A wax mold multi-station synchronous size online detection method and system

Abstract

The application discloses a kind of wax mould multi-station synchronous size online detection method and system, it is related to wax mould detection technical field, the method includes: in multiple wax mould work site deployment detection equipment group, and establish the digital communication of detection equipment group and data processing center;Establish acquisition dataset;Upload to data processing center, establish work site size modeling result;Execute the size evolution process modeling under multiple wax mould work site, configure the deformation mapping model between work site;After mapping unified standard wax injection domain to all wax mould work site upload data, execute image-point cloud's reconstruction feature splicing, establish fusion modeling result;Establish size anomaly identification.The application solves the technical problems that the size detection efficiency of wax mould in the prior art is low and the accuracy of abnormal identification is insufficient due to the data fragmentation between work stations, achieves synchronous online detection and accurate modeling of wax mould multi-station size, improves the technical effects of size detection efficiency and abnormal identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of wax model inspection technology, specifically to a method and system for simultaneous online dimensional inspection of wax models at multiple stations. Background Technology

[0002] In the field of precision casting, wax patterns serve as crucial prototypes for casting formation, and their dimensional accuracy directly impacts the quality of the final casting. Therefore, online inspection of wax pattern dimensions is a vital step in the production process. Currently, traditional wax pattern dimensional inspection often employs a single-station offline inspection method, where inspection is completed at one station before being transferred to the next. This approach is not only inefficient and difficult to adapt to the pace of mass production, but also results in fragmented inspection data from different stations. The lack of correlation analysis of the overall dimensional evolution of the wax pattern leads to an inability to accurately reflect the deformation relationships between different stations, making it prone to misjudgments or omissions of dimensional anomalies. Consequently, it fails to meet the high-efficiency and accuracy requirements of high-precision casting production for wax pattern dimensional inspection.

[0003] Existing technologies suffer from low efficiency in wax mold size detection and insufficient accuracy in anomaly identification due to data fragmentation between workstations. Summary of the Invention

[0004] This application provides a method and system for synchronous online dimensional detection of wax models at multiple workstations, which addresses the technical problems of low efficiency in wax model dimensional detection and insufficient accuracy in anomaly identification due to data fragmentation between workstations in the prior art.

[0005] In view of the above problems, this application provides a method and system for online synchronous dimensional detection of wax molds at multiple stations.

[0006] A first aspect of this application provides a method for simultaneous online dimensional detection of wax molds at multiple stations, the method comprising:

[0007] Inspection equipment groups are deployed at multiple wax model workstations, and digital communication between the inspection equipment groups and the data processing center is established. When the wax model arrives at the positioning space, the corresponding inspection equipment group is automatically activated to perform data acquisition and establish an acquisition dataset, which is equipped with a physical status label. After binding the acquisition dataset with the acquisition workstation and the wax model number, it is uploaded to the data processing center. When the data processing center receives all the data uploaded by the same wax model workstation, it performs independent dimensional modeling for each wax model workstation and establishes workstation dimensional modeling results. Historical datasets are called to perform dimensional evolution process modeling for multiple wax model workstations, and deformation mapping models between workstations are configured based on the modeling results. After mapping all the data uploaded by the wax model workstations to a unified standard wax injection domain using the deformation mapping model, image-point cloud reconstruction feature stitching is performed to establish a fusion modeling result. Dimensional anomaly verification is performed on the workstation dimensional modeling results and the fusion modeling results, and dimensional anomaly identifiers are established.

[0008] A second aspect of this application provides a multi-station synchronous online dimensional detection system for wax molds, the system comprising:

[0009] The system includes the following modules: a digital communication establishment module for deploying inspection equipment groups at multiple wax model workstations and establishing digital communication between these equipment groups and the data processing center; a data acquisition dataset establishment module for automatically activating the corresponding inspection equipment group to acquire data and establish a data acquisition dataset when the wax model arrives at the positioning space, with the dataset having physical status labels; a modeling result establishment module for binding the data acquisition dataset with the acquisition workstation and wax model number, then uploading it to the data processing center; and a deformation mapping model configuration module for calling historical datasets, performing dimensional evolution modeling at multiple wax model workstations, and configuring deformation mapping models between workstations based on the modeling results; a fusion modeling result establishment module for mapping all wax model workstation uploaded data to a unified standard wax injection domain using the deformation mapping model, then performing image-point cloud reconstruction feature stitching to establish a fusion modeling result; and a dimensional anomaly identification module for verifying dimensional anomalies in the workstation dimensional modeling results and the fusion modeling results, and establishing dimensional anomaly identifications.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] A detection equipment group is deployed at multiple wax model workstations, and digital communication is established between the detection equipment group and the data processing center. When the wax model arrives at the positioning space, a data acquisition dataset is created, with physical status labels. This dataset is then bound to the acquisition workstation and the wax model number, and uploaded to the data processing center. Independent dimensional modeling is performed at each wax model workstation, and workstation dimensional modeling results are established. Historical datasets are retrieved to model the dimensional evolution process at multiple wax model workstations, and deformation mapping models between workstations are configured based on the modeling results. Using the deformation mapping model, all uploaded data from the wax model workstations are mapped to a unified standard wax injection domain, and image-point cloud reconstruction feature stitching is performed to establish a fusion modeling result. Dimensional anomaly identification is also established. This achieves synchronous online detection and accurate modeling of wax model dimensions across multiple workstations, improving dimensional detection efficiency and anomaly identification accuracy. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating a multi-station synchronous online dimensional detection method for wax molds provided in this application embodiment;

[0014] Figure 2 This is a schematic diagram of a multi-station synchronous online dimension detection system for wax molds provided in an embodiment of this application.

