Multi-robot quality inspection optimization method and system based on distributed collaboration and knowledge evolution

Abstract

The invention discloses a multi-robot quality inspection optimization method and system based on distributed collaboration and knowledge evolution, and relates to the technical field of multi-robot collaborative quality inspection and distributed quality control. Standardized fragmentation management and cross-node consistency constraint of multi-robot node quality inspection data are realized, and quality inspection result deviation caused by node acquisition difference, processing link abnormity and time slice mismatching can be effectively reduced; through a feature difference degree and processing sequence integrity double-index collaborative judgment mechanism, abnormal fragments can be accurately identified and fine-grained re-division processing is triggered, so that the transitivity amplification of local anomalies to an overall quality inspection conclusion is avoided, and the stability and credibility of a quality inspection result are improved.

Description

Technical Field

[0001] This invention relates to the field of multi-robot collaborative quality inspection and distributed quality control technology, and in particular to a multi-robot quality inspection optimization method and system based on distributed collaboration and knowledge evolution. Background Technology

[0002] With the rapid development of intelligent manufacturing and flexible production models, industrial quality inspection is gradually evolving from traditional manual sampling to robotic, automated, and intelligent processes. Automated quality inspection systems based on robot vision, sensor fusion, and machine learning have become crucial technological supports for improving product consistency and yield in discrete manufacturing, equipment manufacturing, and process industries. In recent years, multi-robot collaborative quality inspection has gradually become a research hotspot, achieving full-coverage inspection of complex, large, or continuous production objects through multi-node parallel data acquisition, distributed computing, and collaborative decision-making. However, with the expansion of collaborative scale and the increase in the complexity of the inspection objects, quality inspection data exhibits significant spatiotemporal heterogeneity, processing sequence differences, and inconsistencies in multi-source features. Existing quality inspection systems still face significant technical bottlenecks in areas such as distributed consistency verification, collaborative stability assessment, and cross-cycle knowledge reuse.

[0003] CN115468560A discloses a quality inspection method, robot, equipment, and medium based on multi-sensor information fusion. By fusing environmental information acquired from visual sensors and navigation information acquired from navigation sensors, it achieves the identification, localization, and enhanced display of target objects in the quality inspection area. This technical solution focuses on the fusion processing of multi-sensor data within a single robot node and the improvement of target localization accuracy, with the core being the optimization of sensor fusion and pose estimation algorithms. However, this comparative technology lacks a unified segmentation management mechanism and consistency comparison strategy for quality inspection data across nodes and time slices; furthermore, in complex production environments, when the acquisition quality of different robot nodes is uneven, data is lost, or the processing link is abnormal, it is difficult to form a stable and evolvable collaborative quality inspection closed loop, resulting in insufficient collaborative robustness and long-term stability.

[0004] CN110866901A discloses a quality inspection method and system based on edge computing technology. This method uploads multiple quality inspection images to an edge server and utilizes a deep convolutional neural network for quality judgment, enabling rapid feedback on production quality. This technical solution addresses the real-time performance and computational load offloading issues in single production lines or local workstation scenarios, playing a positive role in improving quality inspection efficiency. However, this comparative technology still belongs to a typical "centralized intelligent discrimination model." It lacks a unified data sharding structure and cross-node consistency evaluation mechanism designed for multi-robot node parallel collaborative quality inspection scenarios. It also lacks the ability to model and utilize spatiotemporal sharding, the integrity of intra-node processing sequences, and stable consensus across rounds. Furthermore, the quality inspection model focuses on the output of single-round discrimination results and does not establish a knowledge evolution mechanism based on historical successful execution records. This makes it difficult to reuse and self-optimize stable quality inspection consensus formed during long-term operation, and thus fails to meet the needs of continuously evolving collaborative quality inspection in large-scale flexible production environments. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the specification abstract and the title of the invention, to avoid obscuring the purpose of this section, the specification abstract, and the title of the invention. Such simplifications or omissions shall not be used to limit the scope of the invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by this invention is that existing robot quality inspection technology is difficult to support the comprehensive needs of detailed processing of anomalies, consensus formation and continuous self-optimization of quality inspection strategies in large-scale multi-robot parallel quality inspection scenarios.

