A method and system for autonomous detection of ground surfaces of precision workpieces

Abstract

The application provides a precision workpiece grinding processing surface autonomous detection method and system, and belongs to the precision workpiece processing detection technical field. The method comprises the following steps: acquiring multi-modal image data of a target precision workpiece grinding processing surface and grinding parameters in a processing process; performing fusion preprocessing on the multi-modal image data to generate fusion image features; performing feature extraction and cleaning on the grinding parameters to generate grinding process features; inputting the fusion image features into a first-level defect screening model to obtain a first detection result; when a confidence of any first defect exists is greater than a preset first confidence threshold, inputting the grinding process features into a defect mode diagnosis model to obtain a defect mode diagnosis result of a suspected defect area; determining a target defect detection model according to the defect mode diagnosis result; and inputting a part corresponding to the suspected defect area in the fusion image features into the target defect detection model to obtain surface defect geometric feature parameters of the suspected defect area.

Description

Technical Field

[0001] This application relates to the field of precision workpiece machining and inspection technology, and in particular to an autonomous inspection method and system for the grinding surface of precision workpieces. Background Technology

[0002] In modern manufacturing, precision workpieces are widely used in high-end fields such as aerospace and medical devices. The surface quality of their grinding process directly affects product performance and reliability. Currently, automated inspection methods are commonly used in industrial settings. Among them, machine vision-based inspection can quickly identify some surface defects, but it is difficult to effectively distinguish microscopic defects caused by abnormal grinding parameters, and its detection accuracy for complex textured surfaces is limited. Sensor-based inspection methods can acquire information about the processing process, but they cannot intuitively present the defect morphology and are easily affected by environmental noise.

[0003] Therefore, there is an urgent need for an autonomous detection method for the surface of precision workpieces during grinding, which can more accurately and efficiently identify surface defects in order to improve the processing quality and production efficiency of precision workpieces. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method and system for autonomous inspection of the surface of precision workpieces after grinding.

[0005] A first aspect of this application provides a method for autonomous inspection of the ground surface of a precision workpiece, comprising: Acquire multimodal image data of the surface of the target precision workpiece being ground, as well as the grinding parameters during the processing; The multimodal image data is fused and preprocessed to generate fused image features, which are also used to extract and clean the grinding parameters to generate grinding process features. The fused image features are input into the first-level defect screening model to obtain a first detection result. The first detection result includes the location information of at least one suspected defect area and the corresponding first defect existence confidence level. The defect pattern diagnosis module is used to input the grinding process features into the defect pattern diagnosis model when any of the first defects has a confidence level greater than a preset first confidence threshold, and obtain the defect pattern diagnosis result of the suspected defect area. The diagnosis result includes at least one defect pattern and its corresponding pattern confidence level. Based on the defect pattern diagnosis results, the target defect detection model is determined; The fused image features are input into the target defect detection model to obtain the surface defect geometric feature parameters of the suspected defect area.

[0006] A second aspect of this application provides an autonomous inspection system for the surface of precision workpieces after grinding, comprising: The data acquisition module is used to acquire multimodal image data of the surface of the target precision workpiece being ground, as well as grinding parameters during the processing. The data processing module is used to perform fusion preprocessing on the multimodal image data to generate fused image features, and also to extract and clean the grinding parameters to generate grinding process features. The defect screening module is used to input the fused image features into the first-level defect screening model to obtain a first detection result. The first detection result includes the location information of at least one suspected defect area and the corresponding first defect existence confidence level. The defect pattern diagnosis module is used to input the grinding process features into the defect pattern diagnosis model when any of the first defects has a confidence level greater than a preset first confidence threshold, and obtain the defect pattern diagnosis result of the suspected defect area. The diagnosis result includes at least one defect pattern and its corresponding pattern confidence level. The defect detection model determination module is used to determine the target defect detection model based on the defect pattern diagnosis results. The target defect detection module is used to input the fused image features into the target defect detection model to obtain the surface defect geometric feature parameters of the suspected defect area.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for autonomous detection of the surface of a precision workpiece after grinding.

[0008] In a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described autonomous detection system for the grinding surface of a precision workpiece.

