An automated multi-alloy carburizing product post-processing production line

Abstract

The application discloses a multi-element alloy co-permeation product post-processing production line for automatic unloading and sorting, and relates to the technical field of metal surface treatment. The production line comprises a multi-modal perception module, a data processing module, a sorting decision module, a sorting execution module and a re-measurement feedback module. The multi-modal perception module collects workpiece surface image and physical characteristic data, and constructs a calibration gray scalar field. The data processing module constructs a two-dimensional simple complex and obtains a critical complex by discrete Morse simplification, performs persistent homology calculation to extract zero-order connected component number, zero-order scale parameter and first-order hole persistence parameter, and represents residual penetrant cluster and permeation layer discontinuous area. The sorting decision module converts topological parameters into shunt instructions and post-processing process parameters. The sorting execution module guides the workpiece into a corresponding branch and adjusts the post-processing working condition. The re-measurement feedback module re-imaging and re-calculates topological degradation rate to update the mapping rule. The application replaces artificial visual inspection with topological characteristics, and solves the problems of different unloading and sorting standards and solidified process parameters after co-permeation post-processing.

Description

Technical Field

[0001] This application relates to the field of metal surface treatment technology, and in particular to an automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products. Background Technology

[0002] Multi-alloy co-diffusion is a highly efficient metal corrosion protection process widely used in fasteners, rail transit components, and power fittings. After co-diffusion, residual diluent dust usually adheres to the workpiece surface, and due to fluctuations in the temperature and atmosphere fields within the furnace, the uniformity and degree of oxidation of the diluent layer vary among different workpieces. Traditional post-processing production lines rely heavily on manual visual inspection and sorting, and the operating parameters of post-processing equipment (such as shot blasting machines and cleaning machines) are usually fixed preset values, making it impossible to dynamically adjust them according to the actual surface condition of the preceding processes.

[0003] Existing automated vision sorting solutions mostly employ ordinary visible light imaging, grayscale threshold segmentation, or geometric contour recognition methods. These methods struggle to distinguish between high-gloss dust, diffuse reflection infiltration layers, and localized oxide spots on the surface of co-infiltrated workpieces, easily confusing metallic highlights with dust deposits and leading to unstable sorting standards. Applying strong cleaning to workpieces with relatively good infiltration layer conditions may damage the surface infiltration layer; conversely, applying weak cleaning to workpieces with severe dust agglomeration fails to meet corrosion protection standards.

[0004] In addition, existing monitoring methods mostly focus on two-dimensional area or brightness statistics when analyzing surface conditions, lacking means to continuously and quantitatively characterize the connectivity of residual penetrant clusters and the discontinuous pore structure of the penetrant layer. Furthermore, the unloading, sorting, and post-processing stages are isolated from each other, failing to form a closed-loop control based on data feedback. This results in poor production flexibility and difficulty in adapting to the production needs of multiple varieties and small batches.

[0005] Therefore, how to provide an automated unloading and sorting production line that can adaptively adjust sorting strategies and post-processing parameters is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] This invention addresses the aforementioned problems in existing technologies by providing an automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products.

[0007] The objective of this invention is primarily achieved through the following approach:

[0008] An automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products includes:

[0009] The unloading and conveying module is used to receive and transport workpieces after multi-alloy co-diffusion treatment;

[0010] A multimodal sensing module is installed along the unloading and conveying module to collect surface state data and physical feature data of the workpiece flowing through it.

[0011] The data processing module is electrically connected to the multimodal sensing module and is used to process the surface state data and physical feature data to extract feature parameters that characterize the connectivity state of residual infiltration agent on the workpiece surface and the uniformity of the infiltration layer.

[0012] The sorting decision module is electrically connected to the data processing module and is used to receive the feature parameters and generate sorting instructions and corresponding post-processing parameters according to preset mapping rules.

[0013] The sorting execution module is located at the end of the unloading and conveying module and is electrically connected to the sorting decision module. It is used to respond to the sorting command and divert the workpieces to the corresponding processing channels.

[0014] The post-processing module includes multiple parallel post-processing branches for receiving the diverted workpieces and adjusting the operating conditions according to the post-processing process parameters.

[0015] Preferably, the data processing module includes:

[0016] The scalar field construction unit is used to optically enhance the surface state data, extract pixel-level diffusion layer color difference, residual dust brightness and oxidation highlight component, and generate a calibrated grayscale scalar field.

