A non-contact on-line measuring method for profile size deviation of non-ferrous castings

Abstract

The present application relates to the technical field of optical size measurement of castings, and discloses a non-ferrous metal casting profile size deviation online measurement method; the method first acquires the CAD theoretical model corresponding to the current casting variety, the gating system design, the mold cavity partition atlas and the process parameters, constructs a process prior information library and determines the prior area of interference features; collects the casting surface measurement point cloud and multi-view images, establishes a coarse registration benchmark based on the high rigidity features of the body, discriminates the geometric shape differences of the coarse registered point cloud, checks the prior position attribution and checks the multi-view consistency, divides the body feature points and the interference feature points, configures registration weights according to the point types and performs weighted fine registration, and calculates the normal signed distance of the body feature points relative to the CAD theoretical surface after the self-checking is passed; in the above manner, the influence of non-design profile residues on the registration benchmark can be reduced, and stable online measurement of the real manufacturing deviation of the casting can be realized.

Description

Technical Field

[0001] This invention relates to the field of optical dimension measurement technology for castings, specifically to an online method for measuring the dimensional deviation of non-ferrous metal castings. Background Technology

[0002] Non-ferrous metal castings are widely used in automotive parts, rail transportation, aerospace, and motor housings. Deviations in their dimensional profiles can affect subsequent machining allowances, assembly accuracy, sealing performance, and overall machine reliability. To detect dimensional anomalies before machining, existing production lines typically acquire 3D surface data of the castings using line laser scanning, structured light scanning, or multi-angle image reconstruction after demolding, cooling, shot blasting, or sandblasting. The measured point cloud is then registered with a CAD theoretical model, and the distance deviation after registration is used to determine if the casting is out of tolerance. For example, Chinese invention patent application CN107170035B discloses… A method and system for scanning highly reflective objects are disclosed, which acquires geometric data of objects by setting reference objects, taking pictures from multiple angles, and spraying matte coatings or projecting light spots. Another example is the authorized Chinese invention patent application CN107869955B, which discloses a laser 3D scanner system and its usage method, which acquires point clouds of object surfaces using lasers and determines the spatial position of the scanner by combining visual image markers. While these technologies can improve the 3D data acquisition or scanning positioning process, they primarily focus on reflection suppression, point cloud acquisition, and scanning positioning, without addressing the impact of non-designed contour residues in casting production lines on the point cloud registration reference.

[0003] In actual production, the flash on the parting surface, residual protrusions at the gate or riser cut-off points, burrs on the parting line, and molding sand residues adhering to the cavities or deep rib roots are not the design contours defined by the CAD theoretical model. However, they are usually in continuous contact with the surface of the casting body, and the amount of protrusion may be close to the allowable manufacturing deviation of the casting. If ordinary overall registration treats the above-mentioned interference points and body points as valid contour points for matching, it will cause the rotation matrix and translation vector to be pulled off by the interference points. This will cause the deviation results to be mixed with the body manufacturing deviation, the registration datum drift, and the amount of interference residue protrusion. This will result in inconsistent repeated measurement results, unstable batch out-of-tolerance judgment, and weaken the indicative role of the deviation chromatogram in indicating the mold wear location and process drift direction. Therefore, it is necessary to provide an online measurement method that can identify and suppress the influence of interference feature points such as flash, residual protrusions, burrs, and molding sand residues during the registration process. This will allow the registration results to be mainly determined by the true contour of the casting body, thereby separating the true manufacturing deviation from the interference protrusion and improving the reliability of online out-of-tolerance judgment and mold maintenance verification. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an online method for measuring the dimensional deviation of non-ferrous metal castings, thus solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for online measurement of dimensional deviations in the profile of non-ferrous metal castings, comprising:

[0007] Obtain the CAD theoretical model, gating system design, mold cavity partition map and process parameters corresponding to the current casting type, construct a process prior information database, and determine the prior regions of interference features based on the process prior information database;

[0008] Collect measurement point clouds and multi-view images of the casting surface, perform point cloud fusion on the measurement point clouds of the casting surface, and establish a coarse registration benchmark based on the high stiffness registration features of the body and the corresponding features in the CAD theoretical model.

[0009] After coarse registration, geometric morphological differences, prior location attribution checks, and multi-view consistency verifications are performed on the point cloud to divide the point cloud points into ontological feature points and interference feature points.

[0010] Registration weights are configured based on the ontological feature points and the interference feature points, and the fused point cloud and the CAD theoretical model are then finely registered using weighted averages based on the configured registration weights.

[0011] After the weighted fine registration passes self-verification, the signed normal distance of the feature points relative to the CAD theoretical surface is calculated only for the feature points of the body, and the actual manufacturing deviation, deviation chromatogram and interference point removal record are generated.

[0012] Furthermore, a process prior information database is constructed, including:

[0013] Perform consistency checks on the CAD theoretical model version, mold cavity version, and casting type;

[0014] After verification, the CAD theoretical model, gating system design documents, mold cavity partition diagram and process parameters are analyzed to form a process prior information database, which includes body surface data, body registration feature data, parting surface trajectory data, gating and riser location data, mold parting line trajectory data, molding sand residue candidate area data and tolerance reference data.

[0015] Furthermore, the prior regions of interference features are determined, including:

[0016] Generate flash prior regions along both sides of the parting line or in the direction of the outer normal;

[0017] A residual raised prior region is generated on the outside of the gate or riser section;

[0018] Generate a priori areas for seam cut along both sides of the mold parting line;

[0019] The preliminary area for generating molding sand residue is generated based on the cavity depth, the radius of the reinforcing bar root, the position of the sand core positioning surface, the sandblasting path, and the cleaning process record.

[0020] Furthermore, point cloud data and multi-view images of the casting surface are acquired, including:

[0021] In response to the casting arrival signal and the type identifier, the type identifier is matched with the information in the CAD theoretical model;

[0022] After successful matching, read the furnace process file and bind the furnace process file to the measurement point cloud of the casting surface.

[0023] Once the surface temperature of the casting enters the allowable measurement range and the rate of temperature change is lower than the on-site configuration conditions, the clamping turntable is controlled to fix the casting, and the surface measurement point cloud and multi-view images of the casting are acquired.

[0024] Establish a mapping relationship between image timestamps and point cloud timestamps.

[0025] Further, point cloud fusion is performed on the measured point cloud of the casting surface, including:

[0026] Read the turntable angle value, scan the device posture and calibration matrix corresponding to each timestamp, and convert the point cloud collected from different perspectives to a unified casting measurement coordinate system;

[0027] Perform continuity checks on the common overlapping areas between adjacent viewpoints;

[0028] When the deviation in the public area exceeds the allowable range of the equipment calibration, the control turntable returns to the corresponding angle range for supplementary scanning;

[0029] If the continuity requirement is still not met after rescanning, the corresponding area is marked as locally missing, and the missing area information is recorded.

[0030] Furthermore, a coarse registration benchmark is established, including:

[0031] Extract the high-stiffness registration features of the body corresponding to the body registration feature data from the fused point cloud. The high-stiffness registration features of the body include at least one of the following: design reference surface, locating pin hole, bolt hole array or assembly stop.

[0032] The high-rigidity registration features of the body are matched with the corresponding theoretical features in the CAD theoretical model, and the rigid body transformation from the fused point cloud coordinate system to the CAD model coordinate system is solved to obtain the coarse registration matrix.

[0033] Perform coarse registration self-verification after coarse registration is completed;

[0034] When the body high-stiffness registration features are insufficient or the alignment error exceeds the allowable range of the equipment calibration, perform at least one of the following processes: re-extract the body high-stiffness registration features, degrade coarse registration, rescan key viewpoints, or transfer to manual confirmation.

[0035] Furthermore, the coarsely registered point cloud is subjected to geometric morphological difference discrimination, prior location assignment check, and multi-view consistency verification, including:

[0036] After coarse registration, obtain the normal deviation, neighborhood connectivity, neighborhood thickness, local curvature change, local normal direction difference, and neighborhood point density of point cloud points or local point clusters relative to the CAD theoretical surface, and screen candidate interference points.

[0037] Spatial assignment checks are performed on the candidate point coordinates and the prior areas of the flash, residual protrusions, seams, and molding sand fragments.

[0038] Candidate points are back-projected onto image planes from multiple perspectives, and multi-view consistency is checked based on the visibility, edge continuity, and color consistency of local image regions.

[0039] Furthermore, the point cloud points are divided into ontological feature points and interference feature points, including:

[0040] A status identifier is generated for the geometric shape difference discrimination result, the prior position attribution check result, and the multi-view consistency verification result, respectively. The status identifier includes pointing to the main body, pointing to interference, and pending verification.

[0041] If at least two status indicators point to interference and the candidate point does not belong to the critical assembly reference area, the candidate point is determined as the interference feature point. The critical assembly reference area includes the design reference surface, the locating pin hole, the bolt hole array, or the area where the assembly stop is located.

[0042] If at least two status indicators point to the body, or if the candidate point belongs to the key assembly reference area and the multi-view image shows that it is continuous with the body contour, the candidate point is determined as the body feature point.

[0043] In the event of a conflict in the status identifier, the point type of the candidate point is determined based on the region where the candidate point is located, the distance between the candidate point and the key assembly reference region, the type of the interference feature prior region where the candidate point is located, and the performance of the candidate point in the multi-view image.

[0044] Further, registration weights are configured based on the ontological feature points and the interference feature points, including:

[0045] Configure normal registration weights for the ontological feature points;

[0046] Assign zero weights to high-confidence fly edges and high-confidence residual convex points, or remove high-confidence fly edges and high-confidence residual convex points from the fine registration participation points;

[0047] Assign registration weights lower than those for the ontological feature points to high-confidence burr points, molding sand fragment points, and medium-confidence interference points;

[0048] For points of doubt retained through arbitration, assign transitional weights between the weights of ontological feature points and the weights of interfering feature points.

