distributions in the fusion feature map, and to identify spatial distribution features of suspected processing defect regions by combining the texture features, the spectral reflection characteristic distributions, and a preset processing defect knowledge graph.

[0110] The processing module 202 is also used to control a three-dimensional line scanning module to perform local three-dimensional point cloud reconstruction on the suspected processing defect regions based on the spatial distribution features of the suspected processing defect regions, and to generate sub-pixel level three-dimensional topographic data.

[0111] In addition, the processing module 202 is further configured to perform non-rigid registration between the three-dimensional topography data and a reference CAD model of the target precision part, and determine a three-dimensional space deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model, and further generate a topography deviation field of the suspected machining defect region.

[0112] The execution module 203 is mainly configured to perform machining quality classification on the target precision part based on the local surface instability features of the suspected machining defect region in the topography deviation field and the multi-spectrum feature vectors in the fusion feature map.

[0113] In addition, the present application further provides a computer device, which comprises a memory and a processor, the memory stores code, and the processor is configured to acquire the code and execute the above-mentioned machining quality monitoring method for precision parts with complex structures.

[0114] In some embodiments, the reference Figure 4 The figure is a structural schematic diagram of a computer device for implementing the machining quality monitoring method for precision parts with complex structures according to some embodiments of the present application. The machining quality monitoring method for precision parts with complex structures in the above-mentioned embodiments can be implemented by the computer device shown in the figure, which comprises at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304. Figure 4

[0115] The processor 301 can be a general central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more circuits for controlling the execution of the machining quality monitoring method for precision parts with complex structures in the present application.

[0116] The communication bus 302 can be used to transmit information between the above-mentioned components.

[0117] ​The memory 303 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM), or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magneto-optical disk, a magnetic disk or other magnetic storage device, or any other medium capable of storing desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto. The memory 303 can exist independently of the processor 301 and be connected to the processor 301 through the communication bus 302. The memory 303 can also be integrated with the processor 301.

[0118] The memory 303 is configured to store program codes for implementing the solutions of the present application, and the processor 301 is configured to control the execution of the program codes. The program codes can include one or more software modules. The determination of the processing quality monitoring method for the complex structure precision part in the above embodiments can be implemented by one or more software modules in the program codes of the processor 301 and the memory 303.

[0119] The communication interface 304 is configured to communicate with other devices or communication networks, such as an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), etc., using any transceiver-like mechanism.

[0120] In a specific implementation, as an embodiment, the computer device can include a plurality of processors, each of which can be a single-CPU processor or a multi-CPU processor. The processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0121] The computer device described above can be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device can be a desktop computer, a laptop computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of the computer device.

[0122] In addition, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned machining quality monitoring method for complex structure precision parts.

[0123] Although the preferred embodiments of the present application have been described, those skilled in the art who are familiar with the basic inventive concept can make additional changes and modifications to the embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0124] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for monitoring the machining quality of a complex-structure precision part, characterized in that, The method comprises the following steps: obtaining a multi-spectral image sequence of a target precision part after processing, and performing feature fusion on the multi-spectral image sequence to generate a fusion feature map representing the surface features of the target precision part; wherein, spatial registration is performed on each band monitoring image in the multi-spectral image sequence to obtain a registered image sequence; principal component analysis algorithm is used to perform feature dimension reduction on each registered image in the registered image sequence to obtain a dimension-reduced feature matrix corresponding to each registered image; and a fusion feature map representing the surface features of the target precision part is generated based on all the dimension-reduced feature matrices; extracting texture features and spectral reflectance characteristics distribution in the fusion feature map, and identifying the spatial distribution features of the suspected machining defect area by combining the texture features and the spectral reflectance characteristics distribution with a preset machining defect knowledge graph; based on the spatial distribution features of the suspected machining defect area, controlling a three-dimensional line scanning module to perform local three-dimensional point cloud reconstruction on the suspected machining defect area to generate sub-pixel level three-dimensional topography data; performing non-rigid registration on the three-dimensional topography data and a reference CAD model of the target precision part, and determining a three-dimensional space deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model, and further generating a topography deviation field of the suspected machining defect area; based on the local surface instability features of the suspected machining defect area in the topography deviation field and the multi-spectral feature vectors in the fusion feature map, classifying the machining quality of the target precision part.