[0017] The topology analysis unit is used to construct a two-dimensional simple complex based on pixel adjacency relationships, and eliminate non-critical units based on the pairing rules of the discrete Morse vector field, while outputting critical units that preserve the complex.

[0018] A persistent cohomology unit is used to perform persistent cohomology calculations on the critical unit while preserving the complex, and outputs the number of zero-order cohomology connected branches, the zero-order cohomology scale parameter, and the first-order cohomology hole persistence parameter.

[0019] The zero-order coherent connectivity branch number and zero-order coherent scale parameter are used to characterize the connectivity zone between residual infiltrate clusters and dust adhesion, while the first-order coherent pore durability parameter is used to characterize the discontinuous or oxidized isolation zone of the infiltrate layer.

[0020] Preferably, the continuous cohomology unit executes an improved continuous cohomology algorithm including:

[0021] Extract the critical units and retain the scalar values ​​corresponding to each critical unit in the complex. Arrange them into multiple thresholds according to their size. Under each threshold, retain the critical units and algebraic connection relationships that are not less than the current threshold to obtain a multi-threshold filtering sequence.

[0022] Construct a boundary matrix and perform column elimination. During the elimination process, calculate the temporary lifetime of the homology generators, remove generators with a temporary lifetime lower than a preset noise threshold, and write generators with a temporary lifetime not lower than the noise threshold into a persistent bar chart.

[0023] The number of zero-order homology connected branches is counted based on the persistent bar graph, and the threshold lifetime of each connected branch is weighted with the number of associated critical edges to obtain the zero-order homology scale parameter. The threshold lifetime of each first-order hole is weighted with the number of associated critical surfaces to obtain the first-order homology hole persistence parameter.

[0024] Preferably, the preset mapping rule is as follows:

[0025] The number of zero-order homophonic connected branches is used to map the diversion ratio of the cleaning branch. The larger the number of branches, the more residual permeating agent clusters there are, and the higher the proportion of the strong guiding cleaning branch.

[0026] The zero-order coherence scale parameter is used to map the cleaning intensity level. The larger the scale parameter, the larger the dust adhesion area, and the greater the processing time and shot blasting or airflow intensity of the corresponding branch.

[0027] The first-order homogeneous pore durability parameter is used to map the re-infiltration repair or scrapping determination threshold. When the pore durability exceeds the re-infiltration threshold, a re-infiltration repair command is generated; when it exceeds the scrapping threshold, a scrapping command is generated.

[0028] Preferably, the post-processing module includes:

[0029] Premium product line, used for low-pressure cleaning and passivation of workpieces with excellent surface condition;

[0030] Standard cleaning branch line is used to perform graded shot blasting or airflow dust removal according to the mapped zero-order parameters;

[0031] The strong cleaning branch is activated when the zero-order scale parameter exceeds the set value, thereby increasing the cleaning intensity and extending the processing time.

[0032] The re-infiltration repair branch is used to return workpieces with first-order pores exceeding the threshold to secondary co-infiltration or local re-infiltration.

[0033] Scrapped branch lines are used to isolate workpieces that are severely oxidized or geometrically defective.

[0034] Preferably, the system also includes a retest feedback module, which is electrically connected to the post-processing module and the sorting decision module. This module is used to re-acquire the workpiece surface image and recalculate the feature parameters after post-processing, and update the preset mapping rules and noise thresholds according to the disappearance rate of the topological features.

[0035] As a preferred option, it also includes an energy consumption coordination module, which generates a joint control curve of cleaning equipment power, cleaning flow rate and conveying cycle time based on the total number of characteristic parameters of the current batch and the number of branch queues, so that the strong cleaning branch line starts full-load operation only when the zero-order scale parameter exceeds the energy-saving trigger value.

[0036] In summary, compared with the prior art, the present invention has the following beneficial technical effects:

[0037] (1) This invention adopts a technical route that combines multimodal sensing, dynamic weighting of gray-scale scalar field, two-dimensional simple complex construction and discrete Morse simplification to realize continuous topological characterization of residual diffusion agent clusters, uneven diffusion layer and oxide isolation zone on the surface of multi-element alloy co-diffusion workpiece, effectively improving the accuracy of surface state recognition and improving the detection inaccuracy problem of existing gray-scale threshold or area determination methods.

[0038] (2) By mapping the number of zero-order cohomology connected branches, the zero-order cohomology scale parameter and the first-order cohomology hole durability parameter to the cleaning branch diversion ratio, cleaning intensity level and re-infiltration repair / scrapping judgment threshold respectively, the present invention realizes the precise correspondence control between surface topology monitoring results and post-processing execution parameters, which significantly improves the pertinence of sorting timing judgment and process parameter configuration.