[0049] Furthermore, based on the configured registration weights, a weighted fine registration is performed between the fused point cloud and the CAD theoretical model, including:

[0050] The point cloud after configuring the registration weights is used as the object to be registered, and the triangular mesh surface of the CAD theoretical model is used as the registration target.

[0051] Determine the nearest projection point or corresponding surface patch of the point cloud point to be registered on the CAD theoretical surface, and calculate the weighted distance based on the registration weight of the point cloud point;

[0052] Find the rotation matrix and translation vector that make the weighted distance converge, and obtain the fine registration matrix;

[0053] Convergence is determined based on the weighted distance change, the fine registration matrix change, the body high-stiffness registration feature residual, and the number of iterations. After the convergence condition is met, self-verification is performed based on at least one of the body high-stiffness registration feature residual, the overlap error of the common area from different perspectives, and the consistency of the deviation distribution.

[0054] Furthermore, generate actual manufacturing deviations, deviation chromatograms, and interference point removal records, including:

[0055] The feature points of the body are transformed to the coordinate system of the CAD model through a fine registration matrix, the nearest projection point of the feature point on the CAD theoretical surface is determined, and the normal direction of the surface where the projection point is located is obtained.

[0056] Taking the outward normal direction of the CAD theoretical surface as the positive direction, calculate the signed distance of the body feature point along the normal direction relative to the CAD theoretical surface, and determine the signed distance as the actual manufacturing deviation;

[0057] Compare the actual manufacturing deviations with the tolerance reference data to generate status indicators such as qualified, warning, out of tolerance, or pending re-inspection.

[0058] Map out-of-tolerance points or out-of-tolerance areas to the surface of the CAD model to generate a deviation chromatogram;

[0059] Point cloud points marked as interference feature points are removed and recorded as interference points.

[0060] Furthermore, the method includes:

[0061] In the process of acquiring the CAD theoretical model, gating system design, mold cavity partition map and process parameters corresponding to the current casting type, version verification is performed. If any abnormality occurs during the version verification, the acquisition or fusion of the casting surface measurement point cloud, the establishment of the coarse registration benchmark, the division of the point cloud points, the weighted fine registration or the self-verification process, abnormality and downgrade processing is performed and an abnormality record is generated.

[0062] The anomalies include at least one of the following: inconsistent model versions, excessive temperature, missing local point clouds, insufficient high-stiffness registration features of the ontology, conflicting discrimination results, non-convergence of weighted fine registration, or failure of self-verification.

[0063] The anomaly and degradation handling includes at least one of the following: pausing measurement, rescanning, degraded coarse registration, arbitration, weight adjustment, rollback to coarse registration, manual takeover, or pending re-inspection.

[0064] Compared with the prior art, the present invention provides an online method for measuring the dimensional deviation of non-ferrous metal castings, which has the following advantages:

[0065] 1. This invention identifies the prior regions of interference features such as flash, residual protrusions, seams, and molding sand residue by constructing a prior information database. A coarse registration datum is established based on the high-rigidity registration features of the casting body, ensuring that the initial datum for point cloud registration is primarily determined by the stable structure of the casting body. Furthermore, through geometrical difference discrimination, prior position assignment checks, and multi-view consistency verification, the point cloud points are divided into body feature points and interference feature points. Weighted fine registration is then performed after configuring registration weights according to point type, preventing interference feature points from dominating the registration process as ordinary valid contour points. After the weighted fine registration passes self-verification, only the signed normal distance of the body feature points relative to the CAD theoretical surface is calculated. This reduces the bias of non-design contour residues on the registration matrix, separates the actual manufacturing deviation of the casting body from interference protrusions, improves the consistency of repeated measurements, the stability of online out-of-tolerance judgment, and the reliability of the deviation chromatogram in indicating mold wear location and process drift direction.

[0066] 2. This invention, by associating and processing CAD theoretical models, gating system designs, mold cavity partition maps, process parameters, casting surface measurement point clouds, and multi-view images within the same measurement process, transforms the prone locations of flash, residual protrusions, seams, and molding sand residues into callable spatial discrimination conditions. Combined with model version verification, point cloud fusion verification, registration self-verification, and anomaly recording, it can handle situations such as casting type switching, model version inconsistency, local point cloud missing, point cloud discrimination conflict, or abnormal registration status by pausing, rescanning, reverting, manual confirmation, or awaiting re-inspection. This reduces the reliance on manual experience and the adequacy of previous cleaning in the online inspection process, and improves the automation level, anomaly handling reliability, and inspection process traceability of the contour dimension measurement process in continuous casting production lines. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of the online measurement system for the contour dimensional deviation of non-ferrous metal castings according to the present invention;

[0068] Figure 2 This is a schematic diagram of the online measurement method for the contour dimensional deviation of non-ferrous metal castings according to the present invention;

[0069] Figure 3 This is a schematic diagram illustrating the construction of the prior region for interference features in this invention.

[0070] Figure 4 This is a schematic diagram illustrating the establishment of a coarse registration reference based on the high-stiffness registration feature of the body according to the present invention.

[0071] Figure 5 This is a schematic diagram illustrating the interference feature point identification and point type classification of the present invention;

[0072] Figure 6 This is a schematic diagram of the weighted fine registration of the present invention;

[0073] Figure 7 This is a schematic diagram illustrating the actual manufacturing deviation calculation and deviation output of the present invention;

[0074] Figure 8 This is a schematic diagram of the closed-loop feedback of the present invention;

[0075] Figure 9 This is a schematic diagram of the abnormal degradation processing of the present invention. Detailed Implementation

[0076] To make the purpose, technical solution, and beneficial effects of this application clearer, the following description, in conjunction with the accompanying drawings and specific embodiments, further illustrates this application. It should be understood that the following specific embodiments are only used to illustrate the technical solution of this application and should not be construed as limiting the scope of protection of this application. Without departing from the inventive concept of this application, those skilled in the art can make equivalent substitutions or adaptive adjustments to the execution subject, acquisition device, parameter source, data storage method, interface form, and output form according to the actual application scenario. All such substitutions or adjustments should be understood as falling within the scope of implementation of this application.

[0077] Please see Figure 1 - Figure 9 This invention provides an online method for measuring the dimensional deviation of non-ferrous metal castings.

[0078] In the online contour dimension inspection of non-ferrous metal castings, the existing inspection method usually involves collecting point clouds on the surface of the casting after demolding, cooling, shot blasting or sandblasting. The collected point clouds are then rigidly registered with the corresponding CAD theoretical model, and the contour dimension deviation is determined based on the distance between the registered point cloud and the surface of the CAD theoretical model. For castings with less surface residue, regular contours and thorough cleaning, this method can meet certain online inspection requirements. However, in the continuous production process of non-ferrous metal castings such as aluminum alloys, copper alloys, and magnesium alloys, after demolding, gating and riser removal, shot blasting or sandblasting, thin sheet-like flash may still exist at the parting surface, blocky residual protrusions may still exist at the gating or riser removal, strip-like burrs may still exist at the mold parting line, and molding sand residue may still adhere to the cavity, deep rib root or sand core positioning surface.

[0079] The aforementioned flash, residual protrusions, seams, and molding sand remnants are not part of the design contour defined by the CAD theoretical model. However, they are usually continuously connected to the surface of the casting body, and their thickness, height, or protrusion may be close to the allowable manufacturing deviation of the casting. If the aforementioned point cloud points and the point cloud points of the casting body are included as valid contour points for matching during the registration process, the rotation matrix and translation vector obtained from the registration may be offset by interference points. This results in the deviation calculated subsequently including the manufacturing deviation of the casting body, the offset of the registration datum, and the protrusion of the interference residue. This can easily lead to repeated inspection results for the same casting. Inconsistencies can easily lead to unstable determination of out-of-tolerance values ​​for castings in the same batch, and affect the accuracy of the deviation chromatogram in indicating the location of mold wear and the direction of process drift. Based on this, the main technical problem solved by this implementation is: in the online three-dimensional measurement scenario of non-ferrous metal castings, how to combine the surface measurement point cloud of the casting, the CAD theoretical model and the prior information of the casting process to identify and suppress the influence of interfering feature points such as flash, residual protrusions, seams and molding sand fragments on the registration datum, so that the registration result is mainly determined by the real contour of the casting body, and the output contour dimension deviation can reflect the real manufacturing deviation of the casting body.

[0080] The implementation process of this embodiment may include the following: The processing equipment first acquires the CAD theoretical model, gating system design, mold cavity partition map, mold status and furnace process file corresponding to the current casting type, and establishes a process prior information database that can be called by subsequent point cloud processing and registration process based on the above information; According to the process prior information database, the processing equipment determines the positions that are prone to interference residues, such as parting surface, gating and riser, mold parting line, cavity, deep rib root and sand core positioning surface, as the positions prone to interference residues, and generates the corresponding interference feature prior area in the point cloud coordinate system or CAD model coordinate system;

[0081] During the data acquisition phase, the processing equipment acquires the measurement point cloud of the casting surface through the 3D scanning module, acquires images of the casting from multiple perspectives through the multi-view image module, and acquires the surface temperature distribution of the casting through the temperature acquisition module. Before the point cloud enters the fine registration, the processing equipment first extracts stable structural features such as reference planes, positioning pin holes, bolt hole arrays, assembly stops, or large-area plane normals from the casting point cloud, and matches these stable structural features with the corresponding features in the CAD theoretical model to establish a coarse registration reference, so as to reduce the influence of burrs, residual protrusions, and other external protrusions on the initial registration pose.