[0015] Figure labeling: Digital communication establishment module 10, data acquisition and dataset establishment module 20, modeling result establishment module 30, deformation mapping model configuration module 40, fusion modeling result establishment module 50, and size anomaly identification establishment module 60. Detailed Implementation

[0016] This application provides a method and system for synchronous online dimensional detection of wax molds at multiple workstations, which addresses the technical problems of low efficiency in wax mold dimensional detection and insufficient accuracy in anomaly identification due to data fragmentation between workstations in the prior art.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides a method for simultaneous online dimensional detection of wax molds at multiple stations, the method comprising:

[0019] Step S100: Deploy inspection equipment groups at multiple wax mold workstations and establish digital communication between the inspection equipment groups and the data processing center.

[0020] Specifically, inspection equipment groups are deployed at multiple wax model stations, such as the post-wax injection station, the post-cooling station, and the pre-tree assembly station. These inspection equipment groups are used to collect data from the wax models at each station. At the same time, a digital communication connection is established between the inspection equipment groups and the data processing center to ensure that the data collected by the inspection equipment groups at each station can be transmitted to the data processing center stably and efficiently. This provides basic communication support for subsequent processes such as dataset establishment, dimensional modeling, and anomaly detection. It is the preliminary communication and equipment deployment preparation stage for realizing synchronous online dimensional inspection of wax models at multiple stations.

[0021] Step S200: When the wax model arrives at the positioning space, the corresponding detection equipment group is automatically started to perform data acquisition and establish a data acquisition dataset. The data acquisition dataset is equipped with physical status labels.

[0022] Specifically, when the wax model reaches the positioning space of the corresponding station among multiple wax model stations such as the post-wax injection station, the post-cooling station, and the pre-tree assembly station, the detection equipment group deployed at that station is automatically triggered to start data acquisition. By collecting point cloud, image and other data of the wax model, a data acquisition dataset is formed. This data acquisition dataset is set with physical status labels to identify the physical attribute status of the wax model at the current station. This provides the original data foundation with status labels for subsequent binding with the acquisition station and wax model number, as well as modeling and analysis in the data processing center.

[0023] Step S300: After binding the collected dataset with the collection station and wax model number, upload it to the data processing center. When the data processing center receives the uploaded data of all wax model stations for the same wax model, it performs independent dimension modeling for each wax model station and establishes the station dimension modeling results.

[0024] Specifically, the collected datasets from each wax model station, such as the post-wax injection station, the post-cooling station, and the pre-tree assembly station, are bound with physical state labels and linked to the corresponding collection station information and wax model number to form associated data before being uploaded to the data processing center. After the data processing center receives all the data uploaded from all wax model stations for the same wax model, it processes the uploaded data for each wax model station separately. First, it denoises the point cloud and image, identifies the edge contour and surface corner of the wax model to establish a structural contour point set, and then calculates the regional complexity by combining it with the calibration modeling data of the wax model and clusters it into regular and complex partitions. Subsequently, it maps the structural contour point set to the same coordinate system as the calibration modeling data, performs structural modeling based on the structural contour point set in the regular and complex partitions respectively, and merges the results to finally establish the station size modeling results for each station.

[0025] Step S400: Call the historical dataset, perform dimensional evolution process modeling under multiple wax mold stations, and configure the deformation mapping model between stations based on the modeling results.

[0026] Specifically, historical datasets, containing relevant data on past wax patterns at various workstations, are used to model the dimensional evolution process at multiple wax pattern workstations, including the post-wax injection workstation, the post-cooling workstation, and the pre-tree assembly workstation. During the modeling process, the historical datasets are first preprocessed to extract environmental and process parameters. Then, dynamic regression analysis is used to fit the dimensional deformation changes at multiple wax pattern workstations to establish the evolution law of size-process-environment. Finally, based on this evolution law, a deformation mapping model between workstations is configured to provide model support for mapping the data of each workstation to a unified standard wax injection domain.

[0027] Step S500: After mapping all the data uploaded from the wax model workstations to a unified standard wax injection domain using the deformation mapping model, perform image-point cloud reconstruction feature stitching to establish a fusion modeling result.

[0028] Specifically, the workstation corresponding to the unified standard wax injection area is first located and marked as the calibration workstation. The calibration workstation is then used as the target workstation to configure the mapping parameters of the deformation mapping model. Subsequently, these mapping parameters are used to map and transform the data uploaded from all wax model workstations, including the post-wax injection workstation, the post-cooling workstation, and the pre-tree assembly workstation. Data trust identifiers are configured for the transformed data, and the trust level of the data at the same location is verified by combining the data uploaded from the wax model workstations within the calibration workstation to establish a verification dataset. Finally, image-point cloud reconstruction feature stitching is performed based on the verification dataset to establish a fusion modeling result, thereby realizing the integration and modeling of data from different workstations under a unified standard.

[0029] Step S600: Perform dimensional anomaly verification on the workstation dimensional modeling results and the fused modeling results respectively, and establish dimensional anomaly identifiers.

[0030] Specifically, dimensional anomaly verification was performed on the independent dimensional modeling results and the integrated fusion modeling results for each wax model station, including the post-wax injection station, the post-cooling station, and the pre-tree assembly station. Before verification, non-critical inspection structures such as injection ports, wax discharge ports, and gating systems in the wax model structure needed to be located. By calling the structural annotation information in the theoretical CAD model of the wax model, the spatial coordinate range and geometric features of these non-critical structures were identified and defined as non-dimensional verification areas. In the subsequent verification process, only the dimensional comparison analysis of the modeling result data corresponding to the wax model part body was performed, and the modeling data of the non-dimensional verification areas was automatically excluded to avoid interference from structural data such as injection ports and wax discharge ports that have no inspection requirements, thus preventing them from interfering with the normal dimensional anomaly judgment. After completing the area exclusion, the modeling result of the part body is compared with the preset dimensional standard parameters to generate a dimensional anomaly label containing the dimensional deviation value and the location of the abnormal area. It is then determined whether the label meets the preset anomaly threshold. If the threshold is met, the dimensional anomaly label is bound to the corresponding wax model, and the wax model is transferred to the anomaly detection station. At the anomaly detection station, a second inspection is performed based on the dimensional anomaly label, and the corresponding wax model is managed according to the results of the second inspection. This completes the establishment of the dimensional anomaly label and enables accurate identification and subsequent processing of dimensional anomalies in the wax model.