[0039] (3) The improved continuous homology algorithm in this invention introduces a dynamic truncation mechanism for dust specular noise threshold during the elimination process of the boundary matrix column and incorporates the number of associated critical edge units and critical surface units into the parameter calculation. This can reduce the interference of topological generators corresponding to specular noise and improve the ability of zero-order homology scale parameters and first-order homology hole persistence parameters to characterize the distribution of real surface defects.

[0040] (4) The present invention forms a data closed loop of perception, decision-making, execution, retesting and rule updating through the retest feedback module, and dynamically generates equipment power curves based on the total batch topology using the energy consumption coordination module, which improves the stability of sorting quality while optimizing production energy consumption. Attached Figure Description

[0041] Figure 1 This is a block diagram of a production line system architecture according to an embodiment of the present invention.

[0042] Figure 2 This is a flowchart of the calibration grayscale scalar field construction and topology analysis according to an embodiment of the present invention.

[0043] Figure 3 This is an improved continuous coherence algorithm and sorting decision control flowchart according to an embodiment of the present invention. Detailed Implementation

[0044] The technical solution of the present invention will be further described in detail below through specific embodiments and in conjunction with the accompanying drawings. It should be understood that the implementation of the present invention is not limited to the following embodiments, and any modifications and / or alterations made to the present invention will fall within the protection scope of the present invention.

[0045] Example 1: Reference Figures 1 to 3 An automated unloading and sorting post-processing production line for multi-alloy co-diffusion products includes an unloading and conveying module, a multimodal sensing module, a data processing module, a sorting decision module, a sorting execution module, a post-processing module, a retest feedback module, and an energy consumption coordination module.

[0046] The unloading and conveying module is used to receive and transport workpieces after multi-element alloy co-diffusion treatment. After being unloaded from the co-diffusion furnace, the workpiece enters the unloading and conveying area, and sequentially passes through the multi-modal sensing and acquisition area and the sorting and execution area on the conveying path, and is finally introduced into the corresponding post-processing branch line.

[0047] The multimodal sensing module is set along the unloading and conveying module and is specifically configured to: physically calibrate the workpiece in the acquisition area and bind each acquisition area to the corresponding spatial coordinates to obtain the workpiece surface spatial coordinate set; set the polarization band and polarized light incident angle in the calibration area; set the sampling frequency according to the unloading and conveying speed and the spatial overlap rate of adjacent frames; acquire the workpiece surface polarization image sequence according to the set polarization band, incident angle and sampling frequency, and simultaneously acquire infrared surface temperature data and dust optical response data; write the acquired data into the corresponding spatial coordinates and workpiece spatial domain identifier to obtain the unloading sensing sequence.

[0048] In Example 1, the polarization band is set to 520nm to 560nm, the incident angle is adaptively adjusted according to the local curvature of the workpiece surface and stabilized in the range of 46° to 59°, the sampling frequency is set to 18Hz to 24Hz according to the transmission rate, and the spatial overlap rate of adjacent frames is maintained at more than 76% to ensure that a single workpiece is continuously and stably acquired.

[0049] The data processing module is electrically connected to the multimodal sensing module and includes a scalar field construction unit, a topology analysis unit, and a continuous coherence unit, which are used to extract characteristic parameters that characterize the connectivity state of residual penetrant on the workpiece surface and the uniformity of the penetrant layer.

[0050] The scalar field building unit is specifically configured as follows:

[0051] The unloading sensing sequence is separated into single-piece area images according to the workpiece spatial domain identifier; the high brightness response peak, low brightness response valley and stable grayscale component are extracted from the continuous multi-frame images; the polarization degree, polarization angle and diffuse reflection component are calculated to obtain the physical parameter sequence, where: polarization degree = (high brightness response peak - low brightness response valley) / (high brightness response peak + low brightness response valley), polarization angle = the direction corresponding to the high brightness response peak, and diffuse reflection component = stable grayscale component;

[0052] Based on the geometric curvature of the unloading area and the surface curvature of the workpiece, the polarization degree, polarization angle, and diffuse reflection components in the physical parameter sequence are differentially weighted to obtain the enhancement coefficient matrix; the corresponding pixel grayscale scalar values ​​are written into the enhancement coefficient matrix to obtain the calibration grayscale scalar field.