[0082] After coarse registration, the processing device sequentially performs geometric morphology difference discrimination, prior location attribution check, and multi-view consistency verification on the coarsely registered point cloud. Combining the discrimination results and arbitration rules, it divides the point cloud points into ontological feature points and interfering feature points. In the fine registration stage, the processing device assigns normal registration weights to ontological feature points and assigns zero or low weights to high-confidence interfering feature points, ensuring that interfering feature points do not participate in corresponding point matching, or participate in corresponding point matching with a weight lower than that of ontological feature points. After fine registration passes self-verification, the processing device only performs on-the-body registration. The feature points are calculated with their normal signed distance relative to the CAD theoretical surface, and this normal signed distance is used as the actual manufacturing deviation of the casting body. Subsequently, the processing equipment maps the actual manufacturing deviation to the mold cavity partition and furnace process file, generating deviation chromatograms, out-of-tolerance judgment results, interference point removal records, and mold maintenance or process review suggestions. Through the above processing, the casting process prior information, point cloud point classification, weighted registration, deviation separation, and feedback closed loop can work together to reduce the impact of interference feature points on the registration datum within the online detection cycle.

[0083] In this embodiment, the casting surface measurement point cloud is three-dimensional point data of the outer surface of the casting acquired by a line laser three-dimensional scanning device, a surface structured light scanning device, or other three-dimensional measuring devices. The casting surface measurement point cloud includes at least the following fields: point coordinates, normal, reflection intensity, timestamp, viewpoint number, and acquisition device number. The CAD theoretical model is three-dimensional model data corresponding to the current casting type, mold cavity version, and design status. The CAD theoretical model may include model information such as body surface, design datum, parting line, gate position, riser position, mold parting line, tolerance range, machining reserved surface, assembly stop, and hole array. The process prior information database is a callable data set formed by the CAD theoretical model, gating system design, mold cavity partition map, process parameters, and historical defect records, used to determine the spatial areas where interference features may occur. The high-rigidity registration features of the casting body are structural features that are stable in position, have high structural rigidity, and are not easily covered by flash, residual protrusions, seams, and molding sand fragments. Examples include design reference surfaces, locating pin holes, bolt hole arrays, assembly stops, large-area plane normals, and the roots of symmetrical stiffeners. Interference feature points are point cloud points that are not part of the design contour, such as flash, residual protrusions, seams, or molding sand fragments. When these point cloud points participate in ordinary overall registration, they may cause the rotation matrix and translation vector obtained from the registration to be offset. The actual manufacturing deviation is the signed distance of the normal of the feature points of the casting body relative to the CAD theoretical surface after eliminating or reducing the influence of interference feature points and completing fine registration. The deviation chromatogram is a deviation distribution map formed after mapping the actual manufacturing deviation to the CAD model surface or the casting measurement area. The deviation chromatogram is associated with the deviation data archive and the out-of-tolerance judgment results.

[0084] like Figure 1 As shown, in one possible implementation environment, the online measurement unit 100 is located between the casting production line offline station and the machining station, and is used to complete online measurement of contour dimensions, interference feature identification, registration processing, and deviation output before the casting enters the machining process. The online measurement unit 100 may include a 3D scanning module 110, a multi-view image module 120, a clamping turntable module 130, a model and prior library module 140, a registration processing module 150, an interference identification module 160, a deviation output module 170, a closed-loop feedback module 180, and an anomaly and degradation module 190. The 3D scanning module 110 is used to acquire the measurement point cloud of the casting surface, and can use a line laser 3D scanning device or a surface structured light scanning device. The device includes a robot-mounted scanning head or a fixed gantry scanning head; a multi-view image module 120 is used to acquire image sequences of the casting from the front, side, upper oblique, lower oblique, or other visible viewpoints, so as to verify the visibility, edge continuity, and color consistency of candidate interference points in subsequent processing; a clamping turntable module 130 is used to clamp and rotate the casting according to the clamping instructions issued by the industrial controller, and output operating information such as clamping status, turntable angle, and turntable encoder data; a model and prior library module 140 is used to store and call data such as CAD theoretical models, gating system design files, mold cavity partition maps, process prior areas, color reference libraries, and historical deviation data archives;

[0085] The registration processing module 150 performs point cloud fusion, coarse registration, weighted fine registration, convergence judgment, and registration self-verification on the measured point cloud of the casting surface; the interference identification module 160 identifies interference feature points such as flash, residual protrusions, seams, and molding sand fragments based on geometric morphology differences, prior position attribution, multi-view consistency, and arbitration rules; the deviation output module 170 calculates the actual manufacturing deviation based on the fine registration results and generates a deviation chromatogram, out-of-tolerance judgment results, and interference point removal records; the closed-loop feedback module 180 maps the deviation area to the mold cavity partition and outputs maintenance suggestions, review suggestions, or pending inspection status to the casting production line control system, mold management system, or quality management system; the anomaly and degradation module 190 performs pause, rescan, rollback, version check, manual takeover, or pending inspection processing when there are problems such as excessive temperature, missing point cloud, synchronization anomaly, model version conflict, coarse registration failure, fine registration non-convergence, or equipment failure.

[0086] In one possible implementation, the 3D scanning module 110, the multi-view image module 120, and the clamping turntable module 130 can be connected to the industrial controller via industrial Ethernet, fieldbus, fiber optic, or other industrial communication interfaces. The industrial controller can serve as the main execution unit of the online measurement unit 100, or it can be handled by an edge computing device, production line server, industrial robot controller, embedded controller, or other electronic devices with data processing and communication capabilities. At the start of the measurement cycle, the industrial controller initiates the online measurement process based on the casting entry trigger signal and reads the corresponding CAD theoretical model and process prior information based on the product number or mold number. After scanning, the industrial controller performs point cloud fusion and coarse registration. After coarse registration passes self-verification, the industrial controller performs interference feature point identification. After interference identification, the industrial controller performs weighted fine registration. After fine registration passes self-verification, the industrial controller outputs the actual manufacturing deviation, deviation chromatogram, out-of-tolerance judgment result, and closed-loop feedback information. Through the coordination between the above data acquisition, model calling, registration processing, interference identification, and deviation output, point cloud, image, process parameters, mold status, and temperature data can form a traceable data chain within the same measurement cycle, thereby avoiding inconsistencies in measurement objects, model versions, or process status caused by independent acquisition by different devices.

[0087] like Figure 2 As shown, the online measurement method for the contour dimension deviation of non-ferrous metal castings provided in this application embodiment may include S101 to S106; without affecting the technical effect of this application, some verification steps, abnormal handling steps and feedback steps may be executed in parallel, or executed in advance according to the measurement cycle and equipment status.

[0088] In S101, the processing equipment acquires the CAD theoretical model, gating system design, mold cavity diagram, and mold status corresponding to the current casting type, and constructs a process prior information database. Specifically, when the production line is started for the first time, the casting type is switched, mold maintenance is completed, the system is restarted, or the measurement cycle begins, the industrial controller reads the CAD theoretical model file corresponding to the current casting type from the model and prior information database module 140. The industrial controller can also read production information from the casting production line management system, which includes at least fields such as type number, furnace number, process route, and current workstation status; and read mold information from the mold management system, which includes at least fields such as mold number, cavity version number, cumulative number of uses, and last maintenance time.

[0089] The industrial controller performs a consistency check on the CAD theoretical model version, mold cavity version, and casting type delivered from the production line. If the CAD theoretical model version, mold cavity version, and casting type are consistent, the process prior information database construction process is allowed. If any information is inconsistent, the exception and degradation module 190 suspends the current casting scanning process, outputs a model version verification request, and does not enter the registration process until the correct model is confirmed. Through the above version verification, global deviation misjudgment caused by incorrect model calls can be avoided.

[0090] When constructing the process prior information database, the industrial controller can analyze the CAD theoretical model, gating system design documents, mold cavity diagrams, and process parameters to generate data content related to subsequent point cloud processing. Among these, the triangular mesh surfaces in the CAD theoretical model can form body surface data; design datum surfaces, locating pin holes, bolt hole arrays, assembly stops, large-area plane normals, and machining reserved surfaces can form body registration feature data; parting line lines can form parting line trajectory data; gate and riser sections can form gate and riser location data; parting line lines can form parting line trajectory data; cavities, deep rib roots, and sand core locating surfaces can form sand residue candidate area data; and design tolerance ranges can form tolerance reference data. The above data can be stored in the same database table, or separately in model files, configuration files, or process parameter tables, as long as the industrial controller can read and call them within the measurement cycle.

[0091] like Figure 3 As shown, the industrial controller constructs a priori regions of interference characteristics based on the process prior information database. For the parting line, the industrial controller can generate a thin-shell-shaped flash priori region along both sides of the parting line or the outer normal direction. The outer thickness of this flash priori region can be determined based on the casting process manual, statistics of flash residual thickness of similar castings, mold closing accuracy, and on-site calibration data. For example, the outer thickness can be set from 0.1 mm to 2 mm and adjusted according to the casting size, casting material, mold accuracy, and historical flash residue.

[0092] For the gate and riser sections, the industrial controller can generate blocky, columnar, or hemispherical residual protrusion prior regions outside the gate or riser sections. The radius or height of these residual protrusion prior regions can be determined based on statistics of residual height from the gate and riser removal process, cutting equipment accuracy, and historical residual protrusion height records. For example, the radius of the residual protrusion prior region can be set from 2 mm to 8 mm. For the parting line, the industrial controller can generate strip-shaped burr prior regions along both sides of the parting line. The width of these burr prior regions can be determined based on parting accuracy, mold clearance, casting process records, and historical burr width statistics. For example, the width of the pre-test area for the seam can be set to 0.5 mm to 3 mm; for cavities, deep rib roots, or sand core positioning surfaces, the industrial controller can combine the cavity depth, rib root radius, sand core positioning surface position, sandblasting path, and cleaning process records to generate a pre-test area for molding sand debris; for example, areas where the cavity depth exceeds the on-site configuration threshold and the sandblasting path is obstructed can be identified as candidate areas for molding sand debris; the above values ​​are all within the feasible engineering range, and the specific values ​​can be determined by equipment calibration results, on-site test results, historical defect distribution, or manual configuration tables, and should not be construed as limiting the scope of protection of this application.