[0031] In one possible implementation, step S300 further includes:

[0032] Step S310: After denoising the data uploaded from the wax model station by processing point cloud and image, identify the edge contour and surface corner of the wax model and establish a structural contour point set.

[0033] Step S320: Obtain the calibration modeling data of the wax model, calculate the regional complexity based on the calibration modeling data, perform clustering and partitioning using the regional complexity calculation results, and configure regular partitioning and complex partitioning.

[0034] Step S330: After mapping the structural contour point set to the same coordinate system of the calibration modeling data, structural modeling based on the structural contour point set is performed using the regular partitioning and the complex partitioning respectively. After fusing the structural modeling results, the workstation dimension modeling results are established.

[0035] Specifically, Gaussian filtering is used to denoise the point cloud and images in the data uploaded from the wax model station, removing noise interference to improve data quality. Then, the Canny operator is used to identify the edge contour of the wax model, and Harris corner detection is used to accurately locate the surface corners of the wax model. Finally, the coordinate information of the identified edge contour and surface corner is integrated to construct a structural contour point set that can fully reflect the geometric features of the wax model, providing accurate basic data support for subsequent structural modeling.

[0036] The calibration modeling data of the wax model is obtained, including benchmark information such as standard geometric parameters and structural features. When calculating the regional complexity, the number of contour line segments, the density of corner points, and the curvature change range of each region in the calibration modeling data are statistically analyzed. These parameters are then weighted and summed to obtain the regional complexity value of each region. Subsequently, the K-means clustering algorithm is used to perform cluster analysis on the complexity values ​​of all regions. Two cluster centers are set: regions with lower regional complexity values, i.e., fewer contour line segments, sparse corner points, and gentle curvature changes, are clustered into one class and configured as regular partitions; regions with higher regional complexity values, i.e., more contour line segments, dense corner points, and drastic curvature changes, are clustered into another class and configured as complex partitions. This achieves algorithm-based partitioning.

[0037] The structural contour point set, established by identifying edge contours and surface corners, is mapped to the same coordinate system as the wax model calibration modeling data, ensuring consistency in spatial location. Then, structural modeling is performed based on the structural contour point set for both regular and complex partitions. Within regular partitions, random contour point sets are collected under preset distance constraints. The spatial distribution of the random collection results is used to match the calibration geometric model template, and the first structural model is established by iteratively fitting and optimizing parameters. Within complex partitions, the regional complexity value is first obtained, and the fitting weights are dynamically configured. After initializing the complexity partition channel, differential modeling of the complexity interval is performed to generate the second structural model. Finally, the first and second structural models are merged to form the workstation dimension modeling result for the wax model station.

[0038] In one possible implementation, step S330 further includes:

[0039] Step S331: Collect random contour point sets under preset distance constraints within the rule partition. After matching and calibrating the geometric model template according to the spatial distribution of the random collection results, optimize the parameters through iterative fitting to establish the first structural model.

[0040] Step S332: Obtain the regional complexity value of the complex partition, and dynamically configure the fitting weights based on the regional complexity value.

[0041] Step S333: After initializing the complexity partition channels according to the dynamically configured fitting weights, perform differential modeling for each complexity interval, and generate the second structural model based on the differential modeling results.

[0042] Step S334: After fusing the first structural model and the second structural model, the structural modeling is completed.

[0043] Specifically, within a regular partition, according to a preset distance threshold, such as a constraint that the distance between adjacent points does not exceed 0.5mm, a set of contour points within the partition is randomly collected to ensure that the spatial distribution of the point set can cover the main structural features of the regular area. The spatial coordinate distribution of the collected random contour point set is compared and matched with the calibration geometric model template in the calibration modeling data, such as a standard cylinder, cuboid, or other regular geometric model, to determine the most suitable template type. Then, the least squares method is used for iterative fitting. By continuously adjusting the size, position, and other parameters of the template, the spatial deviation between the template and the random contour point set is minimized, thus completing parameter optimization. Finally, based on the optimized parameters, a first structural model that can accurately reflect the structure of the regular partition is established.

[0044] By statistically analyzing the number of contour lines, the distribution density of corner points, and the variation of surface curvature within complex partitions, the region complexity value is calculated according to a preset weight ratio. Subsequently, different intervals are divided based on the magnitude of the region complexity value, and a corresponding fitting weight value is assigned to each interval. The interval with a higher region complexity value corresponds to a larger fitting weight, and vice versa. This achieves dynamic configuration of fitting weights based on the region complexity value, enabling the fitting weights to adapt to partition structures with different levels of complexity.

[0045] The complexity partitioning channel is initialized based on dynamically configured fitting weights. This channel includes low-complexity, medium-complexity, and high-complexity sub-channels, and the learning intensity is configured for each sub-channel according to the fitting weights. Then, the low-complexity, medium-complexity, and high-complexity sub-channels receive the mapped structural contour point sets, respectively. The low-complexity sub-channel performs regular model residual modeling on these point sets to generate the first modeling result, the medium-complexity sub-channel performs multi-scale adjacency modeling to generate the second modeling result, and the high-complexity sub-channel performs modeling through a deep neural network to generate the third modeling result. Finally, the three modeling results are integrated to generate the second structural model.