[0008] To address the aforementioned technical problems, this invention provides the following technical solution: a multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution, comprising: multiple robot nodes collecting raw data of the same quality inspection object in parallel; the raw data being raw point cloud data; each robot node dividing the raw data into quality inspection data slices based on a pre-loaded space-time mapping table, extracting the quality feature vector of each quality inspection data slice, and generating a slice processing trajectory identifier containing a spatial unit identifier, spatial boundary, time slice number, node processing sequence number, and quality feature vector, forming a slice set and a trajectory identifier set; a collaborative node receiving all sets, grouping them according to the spatial unit identifier and time slice number, and processing the data accordingly. Within the same group, quality feature vectors are compared for consistency, and feature difference is calculated. Simultaneously, the continuity of processing sequence numbers within each robot node is checked to determine the completeness of the processing sequence, generating fragment consistency analysis results. Collaborative nodes filter quality inspection data fragments with feature difference exceeding the difference threshold or sequence completeness below the completeness threshold, perform local re-division based on spatial boundaries, generate fine-grained quality inspection data sub-fragments, and update the trajectory identifier set. Collaborative nodes monitor and analyze the results in multiple rounds of collaborative processing, and aggregate quality inspection data fragments that continuously meet stability conditions into a consensus quality inspection fragment set, which is then distributed to each robot node along with the updated trajectory identifiers to optimize the quality inspection strategy for subsequent tasks.

[0009] As a preferred embodiment of the present invention, the generation of the preloaded space-time mapping table includes: in the initial stage, each robot node retrieves historical quality inspection data fragment records of the same type of quality inspection object from storage; based on the baseline three-dimensional model of the object to be inspected, the detection space is divided into non-overlapping basic spatial units using a cubic mesh with a side length of L millimeters, and each basic spatial unit is assigned a unique spatial unit identifier; the frequency of each basic spatial unit being covered by data fragments in the historical quality inspection data fragment records is counted to generate a spatial region fragment frequency statistics table; a primary threshold and a secondary threshold for fragment frequency are set; basic spatial units with frequencies higher than the primary threshold are marked as first-class spatial units, and those lower than the secondary threshold are marked as second-class spatial units; the classification results and the set of basic spatial units are preloaded to each robot node.

[0010] As a preferred embodiment of the present invention, the generation of the preloaded space-time mapping table further includes: each robot node analyzing historical quality inspection data fragment records, extracting the timestamp sequences of all historical data fragments falling into the classified first-class and second-class spatial units respectively; calculating the arithmetic mean of the intervals between adjacent timestamps in each category of timestamp sequences, filtering out outliers exceeding n times the average interval, and recalculating to obtain the average data acquisition time intervals Δt1 and Δt2 respectively; setting Δt1 as the time slice length L1 corresponding to the first-class spatial unit, and setting Δt2 as the time slice length L2 corresponding to the second-class spatial unit; forming a preloaded space-time mapping table consisting of spatial unit identifier, spatial unit category, and corresponding time slice length, and distributing it to each robot node.

[0011] As a preferred embodiment of the present invention, the following steps are included: The calculation of feature difference includes: collaborative nodes grouping quality inspection data fragments with the same spatial unit identifier and time slice number into the same comparison group; for each comparison group, extracting the quality feature vectors from the processing trajectory identifiers of all fragments within the group, calculating the Euclidean distance between any two feature vectors, and taking the median of all distances as the feature difference within the group; the determination of processing sequence integrity includes: for each comparison group, grouping the quality inspection data fragments according to the robot node identifier; for each robot node's corresponding fragment subgroup: extracting the intra-node processing sequence numbers of all quality inspection data fragments in the fragment subgroup, arranging them in ascending order of value to obtain sequence S; performing continuity analysis on sequence S: if the number of sequence numbers in sequence S is less than 2, the sequence integrity is recorded as 1; otherwise, calculating the interval between adjacent sequence numbers; if all intervals are 1, the sequence integrity is 1; if there are intervals greater than 1, calculating the abnormal interval coefficient = (sum of intervals greater than 1) / (The difference between the maximum and minimum sequence number in sequence S), the sequence completeness is 1 minus the abnormal interval coefficient, and the minimum value is 0; take the minimum value of the sequence completeness of all robot nodes as the sequence completeness of the comparison group.