[0093] In one implementation, the process prior information database can be pre-established by the engineering station during the product introduction stage or the mold online stage, and directly called when the industrial controller performs measurements. The engineering station can pre-generate various interference feature prior regions based on the CAD theoretical model and the gating system design, and save the region data as model additional layers, three-dimensional envelope files, or point cloud spatial index tables. After the casting enters the measurement station, the industrial controller checks the casting type, CAD theoretical model version, and mold cavity version, and calls the corresponding prior region data after the check is passed. This method is suitable for continuous mass production scenarios where the casting type is stable and the model version changes little, and can reduce the amount of data processing during the online measurement cycle.

[0094] In another implementation, the industrial controller can dynamically correct the prior area based on production process data. This production process data includes at least the following fields: cumulative mold usage count, most recent maintenance time, pouring temperature, mold temperature, holding time, cooling time, shot blasting time, sandblasting time, and historical defect records. For example, if a preset number of castings in a mold cavity region exhibit flash residue in the same parting surface area, the industrial controller can expand the flash prior range for that region in subsequent measurement cycles. If a preset number of castings in a gating and riser region do not exhibit residual protrusions after maintenance, the industrial controller can restore the default prior range for that region. The preset number can be determined based on the production line sampling frequency, the number of castings in the same batch, the mold maintenance cycle, and the frequency of historical defects. When performing dynamic correction, the industrial controller can generate correction records, which include at least the correction reason, the range before correction, the range after correction, the applicable mold number, the applicable casting type, and the effective time, to ensure the traceability of the prior area adjustment process.

[0095] In S102, the processing equipment acquires point cloud measurements of the casting surface, multi-view images, and the current process status of the casting. Specifically, after cooling, shot blasting, or sandblasting, the casting enters the online measurement station. The arrival sensor, weighing sensor, RFID reader, or barcode reader sends the casting arrival signal, weight status, and product identification to the industrial controller. The industrial controller matches the product identification with the CAD theoretical model information read in S101. After successful matching, it reads the furnace process file. The furnace process file includes at least the furnace number, pouring temperature, mold temperature, holding time, cooling time, shot blasting time, sandblasting time, and current station time. The above process data is bound to the point cloud measurement results for subsequent prior area correction, temperature compensation, anomaly analysis, and closed-loop feedback.

[0096] Before acquiring point cloud data, the temperature acquisition module obtains the surface temperature distribution of the casting through an infrared thermal imager, a point temperature sensor, or an area array temperature acquisition device. If the surface temperature of the casting is higher than the upper limit of the temperature compensation allowed by the scanning device, or higher than the stable measurement temperature range allowed by the camera, clamping mechanism, and calibration matrix, the industrial controller pauses scanning and waits for cooling. The temperature threshold can be determined according to the scanning device manual, thermal expansion compensation calibration test, on-site safety requirements, and historical measurement stability records. During the pause, the industrial controller rereads the temperature data according to a preset retest cycle. The retest cycle can be determined according to the production line cycle time, casting cooling rate, temperature acquisition module response time, and station buffering capacity. When the surface temperature of the casting enters the allowable measurement range and the temperature change rate is lower than the on-site configuration conditions, the industrial controller resumes scanning. If the surface temperature of the casting cannot enter the allowable measurement range for a long time, the industrial controller marks the casting as pending re-inspection and outputs the waiting reason to the production line control system. Through the above temperature access, retesting, and recovery judgment, the impact of thermal expansion or optical imaging instability under high temperature conditions on point cloud coordinates and registration results can be reduced.

[0097] After receiving the clamping command, the clamping turntable module 130 fixes the casting using multi-point floating jaws, flexible positioning blocks, or adjustable support seats. The clamping position is preferentially selected from process reference bosses, design reference surfaces, stable support surfaces, or high-rigidity areas not involved in deviation calculations, avoiding clamping on parting line flash, residual protrusions from gating and risers, thin-walled free edges, or easily deformable areas. After clamping, the clamping turntable module 130 returns clamping stability signals, jaw positions, and initial turntable angles to the industrial controller. If the clamping force exceeds the preset safety range, or the clamping position overlaps with a priori interference area, the industrial controller readjusts the clamping point or requests manual confirmation. The clamping force safety range can be determined based on the casting material, wall thickness, clamping contact area, jaw calibration records, and repeated measurement deviations after clamping. Through clamping access control, the impact of clamping deformation or clamping obstruction on measurement results can be reduced.

[0098] During the point cloud acquisition stage, the 3D scanning module 110 can adopt a line laser 3D scanning method; the line laser emitter projects narrow-band laser stripes onto the surface of the casting, the camera acquires the stripe deformation image, and the processing equipment reconstructs the 3D points on the surface of the casting based on the laser plane, camera intrinsic and extrinsic parameters, and calibration relationship; to adapt to the coexistence of specular reflection and diffuse reflection on the surface of non-ferrous metals, the industrial controller can divide the scanning area into strong reflective area, normal reflective area, and weak reflective area according to the reflection intensity distribution, and call the corresponding exposure parameters, laser power, or gain parameters respectively; if a certain local area has continuous saturation points or missing points, the industrial controller marks the area and performs supplementary acquisition when the turntable rotates to the adjacent viewpoint; the above parameter adjustment is used to improve the integrity of point cloud acquisition, without changing the processing method of this application based on process prior and weighted registration to suppress the influence of interference feature points.

[0099] In another implementation, the 3D scanning module 110 can use surface structured light scanning. The surface structured light scanning device projects stripes, speckles, or coded light patterns onto the surface of the casting and acquires reflected images through one or more cameras. Then, it reconstructs a 3D point cloud based on phase, coding, or parallax relationships. This method is suitable for shell-type castings with relatively uniform surface reflection or where it is necessary to increase the coverage area of ​​a single acquisition. Regardless of whether line laser scanning or surface structured light scanning is used, as long as the measured point cloud of the casting surface containing coordinates, timestamps, and viewpoint numbers can be output, it can proceed to the subsequent prior discrimination and weighted fine registration process.

[0100] The multi-view image module 120 simultaneously acquires RGB or grayscale images of the casting from multiple perspectives during point cloud acquisition. These multi-view images can come from an auxiliary binocular camera, camera array, fixed industrial camera, or a camera that moves with the robot's scanning head. The industrial controller records fields such as the viewpoint number, exposure parameters, acquisition timestamp, and camera posture for each image and establishes a mapping relationship between the image timestamp and the point cloud timestamp. If the camera exposure is abnormal, the illumination is unstable, or the time deviation between the image and the point cloud exceeds a preset synchronization threshold, the anomaly and degradation module 190 performs re-exposure, time synchronization correction, or discards data from the current sampling period. For example, if the timestamp deviation of the multi-source data exceeds 1 millisecond, synchronization correction can be triggered. This threshold can be determined based on the distributed clock accuracy, scanning speed, turntable speed, and whether the angular displacement or spatial displacement calculated from the time deviation exceeds the allowable error for point cloud fusion. By synchronizing the image with the point cloud, a reliable viewpoint correspondence can be ensured when the point cloud points are subsequently back-projected onto the image plane.

[0101] In S103, the processing device fuses the multi-view point clouds and performs coarse registration based on the high-rigidity registration features of the body. Specifically, after the clamping turntable completes the preset rotation angle or completes a full rotation, the industrial controller reads the turntable angle value corresponding to each timestamp, scans the device posture and calibration matrix, and converts the point clouds collected from different viewpoints to a unified casting measurement coordinate system based on the above data.

[0102] During point cloud fusion, the industrial controller performs continuity checks on the common overlapping areas between adjacent viewpoints. These continuity checks may include common area distance deviation checks, normal consistency checks, viewpoint overlap ratio checks, and edge connectivity checks. The common area distance deviation threshold, normal deviation threshold, and minimum overlap ratio can be determined based on the effective point distance of the scanning device, the turntable positioning accuracy, the camera calibration error, the minimum measurable feature size of the casting, and the repeated measurement results of the standard part. If the common area deviation in a certain angle interval exceeds the allowable range of the equipment calibration, the industrial controller controls the turntable to return to the corresponding angle interval for rescanning. If the continuity requirement is still not met after rescanning, the area is marked as partially missing and excluded in subsequent deviation outputs, or the casting is marked as awaiting re-inspection.

[0103] The local point cloud missing ratio can be determined based on the ratio of the number of sampling locations in the target area where no effective 3D points are formed to the theoretical number of sampling locations in the target area, or it can be determined based on the decrease ratio of the actual number of effective points relative to the number of effective points in similar qualified measurement records. The point cloud missing ratio threshold can be determined based on the effective sampling density of the scanning device, the minimum measurable feature size in the CAD theoretical model, the minimum overlap ratio required for point cloud fusion, and historical successful rescan records. When the local point cloud missing ratio exceeds the field-configured threshold, the industrial controller will trigger rescan first. If the requirements are still not met after rescan, the industrial controller will record the missing area information in the deviation data file. The missing area information includes at least the fields of missing area location, missing ratio, number of rescans, rescan result, and processing result.

[0104] like Figure 4 As shown, in the coarse registration stage, the processing device does not directly perform ordinary nearest neighbor registration using the overall point cloud, but instead prioritizes using the high-rigidity registration features of the body to establish an initial registration datum. The industrial controller extracts the measured features corresponding to the body registration feature data from the fused point cloud. For example, for a bolt hole array, the edge points of the holes can be extracted and the hole centers can be fitted; for a positioning pin hole, the axis of the cylindrical surface or the center of the hole opening can be extracted; for an assembly stop, the center of the stop arc or the stop plane can be extracted; for a design datum plane, a local plane can be fitted and the plane normal can be obtained; for a large area plane or a symmetrical structure, the principal axis direction or the surface normal can be extracted. The industrial controller matches the above measured features with the corresponding theoretical features in the CAD theoretical model and solves the rigid body transformation from the fused point cloud coordinate system to the CAD model coordinate system to obtain the coarse registration matrix. This rigid body transformation can be achieved using conventional point-to-point registration, point-to-surface registration, or pose solving methods based on feature correspondence, as long as the pose transformation relationship between the fused point cloud coordinate system and the CAD model coordinate system can be obtained.