[0053] In Example 1, a higher polarization degree weight is assigned to the bright dust area, a higher color difference weight is assigned to the dark uneven area of ​​the infiltrated layer, and a higher polarization angle weight is assigned to the oxidation area; the polarization degree, polarization angle, and diffuse reflection component weights of the equipment surface area are set to 0.45, 0.30, and 0.25, respectively, and the corresponding weights of the pipeline / irregular part area are set to 0.40, 0.34, and 0.26.

[0054] The topology analysis unit is specifically configured as follows:

[0055] Each pixel in the calibration grayscale scalar field is mapped to a vertex cell, and the grayscale scalar value of each pixel is written into the corresponding vertex cell to obtain a set of vertex cells;

[0056] Within the same spatial domain, every four consecutive vertex units that form an orthogonal pixel grid are defined as a grid unit. The diagonal segmentation direction is determined based on the gradient direction of the calibrated grayscale scalar field within the grid unit and the geometric curvature characteristics of the corresponding spatial coordinate set. The grid unit is then divided into two triangular regions along the determined diagonal, each containing three adjacent vertex units, to obtain the triangulation result.

[0057] Based on the triangulation results, edge elements and face elements are established, and the relationships between edge elements and vertex elements, and between face elements and edge elements are established to obtain a simple associative structure.

[0058] In a simple associative structure, the smaller value of the pixel grayscale scalar value of the two vertex units that constitute the same side unit is written into the side unit, and the smaller value of the pixel grayscale scalar value of the three vertex units that constitute the same face unit is written into the face unit, thus obtaining the grayscale scalar value of the side unit and the grayscale scalar value of the face unit.

[0059] By combining the vertex cell set, the edge cell grayscale scalar value, and the face cell grayscale scalar value, a weighted two-dimensional simplex is obtained.

[0060] In Example 1, each frame's effective region forms approximately 50,000 vertex units, 99,000 edge units, and 49,000 face units. The scalar values ​​of the edge units and face units are all taken as the smaller values ​​among the scalar values ​​that constitute the vertex units, in order to ensure the downward closure property of the scalar field on the simplex complex.

[0061] The topology analysis unit is further configured to eliminate non-critical units based on the discrete Morse vector field pairing rule, and the output critical unit retains the complex shape, specifically:

[0062] In a weighted two-dimensional simplex, the two vertex units corresponding to each edge unit and the edge units corresponding to each face unit are retrieved based on the simple correlation structure to obtain the correlation unit group;

[0063] In the associated unit group, the vertex and edge whose scalar value of the edge unit is the same as the smaller scalar value of the two vertices constituting the edge are identified as vertex-edge candidate pairing units, and the edge and face whose scalar value of the face unit is the same as the smaller scalar value of the edge constituting the face are identified as edge-face candidate pairing units, thus obtaining the candidate pairing set;

[0064] In the candidate pairing set, a directed graph is dynamically constructed according to the descent relationship of the gray scalar value corresponding to the gradient of the scalar field, and loop detection is performed on each candidate pairing. Candidate pairs that generate closed alternating paths are discarded, and candidate pairs that do not generate closed alternating paths and satisfy the one-to-one pairing relationship are retained, thus obtaining a discrete Morse vector field.

[0065] Vertex, edge, and face units entering the discrete Morse vector field are identified as non-critical units, while those not entering the pairing relationship are identified as critical units. Alternating gradient paths connecting critical units of different dimensions are traced in the weighted two-dimensional simplex to obtain a set of alternating gradient paths. The boundary operator of the Morse complex is reconstructed based on the set of alternating gradient paths, and the algebraic connection relationship between critical units is re-established to obtain a critical unit-preserving complex composed of critical units and algebraic connection relationships.

[0066] The continuous cohomology unit is specifically configured to execute an improved continuous cohomology algorithm, including:

[0067] S1, extract critical units and retain the gray scalar values ​​corresponding to each critical unit in the complex. Arrange the scalar values ​​from largest to smallest into multiple thresholds, and retain the critical units whose gray scalar values ​​are not less than the current threshold and the algebraic connection relationships between critical units under each threshold to obtain the multi-threshold filtering sequence and continuous filtering results.

[0068] S2, Based on the continuous filtering results and the Morse complex boundary operator, construct the boundary matrix in the continuous filtering results corresponding to each threshold, and perform column elimination on the boundary matrix to obtain the generation threshold and the elimination threshold of the homology generator.