[0009] The beneficial effects of the autonomous detection method and system for the grinding surface of precision workpieces provided in this application are as follows: By acquiring multimodal image data and grinding parameters and performing fusion processing, this application can comprehensively capture information about the workpiece surface and the processing process, improving the comprehensiveness and accuracy of detection and effectively reducing the probability of missed and false detections. Secondly, a first-level defect screening model is used for preliminary screening to quickly locate suspected defect areas. Combined with subsequent defect pattern diagnosis models and target defect detection models, a hierarchical diagnosis mechanism is constructed, which not only improves detection efficiency but also accurately identifies defect patterns and determines their geometric feature parameters, providing detailed data support for subsequent defect repair and process optimization. In addition, this application reduces manual intervention, reducing labor costs while ensuring the objectivity and accuracy of detection results, thereby improving the production quality and efficiency of precision workpieces. Attached Figure Description

[0010] Figure 1 A flowchart illustrating an autonomous inspection method for the grinding surface of a precision workpiece, provided in an embodiment of this application; Figure 2 This application provides a structural block diagram of an autonomous surface inspection system for precision workpiece grinding, as an embodiment of the present application. Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 This application provides a flowchart illustrating an autonomous inspection method for the surface of a precision workpiece after grinding, as provided in one embodiment. The method includes: S101: Acquire multimodal image data of the surface of the target precision workpiece being ground, as well as grinding parameters during the processing; In this embodiment, the target precision workpiece is the precision workpiece to be inspected. This application can acquire multimodal image data of the ground surface of the target precision workpiece under different spectral bands, different spatial dimensions, and different physical properties using various devices such as a multispectral camera, a 3D laser scanner, and an ultrasonic flaw detector. Specifically, the multispectral camera acquires multi-band image data of the ground surface of the target precision workpiece, including visible light and near-infrared light; the 3D laser scanner acquires the 3D morphology data of the ground surface of the target precision workpiece; and the ultrasonic flaw detector acquires image data corresponding to internal defects in the target precision workpiece. This embodiment monitors and acquires grinding parameters during the processing of the target precision workpiece using sensors; these grinding parameters include: grinding wheel speed, feed rate, grinding depth, grinding force, coolant flow rate, and temperature.

[0014] S102: Perform fusion preprocessing on multimodal image data to generate fused image features; extract and clean features from grinding parameters to generate grinding process features.

[0015] In this embodiment, the fusion preprocessing of multimodal image data includes: using a deep learning-based image registration algorithm, with the three-dimensional topography data as a reference, to accurately register other modal images with the three-dimensional topography data; and using a wavelet transform fusion algorithm to extract and fuse features from the registered multimodal image data to generate fused image features that include multimodal information.

[0016] In this embodiment, feature extraction and cleaning of grinding parameters includes: using wavelet transform to denoise the time series of grinding parameters to remove noise interference; and using principal component analysis to reduce the dimensionality of the denoised grinding parameters and extract the grinding process features.

[0017] S103: Input the fused image features into the first-level defect screening model to obtain the first detection result. The first detection result includes the location information of at least one suspected defect area and the corresponding confidence level of the existence of the first defect.

[0018] In this embodiment, the first-level defect screening model is constructed based on a combination of convolutional neural networks and attention mechanisms. After the fused image features are input into the first-level defect screening model, the model analyzes these features through multiple convolutional layers and attention modules to obtain a first detection result. This first detection result includes the location information of at least one suspected defect region and the corresponding confidence level of the first defect's existence. During the training process of the first-level defect screening model, transfer learning techniques are employed. It is pre-trained using a publicly available image dataset and then adjusted using actual collected images of the surface of precision workpieces undergoing grinding.

[0019] S104: When the confidence level of any first defect is greater than the preset first confidence threshold, the grinding process features are input into the defect mode diagnosis model to obtain the defect mode diagnosis result of the suspected defect area. The diagnosis result includes at least one defect mode and its corresponding mode confidence level. In this embodiment, when the confidence level of any first defect is greater than a preset first confidence threshold, the grinding process features are input into a pre-trained defect pattern diagnosis model. The defect pattern diagnosis model is constructed based on a long short-term memory network. This defect pattern diagnosis model can learn the time series information and nonlinear relationship of the grinding process features to obtain the defect pattern diagnosis result of the suspected defect area. The diagnosis result includes at least one defect pattern and its corresponding pattern confidence level. The defect pattern includes, but is not limited to, grinding burns, cracks, vibration marks, and excessive surface roughness.