[0105] After coarse registration is completed, the industrial controller performs coarse registration self-verification. The coarse registration self-verification may include checking the alignment error of the high-rigidity registration features, checking the relative distance error between features, checking the feature normal angle error, and checking the overall position deviation. If the alignment error of the high-rigidity registration features meets the on-site configuration requirements, it is allowed to proceed to S104. If the number of unrecognizable high-rigidity registration features reaches a preset ratio, or the alignment error exceeds the allowable range of the equipment calibration, the anomaly and degradation module 190 triggers rollback processing. The rollback processing may include re-extracting the high-rigidity registration features, performing degraded coarse registration using the spindle direction and a large-area reference plane, rescanning key perspectives, or transferring to manual confirmation.

[0106] For example, when the number of unrecognizable features in the body registration feature data reaches 30%, the industrial controller can determine that the currently available features are insufficient to stably support coarse registration and attempt to degrade the coarse registration. This percentage can be determined based on the total number of body registration features, the minimum number of constraints required to solve the rigid body pose, the spatial distribution of each feature, and historical coarse registration failure records, and should not be interpreted as a unique limiting value. If the self-verification requirements still cannot be met after degrading the coarse registration, the industrial controller outputs a manual takeover request and saves fields such as the reason for the coarse registration failure, the location of the missing features, the attempted rollback methods, and the processing results in the exception record. Through coarse registration using high-rigidity body registration features, the initial pose can be mainly determined by the stable structure on the casting body, reducing the impact of interference points such as flash and residual protrusions on the registration results in the initial registration stage.

[0107] In one implementation, when the casting lacks a sufficient number of hole arrays or locating pin holes, the industrial controller can select the design reference plane, the normal of the large-area plane, the symmetry axis of the stiffener, and the stable boundary of the outer contour as the high-rigidity registration features of the body. This method is suitable for castings with few holes but obvious planar or stiffener structures, such as battery trays, motor housings, casings, or large thin-walled parts. In another implementation, when the casting has a complex shape, many local obstructions, or a limited single-scan viewing angle, the industrial controller can first obtain the initial pose based on the mechanical positioning relationship of the clamping turntable, and then use the visible high-rigidity registration features of the body for local correction. All of the above methods use the body structure that is not easily covered by flash, residual protrusions, seams, or molding sand fragments as the basis for coarse registration, thereby avoiding the direct influence of interference points on ordinary overall coarse registration.

[0108] In S104, the processing device identifies interference feature points in the coarsely registered point cloud and divides the point cloud points into body feature points and interference feature points.

[0109] like Figure 5 As shown, the interference identification module 160 can sequentially perform geometric shape difference discrimination, prior position attribution check, multi-view consistency verification and arbitration processing.

[0110] Geometric morphology difference discrimination is used to screen candidate interference points from the spatial morphology of the point cloud itself; the industrial controller takes the coarsely registered point cloud points as the processing object, and for a single point or local point cluster, obtains its normal deviation, neighborhood connectivity, neighborhood thickness, local curvature change, local normal direction difference, and neighborhood point density relative to the CAD theoretical surface; thin sheet-like flash usually appears as a group of points extending along the parting surface direction, with a small thickness and abrupt edges between it and the body surface; blocky residual protrusions usually appear as a group of points that bulge outwards locally near the gate or riser, with obvious curvature changes and a shape close to a block or hemispherical shape; strip-like burrs usually appear as a group of points extending along the mold parting line direction. The points are groups of points that are small in size, long in length, and continuous with the edge of the body; molding sand fragments usually appear as isolated small clusters of points located in concave cavities or deep rib roots, and their color or surface roughness may differ from the metal body; the industrial controller can determine the neighborhood radius, normal angle range, and connectivity conditions based on equipment calibration, historical defect statistics, or manual configuration tables; for example, the neighborhood radius can be set to 2 mm to 5 mm, and this range can be determined based on typical flash thickness, residual protrusion convexity scale, point cloud resolution, and local surface change scale; after the first layer of discrimination, the industrial controller can obtain the candidate set of body points, the candidate interference point set, the outlier noise point set, and the point set to be verified.

[0111] The prior location attribution check is used to determine whether a candidate point falls into the interference feature prior area determined by the process prior information database. The industrial controller performs a spatial attribution check on the candidate point coordinates with the flash prior area, residual protrusion prior area, seam prior area, and molding sand residue prior area. If the geometric shape of the candidate point is consistent with the type of its prior area, for example, a group of thin sheet-like points falls into the parting surface flash prior area, or a group of blocky external protrusions falls into the gating and riser residual protrusion prior area, then the industrial controller marks the candidate point or point cluster as a high-confidence point. Disturbance points; if the geometry of a candidate point is partially consistent with but not completely matched with the prior region, such as a strip-shaped group of points close to the mold parting line but with a local normal continuous with the body surface, it is marked as a medium-confidence interference point; if a candidate point is identified as a body point by geometric shape difference, but its position falls into the prior region of interference features, it is marked as a point to be verified and enters the multi-view consistency check; by checking the prior position attribution, the easy-to-occur locations of interference residues in the casting process can be transformed into spatial discrimination conditions, reducing the possibility of misjudgment caused by point cloud morphology alone.

[0112] Multi-view consistency verification is used to verify the points to be verified, candidate points with medium confidence, and candidate points near key benchmarks. The industrial controller back-projects candidate points onto image planes from multiple viewpoints based on the camera calibration matrix and coarse registration matrix, and reads the local image regions around the back-projected locations. The size of these local image regions can be determined based on camera resolution, point cloud projection error, and local texture scale. For example, a 16-pixel by 16-pixel local image window centered on the back-projected pixel can be selected as the analysis area, or it can be adjusted based on camera resolution, lens distortion correction error, point cloud back-projection error, and the image scale of candidate interference points. The industrial controller then uses the image region... The visibility, edge continuity, and color consistency of the domain are judged by rules. If a candidate point is visible in multiple views and its edge is smooth and continuous with the outline of the casting body, and its color or grayscale is close to the reference area of ​​the body, then the candidate point is likely to be judged as a body point. If the candidate point exhibits outward branches, broken edges, strong reflection abrupt changes, abnormal local shadows, or color that is significantly inconsistent with the metal body in multiple views, or cannot match the visible outline of the CAD theory in most views, then the candidate point is likely to be judged as an interference point. The above judgment can be achieved by threshold comparison, interval judgment, table lookup matching, or manual configuration of rules, without relying on model training or sample training.

[0113] To ensure the stable synthesis of multiple discrimination results, the industrial controller can generate status identifiers for geometric shape difference discrimination results, prior position attribution check results, and multi-view consistency verification results. These status identifiers can include those pointing to the main body, those pointing to interference, and those pending verification. If at least two status identifiers point to interference, and the candidate point does not belong to a key assembly reference area such as a design reference surface, locating pin hole, bolt hole array, or assembly stop, the industrial controller marks the candidate point as an interference feature point. If at least two status identifiers point to the main body, or the candidate point belongs to a key assembly reference area and the multi-view image shows its continuity with the main body contour, the industrial controller marks the candidate point as a main body feature point. If the three status identifiers conflict with each other, or if the candidate point is located in a key assembly reference area but simultaneously falls into a priori interference feature area, arbitration processing is initiated, and the sources of each discrimination are saved in the rejection record.

[0114] For candidate points entering the arbitration process, the industrial controller determines the final point type based on the candidate point's location, distance from the key assembly datum, prior region type, and multi-view image performance. The arbitration process prioritizes avoiding the erroneous removal of key body feature points. For questionable points located near the design datum plane, locating pin holes, bolt hole arrays, or assembly stops, even if they partially fall into the prior region of interference features, they can be temporarily retained as low-weight body points and further confirmed during fine registration self-verification. For questionable points far from the key datum, falling into the prior region of interference features, and exhibiting abrupt edge changes or color anomalies in multi-view images, they can be included as low-weight interference points. For isolated noise points, they can be directly removed from the registration point set, and the reason for removal can be recorded. After arbitration, the industrial controller outputs the point type identification result, which includes at least the fields of interference feature point set, body feature point set, point type identifier, confidence level, source of discrimination, and removal record.

[0115] In one implementation, for castings with thin-walled flash or residual protrusions whose geometry closely resembles the actual contour, the industrial controller can add a temperature field verification after multi-view consistency verification. The temperature field verification is used as an auxiliary verification and does not replace the aforementioned geometric difference discrimination, prior location attribution check, and multi-view consistency verification. Specifically, the industrial controller maps the region where the candidate point is located to an infrared temperature distribution map and compares the temperature difference, cooling rate, or thermal distribution continuity between this region and adjacent body regions. Flash, residual protrusions, or seams may have local temperature variations that differ from the body structure due to differences in thickness, volume, and heat dissipation path. If the temperature field verification result is consistent with the geometric shape and prior location discrimination results, the interference confidence of the candidate point is increased. If the temperature field verification result supports its continuity with the body structure, the interference confidence of the candidate point is reduced, and it is handled by arbitration rules. This method is suitable for scenarios where the entry temperature is still high and the interference characteristics and the body's geometric boundaries are not obvious.