[0069] S3, during the elimination of columns in the boundary matrix, the temporary lifetime of the cohomogeneous generator corresponding to the current column vector is calculated in real time. Column vectors with a temporary lifetime lower than the preset dust specular noise threshold are eliminated and removed. Column vectors with a temporary lifetime not lower than the noise threshold are eliminated and written into the persistent bar chart to obtain the denoised persistent bar chart.

[0070] S4. Count the number of zero-order homology connected components in the denoised persistent bar chart, and extract the difference between the generation threshold and the extinction threshold corresponding to each zero-order homology connected component, the number of critical edge units associated with each zero-order homology connected component, the difference between the generation threshold and the extinction threshold corresponding to each first-order homology hole, and the number of critical surface units associated with each first-order homology hole to obtain the parameter calculation set.

[0071] S5, the difference between the generation threshold and the destruction threshold corresponding to each zero-order homophonic connected component and the number of associated critical edge units are weighted and fused to obtain the zero-order homophonic scale parameter; the difference between the generation threshold and the destruction threshold corresponding to each first-order homophonic hole and the number of associated critical surface units are weighted and fused to obtain the first-order homophonic hole persistence parameter.

[0072] Wherein, the temporary lifetime of the homogeneous generator = generation threshold - current extinction threshold; the dust specular noise threshold is used to dynamically truncate topological generators generated by short-lived noise such as metallic specular highlights and dust flicker. In Example 1, a 24-level multi-threshold filtering sequence is constructed, with a maximum number of columns in the boundary matrix of 2048. On average, about 1700 columns of elimination are completed per frame, and column vectors with a temporary lifetime lower than the noise threshold of 0.05 are truncated by an average of about 120 columns.

[0073] The sorting decision module is electrically connected to the data processing module. It receives the number of zero-order cohomological connected branches, the zero-order cohomological scale parameter, and the first-order cohomological hole persistence parameter. Based on a preset mapping rule, it generates sorting instructions and corresponding post-processing parameters. The preset mapping rule is as follows:

[0074] The number of zero-order homophonic connected branches is used to map the diversion ratio of the cleaning branch. The larger the number of branches, the more residual permeating agent clusters there are, and the higher the proportion of the strong directional cleaning branch.

[0075] The zero-order coherence scale parameter is used to map the cleaning intensity level. The larger the scale parameter, the larger the dust adhesion area, and the greater the processing time of the corresponding branch and the intensity of shot blasting or airflow dust removal.

[0076] The first-order coherent pore durability parameter is used to map the re-infiltration repair or scrapping judgment threshold. When the pore durability exceeds the re-infiltration threshold, a re-infiltration repair instruction is generated; when it exceeds the scrapping threshold, a scrapping instruction is generated.

[0077] Surface temperature and infrared radiation intensity are used to correct the cleaning and cooling cycle. The higher the temperature, the greater the flow rate and cooling parameters of the cleaning branch.

[0078] In Example 1, the sorting category determination logic is as follows: if the number of zero-order connected branches ≤ N1 and the zero-order size ≤ S1 and the first-order pores ≤ H1, it is a superior product (only low-pressure cleaning and passivation); if the zero-order parameters are medium, it is a standard cleaned product; if the zero-order clusters are large (size > S2), it is a product requiring strong cleaning and rework; if the first-order pore durability reflects local non-permeability, exposed substrate, and the number of associated surfaces exceeds the repermeability value, it is a repermeability repair product; if it exceeds the scrap value, it is a scrapped product.

[0079] The sorting execution module is located at the end of the unloading conveyor module and is electrically connected to the sorting decision module. It responds to sorting commands by diverting workpieces to the corresponding processing channels. The post-processing module includes multiple parallel post-processing branches, specifically:

[0080] Premium product line, used for low-pressure cleaning and passivation of workpieces with excellent surface condition;

[0081] Standard cleaning branch line is used to perform graded shot blasting or airflow dust removal according to the mapped zero-order parameters;

[0082] The strong cleaning branch is activated when the zero-order scale parameter exceeds the set value, thereby increasing the cleaning intensity and extending the processing time.

[0083] The re-infiltration repair branch is used to return workpieces with first-order pores exceeding the threshold to secondary co-infiltration or local re-infiltration.

[0084] Scrapped branch lines are used to isolate workpieces that are severely oxidized or geometrically defective.

[0085] Each branch line receives process parameters issued by the sorting decision module and calls back the operating status in real time.