[0046] A weighted fusion algorithm is used to fuse the first and second structural models. Based on the importance of regular and complex partitions in the overall structure of the wax model and the modeling accuracy, corresponding fusion weights are assigned to the two models. Coordinate alignment technology is used to ensure that the two models match in the same spatial coordinate system. Then, the structural feature data of the two models are integrated according to their weights. For feature information in overlapping areas, data with higher confidence are selected and retained. Finally, the fused model is smoothed to eliminate feature abrupt changes at the model splicing points, forming a complete and coherent structural model, thereby completing the structural modeling.

[0047] In one possible implementation, step S333 further includes:

[0048] Step S3331: After configuring the learning intensity of the low-complexity sub-channel, medium-complexity sub-channel, and high-complexity sub-channel in the complexity partition channel using the fitting weights, the mapped structural contour point set is received using the low-complexity sub-channel, medium-complexity sub-channel, and high-complexity sub-channel, respectively.

[0049] Step S3332: Use low-complexity sub-channels to perform residual modeling of the rule model and generate the first modeling result.

[0050] Step S3333: Perform multi-scale adjacency modeling using medium-complexity sub-channels to generate the second modeling result.

[0051] Step S3334: Use high-complexity sub-channels to model deep neural networks and generate a third modeling result.

[0052] Step S3335: Generate a second structural model based on the first modeling result, the second modeling result, and the third modeling result.

[0053] Specifically, dynamically configured fitting weights are allocated to low-complexity, medium-complexity, and high-complexity sub-channels within the complexity partition channel according to a preset ratio. For example, the high-complexity sub-channel is allocated 60% weight to set high-intensity learning parameters, such as an iteration step size of 0.01 and a feature extraction dimension of 512. The medium-complexity sub-channel is allocated 30% weight to set medium-intensity parameters, with an iteration step size of 0.05 and a feature extraction dimension of 256. The low-complexity sub-channel is allocated 10% weight to set low-intensity parameters, with an iteration step size of 0.1 and a feature extraction dimension of 128. After configuring the learning intensity, the structural contour point set is transformed to the calibration modeling data coordinate system using a coordinate mapping algorithm. Then, based on the complexity label of the region to which each point belongs, the point set data is transmitted to the corresponding sub-channel buffer, enabling the sub-channel to accurately receive the mapped point set.

[0054] The low-complexity subchannel first loads a pre-defined standard geometric template library, containing basic regular model parameters such as planes, cylinders, and spheres. It then uses Hough transform to identify basic geometric features from the received structural contour point set and matches the optimal template. The least squares method is used to calculate the residual vector between the point set and the matched template, and a residual threshold is set, such as 0.02mm, to filter outliers. The template parameters are then optimized using the Levenberg-Marquardt iterative algorithm to minimize the sum of squared residuals. Finally, the optimized template parameters are integrated with the residual distribution data to generate the first modeling result, which includes basic geometric parameters and local deviation corrections. The deviation corrections are visualized in the form of a color cloud map.

[0055] The medium-complexity sub-channel, for the received set of structural contour points in a medium-complex region, first sets multiple scales, such as neighborhood radii of 1mm, 3mm, and 5mm. At each scale, the K-Dtree algorithm is used to quickly retrieve the set of adjacent points for each point. Local features are extracted from the adjacent point sets at different scales, and feature parameters such as the distance distribution between adjacent points, the angle between normal vectors, and the rate of curvature change are calculated. Then, a feature fusion network is used to perform correlation analysis on the multi-scale features to construct a cross-scale structural correlation model, capturing the structural features in the medium-complex region that contain both certain regularities and local variations. Finally, the modeling results at each scale are integrated to generate a second modeling result that can fully reflect the multi-scale structural features of the medium-complex region.

[0056] The high-complexity sub-channel targets the received high-complexity regional structural contour point set. First, the data undergoes standardization preprocessing, mapping the point set coordinates to the [-1, 1] interval and supplementing feature dimensions such as normal vectors and curvature. Then, it is input into a deep neural network composed of multiple stacked PointNet++ modules. Through hierarchical sampling and feature aggregation operations, it progressively extracts multi-scale features of complex structures from local to global perspectives. Each module includes a sampling layer, employing the FPS algorithm to select key sampling points, grouping layers to construct spherical neighborhoods, and a feature extraction layer. Local features are learned through an MLP network. The network output incorporates an attention mechanism to strengthen the weights of key features, ultimately outputting a high-dimensional structural feature vector through a fully connected layer. This feature vector is then back-projected into three-dimensional space to generate a third modeling result containing fine structural details and complex surface features. Its accuracy is verified using a preset IoU (Intersection over Union) threshold, such as ≥0.95.

[0057] First, the spatial coordinates of the first, second, and third modeling results are aligned to ensure their positions match within the same coordinate system. Then, a feature fusion algorithm is used to integrate the residual results of the regular model in low-complexity regions, the multi-scale adjacency features in medium-complexity regions, and the output features of the deep neural network in high-complexity regions according to regional weights. The weight allocation is consistent with the fitting weights of the corresponding sub-channels. For feature information in overlapping regions, data with higher accuracy is selected and retained through confidence assessment, and Gaussian filtering is used for smooth transition processing. Finally, the integrated feature data is converted into a three-dimensional mesh model. After topology optimization and redundant vertex removal, a second structural model that can completely and coherently reflect the overall structure of the complex partition is generated.