[0116] In another implementation, the interference identification module 160 can identify the connection roots between interference feature points and the casting body. Flash, residual protrusions, and seams are usually continuously connected to the casting body through the connection roots. If only exposed protrusions are removed while the transition points of the connection roots are retained, they may still affect the registration results in fine registration. The industrial controller can track connected points along the candidate interference point group towards the body, identify their connection roots, and mark the connection root points as low-weight interference points or transition points. This method is suitable for scenarios where flash is continuous with the edge of the body and the interference effect is difficult to completely eliminate by removing ordinary outliers. By identifying the connection roots, the set of points participating in high-weight matching in fine registration is closer to the true contour of the casting body, which helps to reduce the registration reference offset.

[0117] In S105, the processing device performs weighted fine registration based on the results of interference feature point identification;

[0118] like Figure 6 As shown, the industrial controller categorizes points based on their type and confidence level output from S104, configuring different registration weights for different point cloud points. Body feature points serve as the primary registration basis, receiving normal registration weights. High-confidence flash points and high-confidence residual convex points pose a higher risk of misaligning the registration benchmark and can be configured with zero weight, preventing them from participating in the matching of corresponding points in fine registration, or they can be directly removed from the set of points participating in fine registration. High-confidence seam points, molding sand fragments, and medium-confidence interference points can be configured with weights lower than body feature points to weaken their impact on the fine registration results. Questionable points temporarily retained after arbitration can be configured with weights between those of body feature points. The transition weight between the weight and the weight of the interference feature point; the specific weight value can be determined based on equipment calibration, field testing and historical operation records; for example, the weight of the body feature point can be set to 1.0, the weight of the high-confidence flash point and residual protrusion point can be set to 0, and the weight of the high-confidence burr point, molding sand fragment point or medium-confidence interference point can be set in the range of 0.05 to 0.2; the above weight values ​​are the feasible engineering range, and the actual values ​​can be determined based on the registration residual stability test, repeated measurement results, risk of erroneous body point rejection, interference point coverage ratio and field calibration records of different casting types, and can be called by the industrial controller as a parameter table.

[0119] In the fine registration process, the industrial controller uses the fused point cloud after weight configuration as the object to be registered and the triangular mesh surface of the CAD theoretical model as the registration target. For the point cloud points participating in the registration, the industrial controller determines the nearest projection point or corresponding surface patch on the CAD theoretical surface and calculates the weighted distance according to the registration weight of the point cloud point. Subsequently, the industrial controller obtains the rotation matrix and translation vector that make the weighted distance meet the convergence condition through point-to-point registration, point-to-surface registration, or other rigid body registration methods, thereby obtaining the fine registration matrix. The above fine registration can be implemented using conventional rigid body registration solution methods in this field, as long as the influence of interfering feature points on the registration result can be reduced according to the point cloud point weights. Unlike ordinary overall registration, this embodiment has adjusted the point type and weight of the points participating in the registration before fine registration by combining process prior information and point cloud point classification results, so that interfering feature points no longer dominate the registration result as ordinary effective contour points.

[0120] The industrial controller performs convergence determination during the fine registration process. Convergence determination can be based on the weighted distance change, the fine registration matrix change, the residual of the body's high-stiffness registration features, and the number of iterations. For example, if the weighted distance change in several consecutive registration processes is lower than the on-site configuration ratio, the fine registration can be determined to have reached the convergence condition. The on-site configuration ratio can be determined based on the fluctuation range of the registration residual obtained from repeated measurements of the same standard part, the measurement accuracy of the scanning equipment, the casting tolerance grade, and the online inspection cycle requirements. If the maximum number of processing iterations is reached but the convergence condition is still not met, or if the weighted distance exhibits continuous oscillation, the anomaly and degradation module 190 triggers a rollback process. The rollback process can include reconfiguring the interference point weights, increasing the weights of the body's high-stiffness registration features, rolling back to the coarse registration result, rescanning key perspectives, or transferring to manual confirmation. Through convergence determination and rollback processing, unreliable bias results can be avoided when the interference point coverage is too high, the point cloud is severely missing, or there are conflicts in the model version.

[0121] After the fine registration meets the convergence condition, the industrial controller performs a fine registration self-verification. This fine registration self-verification can include checking the residual of the high-stiffness registration feature, checking the overlap error of common areas from different perspectives, and checking the consistency of the deviation distribution. Specifically, the industrial controller can check whether the residual between the measured position and the theoretical CAD position of the high-stiffness registration feature is within the allowable range of equipment capability and casting tolerance. It can also check whether the overlap error of point clouds from different perspectives in common areas has decreased, and whether there is a global unidirectional offset in the actual manufacturing deviation distribution that does not conform to the casting structural characteristics. If the residual of the high-stiffness registration feature still exceeds the field configuration threshold after fine registration, the industrial controller determines that the fine registration result is unreliable and triggers anomaly handling. The field configuration threshold can be determined based on the scanning accuracy of the online measurement unit, the casting tolerance level, the repeated measurement results of standard parts, and quality control requirements. Through fine registration self-verification, registration results caused by incorrect weight configuration, abnormal point clouds, or abnormal model versions can be prevented from directly entering the quality judgment process.

[0122] In one implementation, the industrial controller can employ a two-stage weighted fine registration. The first stage uses only high-stiffness registration feature points and high-confidence body surface points for fine registration to obtain a stable body reference. The second stage, while maintaining zero or low weights for interfering feature points, introduces ordinary body points to participate in the registration, improving the overall contour fit. This approach is suitable for scenarios where there are many localized residues on the casting surface, but the high-stiffness registration features are relatively clear. In another implementation, the industrial controller can configure different weights according to the casting region. For example, increasing the weight of body points in the assembly reference area, decreasing the weight of candidate interference points near the parting line and gating system, and adjusting the weight of candidate points for molding sand residues in the cavity area using multi-view visibility. This approach is suitable for scenarios where the distribution of residues varies significantly across different regions of large and complex castings. Both approaches reduce the impact of flash, residual protrusions, seams, and molding sand residues on the fine registration results by adjusting the registration weights of different point cloud points.

[0123] In S106, the processing equipment generates the actual manufacturing deviation based on the fine registration results and forms a closed-loop feedback;

[0124] like Figure 7 As shown, after the fine registration passes self-verification, the industrial controller calculates the actual manufacturing deviation only for the body feature points. Specifically, the industrial controller transforms the body feature points to the CAD model coordinate system using the fine registration matrix, determines the nearest projection point of the body feature point on the CAD theoretical surface, and obtains the normal direction of the surface where the projection point is located. Subsequently, the industrial controller uses the outward normal direction of the CAD theoretical surface as the positive direction, calculates the signed distance of the body feature point along this normal direction relative to the CAD theoretical surface, and uses this signed distance as the actual manufacturing deviation of the casting body. The distance deviating along the outward normal direction of the CAD theoretical surface is recorded as a positive value, and the distance deviating in the opposite direction is recorded as a negative value. Alternatively, the sign can be determined according to the normal direction agreed upon in the product inspection specification. For point cloud points marked as interference feature points, the industrial controller does not include their convexity in the actual manufacturing deviation statistics, but instead records and saves them as interference points. By distinguishing between body feature points and interference feature points, it is possible to avoid misjudging the convexity of flash, residual convex height, burr width, or molding sand residue thickness as casting body contour deviation.

[0125] The industrial controller compares actual manufacturing deviations with tolerance reference data, generating status indicators such as qualified, warning, out of tolerance, or pending re-inspection. The tolerance reference data can come from CAD theoretical models, product inspection specifications, process documents, customer quality requirements, or tolerance configuration tables corresponding to mold cavity versions. For out-of-tolerance points or areas, the industrial controller maps them to the surface of the CAD model and generates a deviation chromatogram. The deviation chromatogram can use different grayscale, color levels, or area filling methods to represent the magnitude and direction of deviations, allowing quality personnel to identify the concentrated locations and trends of deviation changes. The industrial controller can also generate a deviation data archive, which includes at least the following fields: casting number, furnace number, model version, mold number, scan time, precision registration matrix, actual manufacturing deviation, out-of-tolerance area location, interference point rejection record, and anomaly record. The interference point rejection record includes at least the following fields: interference point type, interference point location, source of discrimination, confidence level, weight configuration, and rejection reason. The deviation data archive can be uploaded to the quality management system or stored in a local storage module for subsequent traceability queries, re-inspection comparisons, and process analysis.

[0126] like Figure 8 As shown, the closed-loop feedback module 180 maps the out-of-tolerance area or the concentrated deviation area to the mold cavity partition map, and generates maintenance suggestions or review suggestions by combining the furnace process file and historical deviation records. For example, if a preset number of castings show the same outward expansion deviation in the same cavity partition, and the corresponding area does not belong to the high-confidence interference point area, the closed-loop feedback module 180 generates a mold wear review suggestion. If the deviation suddenly appears in the same furnace or adjacent furnaces and is related to changes in pouring temperature, mold temperature, or cooling time, the closed-loop feedback module 180 generates a process parameter review suggestion. If the current casting interference feature coverage exceeds a set threshold, the closed-loop feedback module 180 generates... The pre-process cleaning process is reviewed and suggested for verification or pending re-inspection. The interference feature coverage rate can be determined based on the proportion of the number of interference feature points to the total number of points in the fused point cloud, or based on the proportion of the area where the interference feature points are located to the area of ​​the effective measurement area. The set threshold can be determined based on historical cleaning anomaly records, interference point coverage ratio statistics, manual re-inspection results, and production line rework standards. The above feedback suggestions are used to assist production line processing and mold maintenance, and do not indicate that this method directly changes the casting process parameters. By associating the actual manufacturing deviation with the mold cavity partition and furnace process file, the online measurement results can be extended from single-piece quality judgment to mold status monitoring and process closed-loop management.