[0086] The retest feedback module is electrically connected to the post-processing module and the sorting decision module, and is specifically configured as follows:

[0087] After processing, the re-photographed image of the workpiece is reconstructed, the grayscale scalar field is calibrated, and the persistent bar chart is recalculated. High-persistency topological features are tracked according to continuous sampling time. The topological feature decay rate per unit time is calculated (feature decay amount = difference between the generation threshold and the decay threshold at the previous sampling time - the corresponding difference at the current sampling time; time decay rate = feature decay amount / unit sampling interval).

[0088] If the retest of the superior grade branch still retains first-order pores exceeding the retest allowable value, then the batch is returned to the strong cleaning or re-infiltration branch; if the zero-order connected branch demise rate of the strong cleaning branch retest meets the nonlinear decay condition for multiple consecutive sampling times, then the cleaning intensity is reduced to form a flexible degradation instruction; and the mapping rules and continuous homogeneous noise threshold are incrementally updated using the retest deviation samples.

[0089] The energy consumption coordination module is used to generate a joint control curve of cleaning equipment power, cleaning flow rate and conveying cycle time based on the total zero-order parameter, total first-order parameter and branch queue number of the current batch. This allows the strong cleaning branch to start full-load operation only when the zero-order scale parameter exceeds the energy-saving trigger value, while the remaining batches operate at reduced power according to the mapped parameters, thereby reducing production energy consumption while ensuring the treatment effect.

[0090] Example 2: To verify the feasibility of this invention in practice, the production line of this invention was deployed in a continuously operating multi-alloy co-infiltration (zinc-aluminum-rare earth) fastener post-processing scenario. Online monitoring and closed-loop sorting control were implemented for five key batches, with data collection covering the planar areas of standard parts, threaded areas, and the surfaces of irregularly shaped parts. In this scenario, existing solutions often employ manual inspection combined with fixed-cycle shot blasting, relying primarily on surface brightness, dust area, and experience time for judgment. This makes it difficult to distinguish between the metallic highlights and dust adhesion on the co-infiltrated surface, and it cannot continuously characterize the dust connectivity and the porous structure of the infiltrated layer, resulting in cleaning timing deviations, insufficient removal of stubborn clumps, and repeated over-cleaning of local surfaces. To address these issues, this example employs a multimodal sensing module, a data processing module, a sorting decision module, a sorting execution module, a post-processing module, a retest feedback module, and an energy consumption coordination module to form a complete processing chain.

[0091] During implementation, the multimodal sensing module establishes acquisition areas along the unloading line, with each acquisition area corresponding to an effective field of view of 320×240 pixels and a polarization band set to 540nm. The scalar field construction unit performs spatial domain separation on the unloading sensing sequence, forming individual region image sequences, and then determines the degree of polarization, polarization angle, and diffuse reflection component to obtain a physical parameter sequence. Based on the workpiece surface curvature, the physical parameter sequence is weighted differently to form an enhancement coefficient matrix and written into the pixel grayscale scalar values ​​to obtain a calibration grayscale scalar field. The topology analysis unit assigns each pixel in the calibration grayscale scalar field as a vertex unit. On every four consecutive orthogonal grid vertices, the diagonal segmentation direction is determined according to the scalar field gradient direction and spatial curvature, dividing each grid unit into two triangular regions. On average, each individual image generates approximately 50,000 vertex units, 99,000 edge units, and 49,000 face units. The continuous cohomology unit extracts scalar values ​​from the complex shape of the critical unit and constructs a 24-level multi-threshold filtering sequence. It then constructs a boundary matrix for each threshold and performs column elimination. After multiple rounds of continuous retesting and iteration, the denoised persistent bar graph stably outputs the number of zero-order cohomological connected branches, the zero-order cohomological scale parameter, and the first-order cohomological porosity parameter. The sorting decision module, based on preset mapping rules, converts the zero-order parameters into shot blasting current (0.5A-2.0A) and cleaning time (30s-120s), and converts the first-order parameters into re-infiltration repair / discarding thresholds.

[0092] Table 1 is a comparison table of system output and actual test results for key batches. The “actual test value” is the reference value obtained by offline manual verification and reference testing equipment, and the “system output value” is the corresponding parameter value output in real time by the production line of this invention.