[0058] In one possible implementation, step S400 further includes:

[0059] Step S410: After preprocessing the historical dataset, extract the environmental and process parameters, and then fit the dimensional deformation changes of multiple wax mold stations through dynamic regression analysis to establish the evolution law of size-process-environment.

[0060] Step S420: Configure the deformation mapping model according to the evolution law.

[0061] Specifically, the Pandas library was used to clean the historical dataset, outliers were identified and removed using box plots, missing data were filled using multiple imputation, and the data were mapped to a unified interval using Min-Max standardization. The Scikit-learn toolkit was used to extract the time-domain features of environmental parameters temperature, humidity, and air pressure (mean, variance, and trend), and the process features of process parameters wax injection pressure, holding time, and wax temperature (peak value, duration, and rate of change). Key features were selected using recursive feature elimination. A dynamic regression model was built based on the statsmodels library, incorporating time lag terms and a sliding window mechanism to fit the dimensional deformation sequences of multiple wax mold stations piecewise. The least squares method was used to estimate the model parameters, and the lag order was optimized using the AIC criterion. Finally, the fitting results of each station were integrated, and the influence coefficients between parameters were visualized using a heatmap, forming a quantified dimensional-process-environment evolution pattern.

[0062] Based on the evolutionary laws of size, process, and environment, a deformation mapping model architecture is constructed with an input layer containing environmental parameters (temperature, humidity, air pressure) and process parameters (wax injection pressure, holding time, etc.) and an output layer containing the deformation variables of the wax mold at each station. The quantified parameter relationships in the evolutionary laws are transformed into the initial weight matrix of the model. The model is trained and optimized through a backpropagation algorithm to keep the deviation between the output size deformation variables and the actual values ​​within a reasonable range. At the same time, a dynamic adjustment mechanism is embedded in the model. When the input parameters exceed the range of historical data, the trend extrapolation logic based on the evolutionary laws is used for adaptive correction. Finally, a model that can accurately map the size deformation of the wax mold according to the real-time environment and process parameters is configured.

[0063] In one possible implementation, step S500 further includes:

[0064] Step S510: After locating the station corresponding to the standard wax injection area, mark it as the calibration station.

[0065] Step S520: Use the calibration station as the target station and configure the mapping parameters of the deformation mapping model.

[0066] Step S530: Perform data mapping transformation on the data uploaded from the wax model station according to the mapping parameters, and use the data mapping transformation result to perform image-point cloud reconstruction feature stitching.

[0067] Specifically, the entire production line's workstation layout is scanned using a laser positioning system. Combined with preset standard wax injection domain 3D model parameters, such as wax injection port location, cavity contour boundary, and workstation coordinate range, the workstation with the highest degree of conformity to the standard wax injection domain parameters is matched in the scanning results. This workstation is then physically marked, such as by affixing a calibration label and recording its 3D coordinate information, equipment number, and corresponding wax injection process standard in the system. It is then officially marked as a calibration workstation, serving as the benchmark reference workstation for all subsequent dimensional deformation mapping and model calibration.

[0068] Using the calibrated workstation as the target workstation, the environmental parameters, process parameters, and corresponding dimensional deformation records from its historical production data are retrieved and input into the deformation mapping model for initial training. Based on the standard wax injection domain parameters of the calibrated workstation, such as cavity dimensional tolerances and wax injection paths, the input weights and feature mapping coefficients of the model are adjusted to prioritize the response of parameters strongly correlated with the characteristics of that workstation. By comparing the difference between the predicted deformation output by the model and the actual deformation, the deviation correction factor and dynamic response threshold in the mapping parameters are iteratively optimized. Finally, the parameter configuration of the deformation mapping model adapted to the calibrated workstation is completed, ensuring that the model's dimensional deformation mapping accuracy for that workstation meets the preset standard.

[0069] Based on the configured mapping parameters, the image data and point cloud data uploaded from each wax model station are mapped and transformed to a unified standard wax injection domain coordinate system, completing the spatial alignment transformation of the data. Subsequently, data trust identifiers are configured for the transformed image and point cloud data, which are generated based on factors such as the accuracy of the data acquisition equipment and the stability of transmission. Using these data trust identifiers, trust verification is performed in conjunction with data from the same location uploaded from the calibration station. By comparing the consistency of data from different sources at the same location, data that meets the trust threshold is selected to construct a verification dataset. Finally, based on the verification dataset, features such as edges and textures in the images and features such as spatial coordinates and normal vectors in the point clouds are extracted. The reconstructed features of the images and point clouds are stitched together using a feature matching algorithm to form complete and reliable 3D reconstruction feature data of the wax model.

[0070] In one possible implementation, step S530 further includes:

[0071] Step S531: Configure a data trust identifier for the data mapping and transformation result.

[0072] Step S532: Use the data trust identifier and the data uploaded from the wax model station in the calibration station to verify the trust level of the data at the same location and establish a verification dataset.

[0073] Step S533: Perform image-point cloud reconstruction feature stitching based on the verification dataset.

[0074] Specifically, based on the image data and point cloud data obtained after data mapping and transformation, the built-in trust identifier configuration module of the data processing center is invoked. This module determines whether there is packet loss or tampering and the initial deviation value between the transformed data and the theoretical coordinates of the standard wax injection area, according to the accuracy level of the corresponding detection equipment group, such as the ranging error of the laser scanner, the pixel resolution of the industrial camera, and the CRC check result during data transmission. It calculates a trust score of 0-100 through a preset weighted algorithm and binds the score as the data trust identifier to the corresponding transformation result. The higher the score, the stronger the data reliability, providing a quantitative basis for subsequent trust verification.