[0127] In one operational example, the aluminum alloy motor end cap enters the online measurement station after shot blasting and sandblasting. After the RFID reader reads the product number, the industrial controller retrieves the corresponding CAD theoretical model, gating system design, and mold cavity partition map, and verifies whether the model version matches the mold cavity version. After verification, the processing equipment determines the prior areas of flash, residual protrusions, seams, and molding sand residue based on the parting surface, gate or riser removal position, mold parting line, and cavity structure in the CAD theoretical model. After the infrared thermal imager detects that the surface temperature of the casting is within the allowable scanning range, the turntable is clamped to fix the casting and rotational scanning is started. The line laser 3D scanning device acquires multi-angle point clouds, and the auxiliary camera acquires multi-view images. After the processing equipment fuses the point clouds from each viewpoint, it solves the coarse registration matrix based on the bolt hole array, locating pin holes, and assembly stops. After the coarse registration is passed, the processing equipment identifies thin-sheet-like external protrusions near the parting surface. The processing equipment identifies a group of points that fall within the prior region of flash and appear as protruding structures bifurcating from the body edge in multiple viewpoints. These points are therefore marked as high-confidence flash points and assigned zero weight. The equipment also identifies a group of blocky external protrusions near the gate removal location. This group falls within the prior region of residual protrusions, and its geometry matches that of the residual protrusions. These points are marked as high-confidence residual protrusion points and assigned zero weight. Subsequently, the processing equipment performs weighted fine registration based on the body points and low-weight points, and performs self-verification using the body's high-stiffness registration features. After the fine registration is successful, the processing equipment calculates the normal signed distance only for the body feature points, generating a true manufacturing deviation chromatogram, and mapping the concentrated deviation areas to the corresponding mold cavity partitions. Through this process, the external protrusion of flash and residual protrusions is not included in the body manufacturing deviation, nor is it used to deviate from the registration benchmark in the fine registration, thus enabling the deviation chromatogram to more stably reflect the dimensional state of the casting body.

[0128] In an abnormal operating condition, after the copper alloy shell enters the online measurement station, the 3D scanning module shows local point cloud gaps deep within the cavity. The industrial controller determines the angular range where the gap occurs based on the point cloud timestamp and the turntable angle, and controls the clamping turntable to return to that angular range for rescanning. If the local point cloud is still missing after rescanning, the industrial controller marks the area as partially missing, outputs a status pending re-inspection, and does not use the area for fine registration or out-of-tolerance judgment. If the number of interference points in the candidate area of ​​the molding sand residue exceeds the on-site configured threshold, the industrial controller suspends the generation of the deviation chromatogram for this casting and outputs a front-end cleaning abnormality alarm to the production line control system. Through the above abnormal handling, unreliable deviation results can be avoided from being output when data is missing or interference coverage is too high.

[0129] In one optional implementation, the process prior information database may include color reference information. The color reference information may be formed from image areas of confirmed body surfaces in the same batch, standard metal surface reference plates, historical qualified casting images, or manually calibrated areas. During multi-view consistency verification, the industrial controller compares the grayscale, color, or texture of the candidate point's back-projected image area with the color reference information. If the candidate point's image area differs significantly from the casting body's metal surface, and the candidate point's position falls within the molding sand residue candidate area, the confidence level of the candidate point being identified as a molding sand residue is increased. This method is suitable for scenarios where molding sand residues and the metal body have distinguishable appearances in the image. If the overall surface grayscale of the casting is similar after sandblasting, the priority of the color reference information in the judgment is reduced to avoid misjudgment due to insufficient color difference.

[0130] In another optional implementation, the deviation output module 170 can generate a registration datum offset record simultaneously after generating the actual manufacturing deviation. The registration datum offset record can be obtained by comparing the pose difference or deviation difference between the ordinary overall registration result and the weighted fine registration result of this embodiment. This record is not directly output as the actual manufacturing deviation, but is used as a detection process record to reflect the degree of influence that interference feature points may have on ordinary overall registration. When the registration datum offset exceeds the field configuration threshold, the deviation output module 170 indicates that there is a significant risk of interference point pull in the casting and saves the interference point removal record together with the deviation data archive. The field configuration threshold can be determined based on the repeated measurement results of the standard part, the historical difference record between ordinary overall registration and weighted fine registration, the casting tolerance grade, and the quality management requirements. In this way, quality personnel can distinguish between measurement datum anomalies and casting body deviations, which is beneficial for subsequent verification.

[0131] In another optional implementation, the closed-loop feedback module 180 can compare the current deviation chromatogram with historical deviation records. The historical deviation records include at least fields such as casting type, mold number, cavity partition, furnace number, production date, deviation area location, deviation direction, and number of consecutive deviations. The processing device can determine whether the current deviation is a persistent deviation, a sudden deviation, or a random deviation based on rules such as the degree of regional overlap, deviation direction, number of consecutive deviations, and occurrence cycle. If the same cavity partition shows a continuous deviation in the same direction in multiple measurement cycles, it is more likely to output a mold wear review suggestion. If the deviation only suddenly appears in a certain furnace and is accompanied by abnormal pouring temperature or cooling time, it is more likely to output a process parameter review suggestion. If the deviation area highly overlaps with the high-confidence interference point area, it outputs a previous cleaning review suggestion or a suggestion to be re-inspected. This processing is based on rule matching and historical record comparison and does not require model training.

[0132] like Figure 9As shown, the embodiments of this application may also include an anomaly and degradation handling process; the anomaly and degradation module 190 runs through the model reading, data acquisition, point cloud fusion, coarse registration, interference identification, fine registration, and deviation output processes; during the model reading stage, if the CAD theoretical model version is inconsistent with the casting type or mold cavity version, the industrial controller suspends the measurement and outputs a model verification request; during the data acquisition stage, if the temperature exceeds the limit, the laser power is lower than the equipment setting range, multiple consecutive frames are lost by the camera, or the turntable encoder angle jumps, the industrial controller stops or pauses the current measurement cycle and generates an anomaly record; the equipment setting range may come from the equipment factory parameters, metrological calibration records, or on-site maintenance configuration table.

[0133] During the point cloud fusion, coarse registration, interference identification, fine registration, and deviation output stages, if local point cloud missingness, insufficient high-stiffness registration features, conflicting discrimination results, non-convergence of fine registration, or self-verification failure occurs, the industrial controller will perform rescanning, degenerate coarse registration, arbitration retention, weight adjustment, rollback to coarse registration, manual intervention, or pending re-inspection processing, respectively. The anomaly record includes at least the anomaly occurrence stage, anomaly type, triggering conditions, processing method, processing result, casting number, model version, and occurrence time. Through the above anomaly handling, the system can select rescanning, pause, rollback, downgrade, or manual confirmation under complex working conditions, avoiding the continued output of formal measurement results when the measurement status is uncertain.

[0134] In this embodiment, thresholds related to prior area size, temperature threshold, time synchronization, point cloud missing ratio, viewpoint overlap error, missing ratio of high-rigidity registration features, image verification window, point weight, fine registration convergence, self-verification, and interference coverage can be stored in a parameter table according to casting type and mold cavity version, and called by the industrial controller at the beginning of the measurement cycle. The above thresholds are not set in isolation based on fixed empirical values, but are determined in combination with scanning device capabilities, point cloud resolution, camera resolution, turntable positioning accuracy, casting tolerance grade, standard part repeated measurement results, historical production records, manual re-inspection results, and on-site calibration results. When adjusting the thresholds, the industrial controller generates a threshold adjustment record, which includes at least the following fields: value before adjustment, value after adjustment, adjustment basis, applicable casting type, applicable mold number, effective time, and operator information, to ensure that the measurement results and parameter change process are traceable.

[0135] Through the above implementation methods, casting process information can be transformed into spatial discrimination conditions in point cloud processing. The high-rigidity registration features of the body can be used to establish a stable coarse registration benchmark. Interference feature points can be eliminated or reduced in weight during fine registration. Deviation calculation can be performed only on the body feature points. Therefore, the influence of flash, residual protrusions, seams, and molding sand residues on the deviation of the registration matrix is ​​reduced. The actual manufacturing deviation of the casting body and the amount of protrusion of interference residues can be recorded separately. Deviation chromatograms and mold cavity partitioning feedback results are also more suitable for online out-of-tolerance judgment, mold maintenance verification, and casting process verification.

[0136] In one device implementation, the online measurement device may include functional modules corresponding to the above method steps; the functional modules may include an acquisition module, a priori processing module, a coarse registration module, an interference identification module, a fine registration module, a deviation output module, and a feedback processing module; the acquisition module is used to acquire the surface measurement point cloud, multi-view images, temperature data, and process status data of the casting; the priori processing module is used to construct or call the process prior information database; the coarse registration module is used to establish an initial registration benchmark based on the high stiffness registration features of the casting body; the interference identification module is used to identify interference feature points such as flash, residual protrusions, seams, and molding sand fragments; the fine registration module is used to perform weighted fine registration according to the point type and weight; the deviation output module is used to generate actual manufacturing deviations, deviation chromatograms, and interference point removal records; the feedback processing module is used to generate mold maintenance suggestions, process review suggestions, or pending inspection status; the above functional modules may be implemented by the same industrial controller software unit, or may be implemented collaboratively by edge computing devices, production line servers, and measurement device controllers.

[0137] In one device implementation, the online measurement device may include a processor, a memory, a communication interface, a data acquisition interface, a control interface, and a display interface. The memory may store CAD theoretical models, a process prior information database, parameter configuration tables, calibration matrices, historical deviation data archives, and computer programs. When the processor executes the program in the memory, it can implement the above-mentioned online measurement method for the contour dimensional deviation of non-ferrous metal castings. The data acquisition interface is used to receive data sent by the 3D scanning module, the multi-view image module, the temperature acquisition module, the positioning sensor, and the turntable encoder. The control interface is used to send control commands to the clamping turntable, the scanning device, the lighting device, or the production line sorting mechanism. The communication interface is used to communicate with the casting production line management system, the mold management system, and the quality management system. The display interface is used to output deviation chromatograms, out-of-tolerance areas, interference point rejection records, and abnormal states. This device implementation belongs to the same inventive concept as the aforementioned method embodiments, and the relevant data acquisition, discrimination, registration, deviation output, and feedback details can be found in the aforementioned method embodiments.