[0093] Table 1 Comparison of System Output and Actual Measured Results for Key Batches

[0094] B01 8 8 0.88 0.91 0.45 0.47 98.5 B02 15 14 1.32 1.28 0.82 0.85 97.2 B03 5 5 0.61 0.63 0.21 0.20 99.1 B04 22 23 1.78 1.74 1.05 1.02 96.8 B05 11 11 1.05 1.08 0.63 0.65 98.0

[0095] As shown in Table 1, the number of zero-order homology connected branches output by the system is basically consistent with the offline verification results, and the absolute error of all five batches does not exceed 1. The average relative error of the zero-order scale parameter is 2.6%, and the average relative error of the first-order homology hole persistence parameter is 3.1%, indicating that the improved continuous homology algorithm has high stability in preserving surface connectivity and hole structure. Batch B04 corresponds to the interface between the thread and the irregular part, where the morphology of dirt and seepage layer is the most complex. The system output is still close to the verification results, indicating that simple complex construction and discrete Morse simplification did not destroy the key topology.

[0096] From the perspective of execution parameters, the sorting instructions based on topology parameter mapping can be stably executed by each post-processing branch. Compared with the previous fixed-cycle shot blasting and fixed-intensity cleaning method used on the same production line, in continuous multi-round sorting operations, the average energy consumption per piece processed in this embodiment is reduced by about 18% from the baseline value, the average processing time per piece is reduced, the rework rate is reduced from 7.3% to 2.1%, and the residual dust coverage and uneven penetration detection rate after post-processing are improved, indicating that there is a good correspondence between the sorting instructions output by the system and the topological state of the workpiece surface.

[0097] The results of the retesting during the sorting process show that the high-durability topological feature time series can continuously reflect the process of residual penetrant peeling and penetrant layer defect re-inspection. In the early stage of the retesting, the time decay rate of the high-durability topological features remained in a high range, indicating that the main body of the residual penetrant clusters was still rapidly decreasing. In the middle and late stages of the retesting, the time decay rate showed a downward trend, indicating that large-area dust agglomerates had been broken up, leaving only localized attachment areas. When nonlinear decay results appeared, the system switched to a flexible degradation mode, and the cleaning intensity was reduced on average. Ultimately, no surface penetrant layer scratches or localized over-cleaning phenomena occurred. Based on the above process data and the results in Table 1, it can be verified that the present invention has complete, stable, and repeatable implementation capabilities in terms of online surface topological characterization, adaptive generation of sorting parameters, execution of retesting, and flexible degradation switching in the post-processing unloading and sorting of multi-element alloy co-diffusion products.

[0098] As a further extension of the present invention, this embodiment provides a principle explanation of the technical effect of the noise threshold truncation mechanism in the "improved continuous cohomology algorithm". In the surface image of a co-infiltrated workpiece, metallic highlights, dust flashes, and local reflections generate a large number of short-lived, small-scale topology generators. If these generators are not truncated, they will directly increase the statistical value of the number of zero-order connected branches and raise the zero-order scale parameter, thereby causing workpieces that should be judged as superior or standard cleaned products to be mistakenly diverted to strong cleaning branches, increasing unnecessary energy consumption and the risk of over-cleaning. The present invention calculates the temporary lifetime in real time during the elimination process of the boundary matrix column and dynamically terminates the elimination and writing of short-lived generators based on a preset dust highlight noise threshold. Only generators with a temporary lifetime not lower than the threshold are written into the persistent bar chart to participate in subsequent parameter calculations, thereby effectively suppressing the interference of highlight noise on topology parameters. This mechanism, combined with discrete Morse simplification, reduces the data processing dimensionality and online computational burden while retaining the key critical units that characterize the real residual permeate clusters and permeate layer pores, thus providing algorithmic feasibility for high frame rate online sorting in the unloading process.

[0099] Those skilled in the art should understand that the modules, units, and algorithm steps of this invention can be implemented in software, hardware, or a combination thereof; the terms "module" and "unit" refer to functional units with corresponding data processing or control functions, which can correspond to independent physical devices or different software functional modules or program code segments within the same processing device. The polarization bands, weights, thresholds, number of units, sampling frequencies, and batch data mentioned in the specification are specific exemplary values ​​to aid in understanding the embodiments of this invention, and are not intended to limit the scope of protection of this invention; in practical applications, these parameters can be adaptively adjusted according to the workpiece material, co-infiltration formula, production line cycle time, and equipment conditions.

[0100] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products, characterized in that, include: The unloading and conveying module is used to receive and transport workpieces after multi-alloy co-diffusion treatment; A multimodal sensing module is installed along the unloading and conveying module to collect surface state data and physical feature data of the workpiece flowing through it. The data processing module is electrically connected to the multimodal sensing module and is used to process the surface state data and physical feature data to extract feature parameters that characterize the connectivity state of residual infiltration agent on the workpiece surface and the uniformity of the infiltration layer. The sorting decision module is electrically connected to the data processing module and is used to receive the feature parameters and generate sorting instructions and corresponding post-processing parameters according to preset mapping rules. The sorting execution module is located at the end of the unloading and conveying module and is electrically connected to the sorting decision module. It is used to respond to the sorting command and divert the workpieces to the corresponding processing channels. The post-processing module includes multiple parallel post-processing branches for receiving the diverted workpieces and adjusting the operating conditions according to the post-processing process parameters.