[0020] S105: Determine the target defect detection model based on the defect pattern diagnosis results; In this embodiment, a target defect detection model is determined from a pre-built defect detection model library based on the defect pattern diagnosis results. The defect detection model library contains multiple defect detection models trained for different defect patterns, and each defect detection model adopts a different network structure and parameter configuration. For example, the defect detection model for grinding burns adopts an infrared image feature analysis network, and the defect detection model for cracks adopts a convolutional neural network that combines edge detection and feature extraction.

[0021] S106: Input the portion of the fused image features corresponding to the suspected defect area into the target defect detection model to obtain the surface defect geometric feature parameters of the suspected defect area.

[0022] In this embodiment, the portion of the fused image features corresponding to the suspected defect area is input into the target defect detection model; the target defect detection model performs in-depth analysis of the input image features and obtains the surface defect geometric feature parameters of the suspected defect area through the model's output layer, wherein the geometric feature parameters include defect length, width, depth, area, curvature, etc.

[0023] In summary, this application can also store the results obtained from each detection, including suspected defect area location information, defect pattern diagnosis results, surface defect geometric feature parameters, etc., into the detection result database; periodically analyze the detection result database, and when the frequency of a certain type of defect is detected to exceed a preset threshold, use new detection data to retrain and adjust the parameters of the first-level defect screening model, defect pattern diagnosis model and target defect detection model to improve the accuracy and adaptability of the detection system.

[0024] For example, in the field of aero-engine manufacturing, engine blades made of titanium alloy are selected as the target precision workpieces.

[0025] During the grinding process, a high-resolution industrial camera is used to acquire visible light images, an infrared thermal imager to acquire infrared thermal images, and a laser confocal microscope to acquire surface micro-morphology images, thereby obtaining multimodal image data. In addition, a speed sensor installed on the grinding machine monitors the grinding wheel speed, a feed rate sensor records the feed rate, a force sensor measures the grinding force, and a temperature sensor monitors the coolant temperature, etc., to obtain grinding parameters.

[0026] In the data preprocessing stage: For multimodal image data, an algorithm based on feature point matching is first used for image registration to unify the coordinate systems of different modal images; then, a multimodal image fusion network is used to fuse the registered images to generate fused image features. For grinding parameters, a median filtering algorithm is used to remove noise from the data, and the data is mapped to the 0-1 interval through normalization. Then, Fourier transform is used to extract frequency domain features, which are combined with the original time domain features to generate grinding process features.

[0027] In the first-level screening stage: the fused image features are input into the pre-trained first-level defect screening model. The first-level defect screening model extracts image features through convolutional layers, reduces dimensions through pooling layers, and finally outputs the location information of suspected defect areas and the corresponding confidence level of the existence of the first defect through fully connected layers.

[0028] In the defect mode diagnosis stage: Assuming the preset first confidence threshold is 0.6, when the confidence of the first defect in a certain area reaches 0.7, the grinding process features are input into the defect mode diagnosis model. The defect mode diagnosis model captures the time series information of grinding parameters and outputs the defect mode diagnosis result of the suspected defect area. For example, the confidence of the mode diagnosed as grinding burn is 0.8, and the confidence of the mode of vibration marks is 0.2.

[0029] In the target defect detection model determination stage: based on the defect pattern diagnosis results, a target defect detection model based on infrared image feature analysis for grinding burns is determined from the model library.

[0030] In the target defect detection stage: the part of the fused image features corresponding to the suspected defect area is input into the target defect detection model. The target defect detection model analyzes the temperature distribution features of the infrared image and combines morphological processing and threshold segmentation algorithms to obtain the surface defect geometric feature parameters of the area; for example, the length of the burn area is 2mm, the width is 1.5mm, and the depth is estimated to be 0.1mm by comparing with the standard sample.