2. The method of claim 1, wherein, extracting the texture features and the spectral reflectance characteristics distribution in the fusion feature map specifically comprises: performing multi-directional filtering processing on the fusion feature map to generate a multi-directional filtering response map; extracting the texture features of the fusion feature map based on the multi-directional filtering response map; performing spectral channel separation on the fusion feature map to obtain a gray value matrix of each spectral channel; calculating the spectral reflectance vectors of the corresponding pixel points according to the gray value matrix of each spectral channel, and constructing the spectral reflectance characteristics distribution of the fusion feature map according to all the spectral reflectance vectors.

3. The method of claim 1, wherein, identifying the spatial distribution features of the suspected machining defect area by combining the texture features and the spectral reflectance characteristics distribution with a preset machining defect knowledge graph specifically comprises: performing feature normalization processing on the texture features and the spectral reflectance characteristics distribution to generate a normalized feature vector; performing similarity matching on the normalized feature vector and the defect feature vectors of each machining defect type in the preset machining defect knowledge graph to obtain a feature similarity set; extracting the suspected machining defect area of the target precision part according to the feature similarity set; performing morphological filtering and connected domain analysis on the suspected machining defect area to further determine the spatial distribution features of the suspected machining defect area.

4. The method of claim 1, wherein, based on the spatial distribution features of the suspected machining defect area, controlling a three-dimensional line scanning module to perform local three-dimensional point cloud reconstruction on the suspected machining defect area to generate sub-pixel level three-dimensional topography data specifically comprises: according to the spatial distribution features of the suspected machining defect area, calculating the scanning path and parameter configuration of the three-dimensional line scanning module; Based on the scanning path and the parameter configuration, a three-dimensional line scanning module is controlled to perform layered scanning on the suspected machining defect area to obtain a sub-pixel level depth image; The depth image is subjected to distortion correction and coordinate conversion to generate initial three-dimensional point cloud data; The initial three-dimensional point cloud data is subjected to denoising processing to generate sub-pixel level three-dimensional topography data.

5. The method of claim 1, wherein, The three-dimensional topography data and a reference CAD model of the target precision part are subjected to non-rigid registration, specifically including: An initial corresponding point pair is constructed according to the three-dimensional topography data and the reference CAD model; An iterative closest point algorithm is used to perform rigid registration on the initial corresponding point pair to obtain an initial registration transformation matrix; A non-rigid transformation model is constructed according to the initial registration transformation matrix; Based on the non-rigid transformation model, the three-dimensional topography data is iteratively optimized with the objective of minimizing point cloud distance error to complete non-rigid registration of the three-dimensional topography data and the reference CAD model.

6. The method of claim 1, wherein, A multi-spectral camera is used to obtain a multi-spectral image sequence of the target precision part after machining.

7. A system for monitoring machining quality of a precision part with a complex structure, for executing the method for monitoring machining quality of a precision part with a complex structure according to any one of claims 1 to 6, characterized in that, The system includes: An acquisition module is configured to acquire a multi-spectral image sequence of the target precision part after machining, and perform feature fusion on the multi-spectral image sequence to generate a fused feature map representing surface features of the target precision part; wherein each band monitoring image in the multi-spectral image sequence is subjected to spatial registration to obtain a registered image sequence; a principal component analysis algorithm is used to perform feature dimension reduction on each registered image in the registered image sequence to obtain a dimension-reduced feature matrix corresponding to each registered image; and a fused feature map representing surface features of the target precision part is generated based on all the dimension-reduced feature matrices; A processing module is configured to extract texture features and spectral reflection characteristic distributions in the fused feature map, and identify spatial distribution features of suspected machining defect areas by combining the texture features and the spectral reflection characteristic distributions with a pre-set machining defect knowledge graph; The processing module is further configured to control a three-dimensional line scanning module to perform local three-dimensional point cloud reconstruction on the suspected machining defect areas based on the spatial distribution features of the suspected machining defect areas to generate sub-pixel level three-dimensional topography data; The processing module is further configured to perform non-rigid registration of the three-dimensional topography data and a reference CAD model of the target precision part, and determine a three-dimensional spatial deviation vector of each point cloud data point in the three-dimensional topography data relative to the reference CAD model to further generate a topography deviation field of the suspected machining defect areas; An execution module is configured to perform machining quality classification of the target precision part based on local surface instability features of suspected machining defect areas in the topography deviation field and multi-spectral feature vectors in the fused feature map.

8. A computer device, comprising: The computer device includes a memory and a processor, the memory stores code, and the processor is configured to acquire the code and execute the machining quality monitoring method for complex structure precision parts according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the machining quality monitoring method for complex structure precision parts according to any one of claims 1 to 6.

Images