[0138] In one implementation of the storage medium, a computer-readable storage medium may store a computer program or control instructions. When the computer program or control instructions are executed by a processor, the processor implements the above-described online measurement method for the contour dimensional deviation of non-ferrous metal castings. The storage medium may be a solid-state drive, an industrial controller memory, a memory card, a read-only memory, a random access memory, or other media capable of storing program instructions. Since the method of this application can be executed by an industrial controller or other processing device, the implementation of each step through program instructions has a clear engineering basis. The storage medium implementation is only used to support the implementation of the method by a processing device and does not introduce technical solutions independent of the method embodiments.

[0139] The above different implementation methods can be combined without conflict. For example, line laser scanning can be combined with temperature field verification, surface structured light scanning can be combined with multi-view image consistency verification, coarse registration of high-stiffness registration features can be combined with connection root recognition, and dynamic prior region correction can be combined with mold cavity partition feedback. Regardless of the combination, the common goal is to avoid interference feature points such as flash, residual protrusions, seams, and molding sand residues from participating in or dominating registration as ordinary real contour points during the online three-dimensional measurement of non-ferrous metal castings, thereby obtaining a more reliable output of real manufacturing deviations.

[0140] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included 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.

[0141] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online measurement of dimensional deviations in the contours of non-ferrous metal castings, characterized in that, include: Obtain the CAD theoretical model, gating system design, mold cavity partition map and process parameters corresponding to the current casting type, construct a process prior information database, and determine the prior regions of interference features based on the process prior information database; Collect measurement point clouds and multi-view images of the casting surface, perform point cloud fusion on the measurement point clouds of the casting surface, and establish a coarse registration benchmark based on the high stiffness registration features of the body and the corresponding features in the CAD theoretical model. After coarse registration, geometric morphological differences, prior location attribution checks, and multi-view consistency verifications are performed on the point cloud to divide the point cloud points into ontological feature points and interference feature points. Registration weights are configured based on the ontological feature points and the interference feature points, and the fused point cloud and the CAD theoretical model are then finely registered using weighted averages based on the configured registration weights. After the weighted fine registration passes self-verification, the signed normal distance of the feature points relative to the CAD theoretical surface is calculated only for the feature points of the body, and the actual manufacturing deviation, deviation chromatogram and interference point removal record are generated.

2. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Construct a process prior information database, including: Perform consistency checks on the CAD theoretical model version, mold cavity version, and casting type; After verification, the CAD theoretical model, gating system design documents, mold cavity partition diagram and process parameters are analyzed to form a process prior information database, which includes body surface data, body registration feature data, parting surface trajectory data, gating and riser location data, mold parting line trajectory data, molding sand residue candidate area data and tolerance reference data.

3. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Determine the prior region of interference features, including: Generate flash prior regions along both sides of the parting line or in the direction of the outer normal; A residual raised prior region is generated on the outside of the gate or riser section; Generate a priori areas for seam cut along both sides of the mold parting line; The preliminary area for generating molding sand residue is generated based on the cavity depth, the radius of the reinforcing bar root, the position of the sand core positioning surface, the sandblasting path, and the cleaning process record.

4. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Acquire point cloud data and multi-view images of the casting surface, including: In response to the casting arrival signal and the type identifier, the type identifier is matched with the information in the CAD theoretical model; After successful matching, read the furnace process file and bind the furnace process file to the measurement point cloud of the casting surface. Once the surface temperature of the casting enters the allowable measurement range and the rate of temperature change is lower than the on-site configuration conditions, the clamping turntable is controlled to fix the casting, and the surface measurement point cloud and multi-view images of the casting are acquired. Establish a mapping relationship between image timestamps and point cloud timestamps.

5. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Point cloud fusion is performed on the measured point cloud of the casting surface, including: Read the turntable angle value, scan the device posture and calibration matrix corresponding to each timestamp, and convert the point cloud collected from different perspectives to a unified casting measurement coordinate system; Perform continuity checks on the common overlapping areas between adjacent viewpoints; When the deviation in the public area exceeds the allowable range of the equipment calibration, the control turntable returns to the corresponding angle range for supplementary scanning; If the continuity requirement is still not met after rescanning, the corresponding area is marked as locally missing, and the missing area information is recorded.

6. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Establishing a coarse registration benchmark includes: Extract the high-stiffness registration features of the body corresponding to the body registration feature data from the fused point cloud. The high-stiffness registration features of the body include at least one of the following: design reference surface, locating pin hole, bolt hole array or assembly stop. The high-rigidity registration features of the body are matched with the corresponding theoretical features in the CAD theoretical model, and the rigid body transformation from the fused point cloud coordinate system to the CAD model coordinate system is solved to obtain the coarse registration matrix. Perform coarse registration self-verification after coarse registration is completed; When the body high-stiffness registration features are insufficient or the alignment error exceeds the allowable range of the equipment calibration, perform at least one of the following processes: re-extract the body high-stiffness registration features, degrade coarse registration, rescan key viewpoints, or transfer to manual confirmation.

7. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, The coarsely registered point cloud is subjected to geometric morphology difference discrimination, prior location assignment check, and multi-view consistency verification, including: After coarse registration, obtain the normal deviation, neighborhood connectivity, neighborhood thickness, local curvature change, local normal direction difference, and neighborhood point density of point cloud points or local point clusters relative to the CAD theoretical surface, and screen candidate interference points. Spatial assignment checks are performed on the candidate point coordinates and the prior areas of the flash, residual protrusions, seams, and molding sand fragments. Candidate points are back-projected onto image planes from multiple perspectives, and multi-view consistency is checked based on the visibility, edge continuity, and color consistency of local image regions.

8. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, The point cloud points are divided into ontological feature points and interference feature points, including: A status identifier is generated for the geometric shape difference discrimination result, the prior position attribution check result, and the multi-view consistency verification result, respectively. The status identifier includes pointing to the main body, pointing to interference, and pending verification. If at least two status indicators point to interference and the candidate point does not belong to the critical assembly reference area, the candidate point is determined as the interference feature point. The critical assembly reference area includes the design reference surface, the locating pin hole, the bolt hole array, or the area where the assembly stop is located. If at least two status indicators point to the body, or if the candidate point belongs to the key assembly reference area and the multi-view image shows that it is continuous with the body contour, the candidate point is determined as the body feature point. In the event of a conflict in the status identifier, the point type of the candidate point is determined based on the region where the candidate point is located, the distance between the candidate point and the key assembly reference region, the type of the interference feature prior region where the candidate point is located, and the performance of the candidate point in the multi-view image.

9. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Configure registration weights based on the ontological feature points and the interference feature points, including: Configure normal registration weights for the ontological feature points; Assign zero weights to high-confidence fly edges and high-confidence residual convex points, or remove high-confidence fly edges and high-confidence residual convex points from the fine registration participation points; Assign registration weights lower than those for the ontological feature points to high-confidence burr points, molding sand fragment points, and medium-confidence interference points; For points of doubt retained through arbitration, assign transitional weights between the weights of ontological feature points and the weights of interfering feature points.

10. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Based on the configured registration weights, a weighted fine registration is performed between the fused point cloud and the CAD theoretical model, including: The point cloud after configuring the registration weights is used as the object to be registered, and the triangular mesh surface of the CAD theoretical model is used as the registration target. Determine the nearest projection point or corresponding surface patch of the point cloud point to be registered on the CAD theoretical surface, and calculate the weighted distance based on the registration weight of the point cloud point; Find the rotation matrix and translation vector that make the weighted distance converge, and obtain the fine registration matrix; Convergence is determined based on the weighted distance change, the fine registration matrix change, the body high-stiffness registration feature residual, and the number of iterations. After the convergence condition is met, self-verification is performed based on at least one of the body high-stiffness registration feature residual, the overlap error of the common area from different perspectives, and the consistency of the deviation distribution.

11. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, Generate actual manufacturing deviations, deviation chromatograms, and interference point removal records, including: The feature points of the body are transformed to the coordinate system of the CAD model through a fine registration matrix, the nearest projection point of the feature point on the CAD theoretical surface is determined, and the normal direction of the surface where the projection point is located is obtained. Taking the outward normal direction of the CAD theoretical surface as the positive direction, calculate the signed distance of the body feature point along the normal direction relative to the CAD theoretical surface, and determine the signed distance as the actual manufacturing deviation; Compare the actual manufacturing deviations with the tolerance reference data to generate status indicators such as qualified, warning, out of tolerance, or pending re-inspection. Map out-of-tolerance points or out-of-tolerance areas to the surface of the CAD model to generate a deviation chromatogram; Point cloud points marked as interference feature points are removed and recorded as interference points.

12. The method for online measurement of the dimensional deviation of non-ferrous metal castings according to claim 1, characterized in that, The method includes: In the process of acquiring the CAD theoretical model, gating system design, mold cavity partition map and process parameters corresponding to the current casting type, version verification is performed. If any abnormality occurs during the version verification, the acquisition or fusion of the casting surface measurement point cloud, the establishment of the coarse registration benchmark, the division of the point cloud points, the weighted fine registration or the self-verification process, abnormality and downgrade processing is performed and an abnormality record is generated. The anomalies include at least one of the following: inconsistent model versions, excessive temperature, missing local point clouds, insufficient high-stiffness registration features of the ontology, conflicting discrimination results, non-convergence of weighted fine registration, or failure of self-verification. The anomaly and degradation handling includes at least one of the following: pausing measurement, rescanning, degraded coarse registration, arbitration, weight adjustment, rollback to coarse registration, manual takeover, or pending re-inspection.

Images