[0031] As can be seen from the above, this application, by acquiring multimodal image data and grinding parameters and fusing them, can comprehensively capture information about the workpiece surface and processing process, improving the comprehensiveness and accuracy of detection and effectively reducing the probability of missed and false detections. Secondly, a first-level defect screening model is used for preliminary screening to quickly locate suspected defect areas. Combined with subsequent defect pattern diagnosis models and target defect detection models, a hierarchical diagnosis mechanism is constructed, which not only improves detection efficiency but also accurately identifies defect patterns and determines their geometric feature parameters, providing detailed data support for subsequent defect repair and process optimization. Furthermore, this application reduces manual intervention, lowering labor costs while ensuring the objectivity and accuracy of detection results, thereby improving the production quality and efficiency of precision workpieces.

[0032] In one embodiment of this application, before obtaining the defect mode diagnosis result of the suspected defect area based on inputting the grinding process features into the defect mode diagnosis model, the method further includes: Obtain the defect pattern labels and corresponding historical grinding parameters of the surface of the precision workpiece in the past. Based on defect mode labels, historical grinding parameters are classified to determine the grinding parameter range corresponding to each type of defect mode.

[0033] In this embodiment, historical precision workpiece grinding data is obtained from the enterprise's historical production database, along with defect pattern labels and corresponding historical grinding parameters for the ground surfaces of the historical precision workpieces. The defect pattern labels include common defect types such as grinding burns, surface cracks, vibration marks, and dimensional deviations. Historical grinding parameters include parameters such as grinding wheel speed, feed rate, grinding depth, coolant temperature, and spindle power for each machining operation. After obtaining the historical grinding parameters, this embodiment performs outlier removal and data completion operations to ensure data integrity and accuracy. Furthermore, the defect pattern labels undergo standardized processing to eliminate semantic differences. In one embodiment of this application, historical grinding parameters are classified based on defect mode labels to determine the grinding parameter range corresponding to each type of defect mode, including: For each defect mode, the historical grinding parameters of the precision workpieces with that defect mode are analyzed for attribution, and the grinding parameter types related to that defect mode are selected as the relevant grinding parameters for that defect mode. For each defect pattern, cluster analysis is performed on the relevant grinding parameters of historical precision workpieces with that defect pattern to generate the grinding parameter range corresponding to that defect pattern.

[0034] In this embodiment, attribution analysis is a statistical method used to determine which factors or variables contribute to a specific outcome and to what extent. In the context of detecting defects in knitted fabric patterns, attribution analysis aims to identify the relationship between the types of production parameters (such as grinding speed, feed rate, and wheel grit size) and the defect patterns of precision workpieces. Based on the results of the attribution analysis, clustering the relevant production parameters can identify reasonable ranges for these parameters, thereby defining a reasonable range of production parameters for each defect pattern.

[0035] In this embodiment, a clustering algorithm is used to cluster data points in a multi-dimensional parameter space, using historical grinding parameters as input data. Based on the clustering results and defect pattern labels, the historical grinding parameters are divided into different categories, each corresponding to one or more similar defect patterns. During the clustering process, parameters such as the neighborhood radius and minimum number of sample points can be adjusted to optimize the clustering effect and ensure high similarity among data points of the same category. For each type of defect pattern, the mean, standard deviation, maximum, and minimum values ​​of each grinding parameter are calculated. Using the mean as the center and combining it with the standard deviation, the fluctuation range of the parameters is determined, constructing the grinding parameter range corresponding to each type of defect pattern. For example, for the grinding burn defect pattern, reasonable upper and lower limits of the grinding wheel speed and sensitive intervals of the feed rate are determined to form the grinding parameter characteristic interval for this defect pattern. During the classification process, this embodiment can also manually verify and correct the classification results based on domain expert knowledge, and adjust the boundaries of the grinding parameter ranges based on actual processing experience to improve the accuracy and practicality of the classification results.

[0036] In one embodiment of this application, multimodal image data is fused and preprocessed to generate fused image features, including: Spatial registration is performed on multimodal image data to generate a registered multimodal image set; A wavelet transform fusion algorithm is used to fuse features of the registered multimodal image set to generate a fused feature map; The dimensionality of the fused feature map is reduced to obtain the fused image features.

[0037] In this embodiment, a deep learning-based image registration algorithm is used to accurately register other modal images with the 3D topography data as a reference. A wavelet transform fusion algorithm is then employed to extract and fuse features from the registered multimodal image data, generating a fused feature map. Principal component analysis is used to linearly reduce the dimensionality of the fused feature map. By calculating the covariance matrix and eigenvalue decomposition of the feature map, principal components are extracted, reducing the data dimensionality and obtaining the fused image features.