2. The automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products according to claim 1, characterized in that, The data processing module includes: The scalar field construction unit is used to optically enhance the surface state data, extract pixel-level diffusion layer color difference, residual dust brightness and oxidation highlight component, and generate a calibrated grayscale scalar field. The topology analysis unit is used to construct a two-dimensional simple complex based on pixel adjacency relationships, and eliminate non-critical units based on the pairing rules of the discrete Morse vector field, while outputting critical units that preserve the complex. A persistent cohomology unit is used to perform persistent cohomology calculations on the critical unit while preserving the complex, and outputs the number of zero-order cohomology connected branches, the zero-order cohomology scale parameter, and the first-order cohomology hole persistence parameter. The zero-order coherent connectivity branch number and zero-order coherent scale parameter are used to characterize the connectivity zone between residual infiltrate clusters and dust adhesion, while the first-order coherent pore durability parameter is used to characterize the discontinuous or oxidized isolation zone of the infiltrate layer.

3. The automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products according to claim 2, characterized in that, The continuous co-homing unit executes the improved continuous co-homing algorithm, including: Extract the critical units and retain the scalar values ​​corresponding to each critical unit in the complex. Arrange them into multiple thresholds according to their size. Under each threshold, retain the critical units and algebraic connection relationships that are not less than the current threshold to obtain a multi-threshold filtering sequence. Construct a boundary matrix and perform column elimination. During the elimination process, calculate the temporary lifetime of the homology generators, remove generators with a temporary lifetime lower than a preset noise threshold, and write generators with a temporary lifetime not lower than the noise threshold into a persistent bar chart. The number of zero-order homology connected branches is counted based on the persistent bar graph, and the threshold lifetime of each connected branch is weighted with the number of associated critical edges to obtain the zero-order homology scale parameter. The threshold lifetime of each first-order hole is weighted with the number of associated critical surfaces to obtain the first-order homology hole persistence parameter.

4. The automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products according to claim 1, characterized in that, The preset mapping rule is as follows: The number of zero-order homophonic connected branches is used to map the diversion ratio of the cleaning branch. The larger the number of branches, the more residual permeating agent clusters there are, and the higher the proportion of the strong cleaning branch. The zero-order coherence scale parameter is used to map the cleaning intensity level. The larger the scale parameter, the larger the dust adhesion area, and the greater the processing time and shot blasting or airflow intensity of the corresponding branch. The first-order homogeneous pore durability parameter is used to map the re-infiltration repair or scrapping determination threshold. When the pore durability exceeds the re-infiltration threshold, a re-infiltration repair command is generated; when it exceeds the scrapping threshold, a scrapping command is generated.

5. The automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products according to claim 1, characterized in that, The post-processing module includes: Premium product line, used for low-pressure cleaning and passivation of workpieces with excellent surface condition; Standard cleaning branch line is used to perform graded shot blasting or airflow dust removal according to the mapped zero-order parameters; The strong cleaning branch is activated when the zero-order scale parameter exceeds the set value, thereby increasing the cleaning intensity and extending the processing time. The re-infiltration repair branch is used to return workpieces with first-order pores exceeding the threshold to secondary co-infiltration or local re-infiltration. Scrapped branch lines are used to isolate workpieces that are severely oxidized or geometrically defective.

6. The automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products according to claim 1, characterized in that, Also includes: The retest feedback module is electrically connected to the post-processing module and the sorting decision module. It is used to re-acquire the workpiece surface image and recalculate the feature parameters after the post-processing is completed, and update the preset mapping rules and noise threshold according to the disappearance rate of the topological features.

7. The automated unloading and sorting post-processing production line for multi-element alloy co-diffusion products according to claim 1, characterized in that, Also includes: The energy consumption coordination module is used to generate a joint control curve of cleaning equipment power, cleaning flow rate and conveying cycle time based on the total characteristic parameters of the current batch and the number of branch queues, so that the strong cleaning branch line starts full-load operation only when the zero-order scale parameter exceeds the energy-saving trigger value.

Images