[0075] Based on the configured data trust identifier, the benchmark data at the same location as the data uploaded by each wax model station within the calibration station is retrieved, including the standard dimensional parameters and historical reliable detection values ​​at that location. The image and point cloud data at the same location in the data mapping and transformation results are compared with the benchmark data of the calibration station, and the deviation value between the two is calculated. Combined with the quantitative score of the data trust identifier, the data with deviation values ​​within the preset threshold and whose trust identifiers meet the reliability requirements are filtered. The verified data are integrated into a verification dataset to ensure that the dataset can reflect the actual detection situation of each station and is consistent with the benchmark data of the calibration station, providing a reliable data foundation for subsequent image-point cloud reconstruction feature stitching.

[0076] Based on the validation dataset, the edge contours, surface texture features, and feature point coordinates of the wax model in the image data are extracted. At the same time, the three-dimensional spatial coordinates, normal vectors, and curvature features of the point cloud data are extracted. Through feature matching algorithms, the two-dimensional feature points in the image are associated with the corresponding three-dimensional feature points in the point cloud to establish a spatial mapping relationship between the two. The image features and point cloud features are then fused and stitched together using this mapping relationship, so that the detailed information of the image and the spatial structure information of the point cloud complement each other. Finally, a reconstructed feature stitching result that can completely and accurately reflect the three-dimensional structure of the wax model is formed, providing unified and accurate feature data for subsequent fusion modeling.

[0077] In one possible implementation, step S600 further includes:

[0078] Step S610: Determine whether the size anomaly identifier meets the anomaly threshold.

[0079] Step S620: If the size anomaly identifier meets the anomaly threshold, then after binding the size anomaly identifier with the corresponding wax model, the wax model is transferred to the anomaly detection station.

[0080] Step S630: Perform a second inspection based on the dimensional anomaly marker at the anomaly detection station, and manage the corresponding wax mold according to the results of the second inspection.

[0081] Specifically, the system retrieves dimensional anomaly identifiers generated from the workstation dimensional modeling results and fusion modeling results from the data processing center. These identifiers include dimensional deviation values ​​for each part of the wax model, coordinates of the abnormal area, and feature descriptions. A preset anomaly threshold parameter library is then called, storing key dimensional tolerance ranges for different wax model types, such as allowable deviations for cavity diameter and straightness tolerances for contour lines, as well as thresholds for the number of abnormal features and severity grading standards for individual anomalies. A numerical comparison algorithm quantifies and compares the deviation values ​​in the dimensional anomaly identifiers with the corresponding tolerance ranges, counts whether the number of abnormal features exceeds the threshold, and determines the level of a single anomaly based on the severity grading standards. If any indicator meets or exceeds the preset threshold, the dimensional anomaly identifier is determined to meet the anomaly threshold, triggering subsequent anomaly handling procedures.

[0082] When a dimensional anomaly marker is determined to meet the anomaly threshold, the data processing center will associate the anomaly location, dimensional deviation value, and anomaly characteristics contained in the marker with the unique number of the corresponding wax model, forming associated data containing wax model identification information and anomaly details. At the same time, a transfer instruction is sent to the production line's conveyor control module. The real-time position of the wax model is located through the workstation sensor, and the steering mechanism and pushing device of the conveyor track are controlled to divert the wax model with the anomaly marker from the normal production flow and accurately deliver it to the preset anomaly detection workstation for further secondary inspection.

[0083] At the anomaly detection station, the detection equipment, based on the dimensional anomaly markers bound to the wax model, performs secondary precision inspections on the anomaly locations recorded in the markers, such as cavity corners and wax injection port edges, and the anomaly types, such as dimensional deviations and contour deformations. This includes using a high-precision laser diameter gauge to verify key dimensions, capturing detailed images of the anomaly area using a high-resolution industrial camera, and acquiring fine point cloud data of the anomaly area using a 3D scanner. The secondary inspection results are compared and analyzed with the initial inspection data in the dimensional anomaly markers. If the anomaly is confirmed to meet the preset acceptable standards, such as if the deviation is within the repairable range, the wax model is marked as "to be repaired" and transferred to the repair station. If the re-inspection results show that the anomaly exceeds the allowable range and cannot be repaired, it is marked as "unqualified" and a rejection mechanism is triggered. If the re-inspection finds that the initial anomaly marker is a misjudgment, the anomaly marker is removed and the wax model is returned to the normal production process. At the same time, the anomaly detection parameters in the system are updated to optimize the accuracy of subsequent judgments, completing the classification and management of the corresponding wax models.

[0084] In one possible implementation, step S100 further includes:

[0085] Step S110: The multiple wax model stations include a post-wax injection station, a post-cooling station, and a pre-tree assembly station.

[0086] Specifically, the multiple wax model stations cover the stations corresponding to three key stages in the wax model production process. The post-wax injection station is the station where the wax model has just been waxed and formed but has not yet undergone subsequent processing, and is used to detect the dimensional characteristics of the wax model in its initial state. The post-cooling station is the station where the wax model has cooled and solidified, and can detect the dimensional state of the wax model after cooling under the influence of temperature changes. The pre-tree-building station is the station where the wax model is about to enter the tree-building process, and is used to detect the final dimensional state of the wax model after completing the pre-processing and before the tree-building operation. By setting up these three key stations, comprehensive detection of the dimensions at different stages of wax model production can be achieved.

[0087] Example 2, based on the same inventive concept as the multi-station synchronous online dimensional detection method for wax molds in the foregoing examples, such as... Figure 2 As shown, this application provides a multi-station synchronous online dimensional detection system for wax molds. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0088] The digital communication establishment module 10 is used to deploy inspection equipment groups at multiple wax mold workstations and establish digital communication between the inspection equipment groups and the data processing center.

[0089] The dataset creation module 20 is used to automatically start the corresponding detection equipment group to perform data acquisition and create a dataset when the wax model arrives at the positioning space. The dataset is equipped with physical status labels.