[0038] In one embodiment of this application, a wavelet transform fusion algorithm is used to perform feature fusion on a spatially registered image to obtain a feature map, including: Each registered multimodal image is subjected to N-level wavelet decomposition to obtain the low-frequency approximation coefficient sub-band and the high-frequency detail coefficient sub-band in multiple directions for each image. The low-frequency approximation coefficient subbands of all images on the same decomposition layer are fused using a weighted average method to obtain the fused low-frequency subbands; The high-frequency detail coefficient subbands of all images in the same decomposition layer and in the same direction are fused using the local region energy maximization rule to obtain the fused high-frequency subbands. The fused low-frequency subband and the fused high-frequency subband are subjected to inverse wavelet transform and reconstructed into a fused feature map.

[0039] In this embodiment, the complexity of each registered multimodal image is quantitatively evaluated by calculating indicators such as information entropy, gradient energy, and texture complexity. Based on the image complexity evaluation results, an adaptive wavelet decomposition layer mapping function is established. For images with high complexity, the wavelet decomposition layer N is increased to obtain more refined image features; for images with low complexity, the decomposition layer is reduced to avoid computational redundancy caused by excessive decomposition. Simultaneously, upper and lower thresholds for the decomposition layer number are set to prevent unreasonable decomposition layer numbers due to image anomalies. Quality weights are calculated for each registered multimodal image based on image sharpness, contrast, and signal-to-noise ratio. When fusing the low-frequency approximation coefficient subbands of all images at the same decomposition layer using a weighted average method, the weighting coefficients are adjusted based on the image quality weights and the energy distribution of each image's low-frequency subbands. Higher weights are assigned to low-frequency subbands of images with high quality and large energy proportions to ensure that the fused low-frequency subbands retain important information to the greatest extent.

[0040] In this embodiment, in addition to the local region energy maximization rule, a fusion strategy based on saliency detection can also be adopted. Using a visual saliency model, the saliency value of each region in the high-frequency detail coefficient sub-band of each image is calculated, and this saliency value is used as an important basis for fusion. When fusing high-frequency detail coefficient sub-bands of all images at the same decomposition layer but in different directions, for regions with high saliency, image coefficients with larger absolute values ​​in those regions are preferentially selected for fusion. Simultaneously, combining gradient information, when the saliency of multiple images in the same region is similar, coefficients that better reflect edge and detail features are selected based on the gradient direction and magnitude, enhancing the ability of the fused high-frequency sub-bands to express image details. In one embodiment of this application, determining a target defect detection model based on defect pattern diagnosis results includes: Select the defect pattern with the highest pattern confidence from the defect pattern diagnosis results as the target defect pattern; The defect detection model that matches the target defect pattern is used as the target defect detection model.

[0041] In this embodiment, the defect mode with the highest pattern confidence score is selected as the target defect mode from the defect mode diagnosis results. This requires normalizing the pattern confidence scores in the defect mode diagnosis results to eliminate the influence of differences in confidence score scales between different models. Simultaneously, through repeated diagnosis, the variance and coefficient of variation of the confidence score for each defect mode are calculated to evaluate its stability. For defect modes with high confidence scores but poor stability, their weight in the selection of the target defect mode is reduced. Based on historical defect data analysis, the historical occurrence probability of each defect mode under the current workpiece type and processing technology is analyzed. The historical occurrence probability and pattern confidence score are then weighted and fused to obtain a comprehensive evaluation score. For example, defect modes with high historical occurrence probabilities and high current pattern confidence scores are assigned higher comprehensive evaluation scores, making the determination of the target defect mode more closely aligned with actual production conditions. In this embodiment, when multiple defect modes have similar comprehensive evaluation scores, the defect mode with the highest confidence level is no longer simply selected. Domain expert knowledge is introduced, and the impact of different defect modes on the workpiece's function is determined by considering the workpiece's usage requirements and key performance indicators. For example, for precision workpieces subjected to high stress, crack defect modes that affect structural strength are prioritized as target defect modes, even if their confidence level is not the highest. This embodiment can also include related defect modes in the target defect mode set if a defect mode is detected and it has a causal relationship or concurrent possibility with other defect modes. For example, when grinding burn defects are detected, since they may cause changes in material properties leading to crack formation, crack defect modes are also included as target defect modes to improve the comprehensiveness of detection. In this embodiment, in addition to simple matching based on defect pattern names, further classification can be performed based on defect pattern characteristic parameters (such as defect size range, shape characteristics, and occurrence location), detection accuracy requirements, and computational resource consumption. After determining the target defect pattern, multiple candidate defect detection models that meet the requirements of the current detection scenario are selected from a pre-established multi-dimensional matching relationship library between defect detection models and defect patterns. Combining the real-time and accuracy requirements of the current detection task, the weights of each performance index are determined using the analytic hierarchy process (AHP), and the candidate models are comprehensively scored. The defect detection model with the highest comprehensive score is selected as the target defect detection model. The performance indexes are obtained by simulating detection of similar defect cases in historical detection data on the candidate models. The performance indexes include: detection accuracy, recall, and detection time.