[0090] The modeling result establishment module 30 is used to bind the collected dataset with the collection station and wax model number, and then upload it to the data processing center. When the data processing center receives the uploaded data of all wax model stations for the same wax model, it performs independent dimension modeling for each wax model station and establishes the station dimension modeling result.

[0091] The deformation mapping model configuration module 40 is used to call historical datasets, perform dimensional evolution process modeling under multiple wax mold stations, and configure the deformation mapping model between stations based on the modeling results.

[0092] The fusion modeling result establishment module 50 is used to map the data uploaded from all wax mold stations to a unified standard wax injection domain using the deformation mapping model, and then perform image-point cloud reconstruction feature stitching to establish the fusion modeling result.

[0093] The dimension anomaly identification module 60 is used to perform dimension anomaly verification on the workstation dimension modeling results and the fusion modeling results respectively, and to establish dimension anomaly identification.

[0094] Furthermore, the system is also used to implement the following functions:

[0095] After denoising the data uploaded from the wax model station by using point cloud and image processing, the edge contours and surface corners of the wax model are identified, and a structural contour point set is established. The calibration modeling data of the wax model is obtained, and the regional complexity is calculated based on the calibration modeling data. The results of the regional complexity calculation are used for clustering and partitioning, configuring regular partitions and complex partitions. After mapping the structural contour point set to the same coordinate system as the calibration modeling data, structural modeling based on the structural contour point set is performed using the regular partitions and the complex partitions respectively. After fusing the structural modeling results, the station dimension modeling results are established.

[0096] Furthermore, the system is also used to implement the following functions:

[0097] Within the defined rule partition, random contour point sets under preset distance constraints are collected. After matching and calibrating the geometric model template based on the spatial distribution of the random collection results, the parameters are optimized through iterative fitting to establish a first structural model. The regional complexity value of the complex partition is obtained, and the fitting weight is dynamically configured based on the regional complexity value. After initializing the complexity partition channel according to the dynamically configured fitting weight, differential modeling of the complexity interval is performed respectively, and a second structural model is generated based on the differential modeling results. After fusing the first structural model and the second structural model, the structural modeling is completed.

[0098] Furthermore, the system is also used to implement the following functions:

[0099] After configuring the learning intensities of the low-complexity, medium-complexity, and high-complexity sub-channels in the complexity partition channel using the fitted weights, the mapped structural contour point sets are received using the low-complexity, medium-complexity, and high-complexity sub-channels respectively; regular model residual modeling is performed using the low-complexity sub-channel to generate a first modeling result; multi-scale adjacency modeling is performed using the medium-complexity sub-channel to generate a second modeling result; deep neural network modeling is performed using the high-complexity sub-channel to generate a third modeling result; and a second structural model is generated based on the first, second, and third modeling results.

[0100] Furthermore, the system is also used to implement the following functions:

[0101] After preprocessing the historical dataset, environmental and process parameters are extracted, and dynamic regression analysis is used to fit the dimensional deformation changes of multiple wax mold stations to establish the evolution law of size-process-environment; a deformation mapping model is configured according to the evolution law.

[0102] Furthermore, the system is also used to implement the following functions:

[0103] After locating the workstation corresponding to the standard wax injection area, it is marked as the calibration workstation; the calibration workstation is used as the target workstation, and the mapping parameters of the deformation mapping model are configured; the data uploaded by the wax model workstation is converted according to the mapping parameters, and the image-point cloud reconstruction feature stitching is performed using the data conversion result.

[0104] Furthermore, the system is also used to implement the following functions:

[0105] Configure a data trust identifier for the data mapping and transformation result; use the data trust identifier and the data uploaded from the wax model station in the calibration station to perform data trust verification at the same location and establish a verification dataset; perform image-point cloud reconstruction feature stitching based on the verification dataset.

[0106] Furthermore, the system is also used to implement the following functions:

[0107] Determine whether the size anomaly identifier meets the anomaly threshold; if the size anomaly identifier meets the anomaly threshold, bind the size anomaly identifier to the corresponding wax model and transfer the wax model to the anomaly detection station; perform a second re-inspection based on the size anomaly identifier at the anomaly detection station, and manage the corresponding wax model according to the results of the second re-inspection.

[0108] Furthermore, the system is also used to implement the following functions:

[0109] The multiple wax model stations include a post-wax injection station, a post-cooling station, and a pre-tree assembly station.

[0110] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0111] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0112] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for simultaneous online dimensional detection of wax molds at multiple stations, characterized in that, The method includes: Deploy testing equipment groups at multiple wax mold workstations and establish digital communication between the testing equipment groups and the data processing center; Once the wax model reaches the positioning space, the corresponding detection equipment group is automatically activated to perform data acquisition and establish a data acquisition dataset, which is equipped with physical status labels. After the collected dataset is bound to the collection station and wax model number, it is uploaded to the data processing center. When the data processing center receives the uploaded data of all wax model stations for the same wax model, it performs independent dimension modeling for each wax model station and establishes the station dimension modeling results. Call historical datasets to perform dimensional evolution process modeling at multiple wax mold stations, and configure deformation mapping models between stations based on the modeling results; After mapping all the data uploaded from the wax model workstations to a unified standard wax injection domain using the deformation mapping model, image-point cloud reconstruction feature stitching is performed to establish a fusion modeling result; Perform dimensional anomaly verification on the workstation dimensional modeling results and the fused modeling results respectively, and establish dimensional anomaly identifiers; The process of calling historical datasets, performing dimensional evolution modeling at multiple wax mold stations, and configuring deformation mapping models between stations based on the modeling results includes: After preprocessing the historical dataset, environmental and process parameters are extracted, and dynamic regression analysis is used to fit the dimensional deformation changes of multiple wax mold stations to establish the evolution law of size-process-environment. Configure the deformation mapping model according to the aforementioned evolutionary laws; After mapping all wax model station uploaded data to a unified standard wax injection domain using the deformation mapping model, image-point cloud reconstruction feature stitching is performed to establish a fusion modeling result, including: After locating the standard wax injection area, mark the corresponding workstation as the calibration workstation; The calibration station is used as the target station, and the mapping parameters of the deformation mapping model are configured. The data mapping conversion of the wax model station uploaded data is performed according to the mapping parameters, and the image-point cloud reconstruction feature stitching is performed using the data mapping conversion result.