[0042] In one embodiment of this application, the method for autonomous detection of the ground surface of a precision workpiece further includes: When the confidence level of the first defect is greater than the first confidence threshold but less than the preset second confidence threshold, wherein the second confidence threshold is greater than the first confidence threshold; The grinding process features are input into the defect mode diagnosis model, and the defect mode diagnosis results of the suspected defect area are output. Based on the highest pattern confidence in the defect pattern diagnosis results, adjust the image range input to the target defect detection model: If the highest pattern confidence is greater than or equal to the preset pattern confidence threshold, then the part of the fused image features corresponding to the suspected defect area will be input into the target defect detection model. If the highest pattern confidence is less than the preset pattern confidence threshold, the fused image features will be input into the target defect detection model.

[0043] In this embodiment, based on information such as the defect distribution and detection accuracy of different types of precision workpieces in historical inspection data, a model is trained using machine learning algorithms (such as random forest regression and support vector regression) to establish a mapping relationship between workpiece type, processing parameters and the first confidence threshold and the second confidence threshold. During inspection, the first confidence threshold and the second confidence threshold are adjusted according to the current type of precision workpiece and processing parameters to make the threshold setting more in line with the actual inspection requirements. In this embodiment, when the highest pattern confidence score is less than a preset pattern confidence score threshold, a block-based processing strategy is adopted when the fused image features are input into the target defect detection model. The fused image features are divided into multiple overlapping sub-image blocks, and each sub-image block is preprocessed (preprocessing includes normalization and contrast enhancement operations). Then, they are sequentially input into the target defect detection model for detection. Finally, the detection results of each sub-image block are combined to determine the final detection result of the suspected defect region, thereby improving the accuracy and comprehensiveness of the detection.

[0044] A method for autonomous inspection of the grinding surface of a precision workpiece, corresponding to the above embodiment. Figure 2 This is a structural block diagram of an autonomous surface inspection system for precision workpiece grinding, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The autonomous surface inspection system 20 for grinding of precision workpieces includes: a data acquisition module 21, a data processing module 22, a defect screening module 23, a defect pattern diagnosis module 24, a defect detection model determination module 25, and a target defect detection module 26.

[0045] in, Data acquisition module 21 is used to acquire multimodal image data of the grinding surface of the target precision workpiece and grinding parameters during the processing; Data processing module 22 is used to perform fusion preprocessing on multimodal image data to generate fused image features, and also to extract and clean features of grinding parameters to generate grinding process features; Defect screening module 23 is used to input the fused image features into the first-level defect screening model to obtain the first detection result. The first detection result includes the location information of at least one suspected defect area and the corresponding first defect existence confidence level. Defect mode diagnosis module 24 is used to input grinding process features into defect mode diagnosis model when the confidence of any first defect is greater than the preset first confidence threshold, and obtain defect mode diagnosis results of suspected defect areas. The diagnosis results include at least one defect mode and its corresponding mode confidence. The defect detection model determination module 25 is used to determine the target defect detection model based on the defect pattern diagnosis results; The target defect detection module 26 is used to input the fused image features into the target defect detection model to obtain the surface defect geometric feature parameters of the suspected defect area.

[0046] In one embodiment of this application, the autonomous surface inspection system 20 for grinding processes further includes: The grinding parameter range module 27 is used to: obtain the defect mode labels and corresponding historical grinding parameters of the surface of the precision workpiece grinding process before inputting the grinding process features into the defect mode diagnosis model and obtaining the defect mode diagnosis results of the suspected defect area; classify the historical grinding parameters based on the defect mode labels and determine the grinding parameter range corresponding to each type of defect mode.

[0047] In one embodiment of this application, the grinding parameter range module 27 is further specifically used for: For each defect mode, the historical grinding parameters of the precision workpieces with that defect mode are analyzed for attribution, and the grinding parameter types related to that defect mode are selected as the relevant grinding parameters for that defect mode. For each defect pattern, cluster analysis is performed on the relevant grinding parameters of historical precision workpieces with that defect pattern to generate the grinding parameter range corresponding to that defect pattern.

[0048] In one embodiment of this application, the data processing module 22 is specifically used for: Spatial registration is performed on multimodal image data to generate a registered multimodal image set; A wavelet transform fusion algorithm is used to fuse features of the registered multimodal image set to generate a fused feature map; The dimensionality of the fused feature map is reduced to obtain the fused image features.

[0049] In one embodiment of this application, the data processing module 22 is further specifically used for: Each registered multimodal image is subjected to N-level wavelet decomposition to obtain the low-frequency approximation coefficient sub-band and the high-frequency detail coefficient sub-band in multiple directions for each image. The low-frequency approximation coefficient subbands of all images on the same decomposition layer are fused using a weighted average method to obtain the fused low-frequency subbands; The high-frequency detail coefficient subbands of all images in the same decomposition layer and in the same direction are fused using the local region energy maximization rule to obtain the fused high-frequency subbands. The fused low-frequency subband and the fused high-frequency subband are subjected to inverse wavelet transform and reconstructed into a fused feature map.

[0050] In one embodiment of this application, the defect detection model determination module 25 is specifically used for: Select the defect pattern with the highest pattern confidence from the defect pattern diagnosis results as the target defect pattern; The defect detection model that matches the target defect pattern is used as the target defect detection model.

[0051] In one embodiment of this application, the target defect detection module 26 is further specifically used for: When the confidence level of the first defect is greater than the first confidence threshold but less than the preset second confidence threshold, wherein the second confidence threshold is greater than the first confidence threshold; The grinding process features are input into the defect mode diagnosis model, and the defect mode diagnosis results of the suspected defect area are output. Adjust the image range input to the target defect detection model based on the highest pattern confidence in the defect pattern diagnosis results; If the highest pattern confidence is greater than or equal to the preset pattern confidence threshold, then the part of the fused image features corresponding to the suspected defect area will be input into the target defect detection model. If the highest pattern confidence is less than the preset pattern confidence threshold, the fused image features will be input into the target defect detection model.

[0052] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the data acquisition module 21, data processing module 22, defect screening module 23, defect pattern diagnosis module 24, defect detection model determination module 25, and target defect detection module 26 are shown.

[0053] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0054] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0055] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0056] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the autonomous detection method for grinding surfaces of precision workpieces provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0057] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0058] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0059] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0060] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0061] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0063] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for autonomous inspection of the surface of precision workpieces after grinding, characterized in that, include: Acquire multimodal image data of the surface of the target precision workpiece being ground, as well as the grinding parameters during the processing; The multimodal image data is fused and preprocessed to generate fused image features; The grinding parameters are feature extracted and cleaned to generate grinding process features; The fused image features are input into the first-level defect screening model to obtain a first detection result. The first detection result includes the location information of at least one suspected defect area and the corresponding first defect existence confidence level. When the confidence level of any of the first defects is greater than the preset first confidence threshold, the grinding process features are input into the defect mode diagnosis model to obtain the defect mode diagnosis result of the suspected defect area. The diagnosis result includes at least one defect mode and its corresponding mode confidence level. Based on the defect pattern diagnosis results, the target defect detection model is determined; The portion of the fused image features corresponding to the suspected defect region is input into the target defect detection model to obtain the surface defect geometric feature parameters of the suspected defect region.

2. The method for autonomous inspection of the surface of a precision workpiece after grinding according to claim 1, characterized in that, Before inputting the grinding process features into the defect mode diagnosis model to obtain the defect mode diagnosis result of the suspected defect area, the method further includes: Obtain the defect pattern labels and corresponding historical grinding parameters of the surface of the precision workpiece in the past. Based on the defect pattern labels, the historical grinding parameters are classified to determine the grinding parameter ranges corresponding to each type of defect pattern.