2. The method for simultaneous online dimensional detection of wax molds at multiple stations as described in claim 1, characterized in that, When the data processing center receives the uploaded data from all wax model workstations for the same wax model, it performs independent dimensional modeling for each wax model workstation, establishing workstation dimensional modeling results, including: After denoising the data uploaded from the wax model station by using point cloud and image processing, the edge contour and surface corner of the wax model are identified, and a structural contour point set is established. Obtain the calibration modeling data of the wax model, calculate the regional complexity based on the calibration modeling data, perform clustering and partitioning using the regional complexity calculation results, and configure regular partitioning and complex partitioning; After mapping the structural contour point set to the same coordinate system as the calibration modeling data, structural modeling based on the structural contour point set is performed using the regular partitioning and the complex partitioning respectively. After fusing the structural modeling results, the workstation dimension modeling results are established.

3. The method for simultaneous online dimensional detection of wax molds at multiple stations as described in claim 2, characterized in that, The structural modeling based on structural contour point sets using the regular partitioning and the complex partitioning respectively includes: Within the rule partition, a random contour point set under a preset distance constraint is collected. After matching and calibrating the geometric model template according to the spatial distribution of the random collection results, the parameters are optimized through iterative fitting to establish the first structural model. Obtain the regional complexity value of the complex partition, and dynamically configure the fitting weights based on the regional complexity value; After initializing the complexity partition channels according to the dynamically configured fitting weights, differential modeling of the complexity intervals is performed respectively, and a second structural model is generated based on the differential modeling results. After fusing the first structural model and the second structural model, structural modeling is completed.

4. The method for simultaneous online dimensional detection of wax molds at multiple stations as described in claim 3, characterized in that, After initializing the complexity partitioning channels according to the dynamically configured fitting weights, the differential modeling of the complexity intervals is performed respectively, including: After configuring the learning intensity of the low-complexity sub-channel, medium-complexity sub-channel, and high-complexity sub-channel in the complexity partition channel using the fitting weights, the mapped structural contour point sets are received using the low-complexity sub-channel, medium-complexity sub-channel, and high-complexity sub-channel, respectively. The first modeling result is generated by using low-complexity sub-channels to perform residual modeling of the rule model. Multi-scale adjacency modeling is performed using medium-complexity subchannels to generate a second modeling result; High-complexity sub-channels are used to model deep neural networks, generating a third modeling result; A second structural model is generated based on the first modeling result, the second modeling result, and the third modeling result.

5. The method for simultaneous online dimensional detection of wax molds at multiple stations as described in claim 1, characterized in that, The step of performing image-point cloud reconstruction feature stitching using the data mapping transformation results includes: Configure a data trust identifier for the data mapping and transformation result; The trustworthiness of data at the same location is verified by using the data trust identifier and the data uploaded from the wax model station in the calibration station, and a verification dataset is established. Perform image-point cloud reconstruction feature stitching based on the validation dataset.

6. The method for simultaneous online dimensional detection of wax molds at multiple stations as described in claim 1, characterized in that, The establishment of the size anomaly identifier includes: Determine whether the size anomaly indicator meets the anomaly threshold; If the size anomaly identifier meets the anomaly threshold, then the size anomaly identifier is bound to the corresponding wax model, and the wax model is transferred to the anomaly detection station. A second inspection based on the dimensional anomaly marker is performed at the anomaly detection station, and the corresponding wax mold is managed according to the results of the second inspection.

7. The method for simultaneous online dimensional detection of wax molds at multiple stations as described in claim 1, characterized in that, The multiple wax model stations include a post-wax injection station, a post-cooling station, and a pre-tree assembly station.

8. A multi-station synchronous online dimensional detection system for wax molds, characterized in that, The system is used to implement the online dimensional detection method for multi-station synchronous dimensions of wax molds according to any one of claims 1-7, and the system includes: The digital communication establishment module is used to deploy inspection equipment groups at multiple wax mold workstations and establish digital communication between the inspection equipment groups and the data processing center; The data acquisition dataset creation module is used to automatically start the corresponding detection equipment group to perform data acquisition and create a data acquisition dataset when the wax model arrives at the positioning space. The data acquisition dataset is equipped with physical status labels. The modeling result establishment module is used to bind the collected dataset with the collection station and wax model number and upload it to the data processing center. When the data processing center receives the uploaded data of all wax model stations for the same wax model, it performs independent dimension modeling for each wax model station and establishes the station dimension modeling result. The deformation mapping model configuration module is used to call historical datasets, perform dimensional evolution process modeling under multiple wax mold stations, and configure the deformation mapping model between stations based on the modeling results; The fusion modeling result establishment module is used to map the data uploaded from all wax mold stations to a unified standard wax injection domain using the deformation mapping model, and then perform image-point cloud reconstruction feature stitching to establish the fusion modeling result. The dimension anomaly identification module is used to perform dimension anomaly verification on the workstation dimension modeling results and the fused modeling results respectively, and to establish dimension anomaly identification.

Images