3. The method for autonomous inspection of the surface of a precision workpiece after grinding according to claim 2, characterized in that, The process of classifying historical grinding parameters based on the defect mode labels and determining the grinding parameter ranges corresponding to each type of defect mode includes: For each defect mode, the historical grinding parameters of the historical precision workpieces with that defect mode are subjected to attribution analysis, and the grinding parameter types related to that defect mode are selected as the relevant grinding parameters for that defect mode. For each defect pattern, cluster analysis is performed on the relevant grinding parameters of historical precision workpieces with that defect pattern to generate the grinding parameter range corresponding to that defect pattern.

4. The method for autonomous inspection of the surface of a precision workpiece after grinding according to claim 1, characterized in that, The step of fusing and preprocessing the multimodal image data to generate fused image features includes: Spatial registration is performed on the multimodal image data to generate a registered multimodal image set; The registered multimodal image set is fused using a wavelet transform fusion algorithm to generate a fused feature map. The fusion feature map is subjected to dimensionality reduction processing to obtain the fusion image features.

5. The method for autonomous inspection of the surface of a precision workpiece after grinding according to claim 4, characterized in that, The step of using a wavelet transform fusion algorithm to perform feature fusion on the registered multimodal image set to generate a fused feature map includes: Each registered multimodal image is subjected to N-level wavelet decomposition to obtain the low-frequency approximation coefficient sub-band and the high-frequency detail coefficient sub-band in multiple directions for each image. The low-frequency approximation coefficient subbands of all images on the same decomposition layer are fused using a weighted average method to obtain the fused low-frequency subbands; The high-frequency detail coefficient subbands of all images in the same decomposition layer and in the same direction are fused using the local region energy maximization rule to obtain the fused high-frequency subbands. The fused low-frequency subband and the fused high-frequency subband are subjected to inverse wavelet transform to reconstruct the fused feature map.

6. The method for autonomous inspection of the surface of a precision workpiece after grinding according to claim 1, characterized in that, The step of determining the target defect detection model based on the defect pattern diagnosis results includes: The defect pattern with the highest pattern confidence is selected from the defect pattern diagnosis results as the target defect pattern. The defect detection model that matches the target defect pattern is used as the target defect detection model.

7. The method for autonomous inspection of the surface of a precision workpiece after grinding according to claim 1, characterized in that, Also includes: When the confidence level of the first defect is greater than the first confidence threshold but less than the preset second confidence threshold, wherein the second confidence threshold is greater than the first confidence threshold; The grinding process features are input into the defect mode diagnosis model, and the defect mode diagnosis results of the suspected defect area are output. Based on the highest pattern confidence in the defect pattern diagnosis results, adjust the image range input to the target defect detection model; If the highest pattern confidence is greater than or equal to a preset pattern confidence threshold, then the portion of the fused image features corresponding to the suspected defect region is input into the target defect detection model. If the highest pattern confidence is less than the preset pattern confidence threshold, then the fused image features are input into the target defect detection model.

8. An autonomous surface inspection system for precision workpiece grinding, characterized in that, include: The data acquisition module is used to acquire multimodal image data of the surface of the target precision workpiece being ground, as well as grinding parameters during the processing. The data processing module is used to perform fusion preprocessing on the multimodal image data to generate fused image features, and also to extract and clean the grinding parameters to generate grinding process features. The defect screening module is used to input the fused image features into the first-level defect screening model to obtain a first detection result. The first detection result includes the location information of at least one suspected defect area and the corresponding first defect existence confidence level. The defect pattern diagnosis module is used to input the grinding process features into the defect pattern diagnosis model when any of the first defects has a confidence level greater than a preset first confidence threshold, and obtain the defect pattern diagnosis result of the suspected defect area. The diagnosis result includes at least one defect pattern and its corresponding pattern confidence level. The defect detection model determination module is used to determine the target defect detection model based on the defect pattern diagnosis results. The target defect detection module is used to input the fused image features into the target defect detection model to obtain the surface defect geometric feature parameters of the suspected defect area.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images