[0012] As a preferred embodiment of the present invention, the fragment consistency analysis results include: alignment group identifier, feature difference degree, sequence integrity, feature dimension identifier with the largest difference, node identifier of processed sequence number within missing node, and list of missing sequence numbers.

[0013] As a preferred embodiment of the present invention, the local re-division based on spatial boundaries includes: the collaborative node filters out the fragments to be re-divided based on the fragment consistency analysis results; sends a re-division request to the robot node that collects the fragments to be re-divided; wherein the re-division request includes: the trajectory identifier of the fragment to be re-divided, and the spatial boundaries of all other quality inspection data fragments within the same time slice number; after receiving the re-division request, the robot node obtains the corresponding original point cloud data from local storage; and performs the following spatial relationship determination in sequence: if the overlap volume ratio between the spatial boundary and any other quality inspection data fragment spatial boundary exceeds the overlap threshold, it is determined to be data redundancy, and a data merging operation is performed; if the spatial boundary and all other quality inspection data fragment spatial boundaries are not redundant, the data is determined to be redundant, and a data merging operation is performed; if the spatial boundary and any other quality inspection data fragment spatial boundary are ... If the minimum distance between the spatial boundaries of the data fragments is greater than the gap threshold, the fragments are identified as isolated fragments and marked as abnormal. If the minimum distance between the spatial boundary and at least one other quality inspection data fragment spatial boundary is greater than 0 and less than or equal to the gap threshold, or the overlap volume ratio is greater than 0 but less than the overlap threshold, the fragments need to be further divided. If further division is required, the following operations are performed: within the point cloud data to be further divided, the optimal segmentation plane is found using a RANSAC-based plane detection algorithm; the fragment to be further divided is divided into two or more sub-fragments along the segmentation plane; the sub-fragments are pruned and repackaged, and the trajectory identifiers are updated; the segmented sub-fragment data, the updated trajectory identifiers, and the further division analysis results are sent back to the collaborative node.

[0014] As a preferred embodiment of the present invention, the continuous satisfaction of the stability condition means that: in multiple rounds of collaborative processing, the feature difference degree of the same quality inspection data shard is always lower than the feature difference degree stability threshold, and the sequence integrity degree is always higher than the sequence integrity degree stability threshold; the generation of the consensus quality inspection shard set includes: the collaborative node clusters all quality inspection data shards that satisfy the stability condition according to the spatial unit identifier; for each quality inspection data shard in the spatial unit, extracts the quality feature vector, calculates the arithmetic mean as the aggregated feature vector; calculates the Euclidean distance between each shard feature and the aggregated feature, and classifies the quality inspection data shards whose distance is less than the aggregation threshold into one class; merges the quality inspection data shards classified into one class and the corresponding optimized shard processing trajectory identifier into the consensus quality inspection shard set, and distributes it to each robot node as a reliable knowledge template.

[0015] On the other hand, the present invention also provides a multi-robot quality inspection optimization system based on distributed collaboration and knowledge evolution, comprising: The raw data acquisition module is used for multiple robot nodes to acquire raw data of the same quality inspection object in parallel. The slicing and mapping module is used by each robot node to divide the raw data into quality inspection data slices based on a preloaded space-time mapping table, extract the quality feature vector of each quality inspection data slice, and generate slicing processing trajectory identifiers containing spatial unit identifiers, time slice numbers, node processing sequence numbers and quality feature vectors, forming a slice set and a trajectory identifier set. The feature extraction module is used to receive all sets of data from the collaborative nodes, group them according to spatial unit identifiers and time slice numbers, perform consistency comparisons on the quality feature vectors within the same group, calculate the feature difference degree, and simultaneously check the continuity of processing sequence numbers within different robot nodes to determine the integrity of the processing sequence and generate slice consistency analysis results. The trajectory generation module is used to collaboratively filter quality inspection data fragments with feature differences higher than the first threshold or sequence integrity lower than the second threshold, perform local re-division based on spatial boundaries, generate fine-grained quality inspection data sub-fragments, and update the trajectory identifier set. The sharding and aggregation module is used by collaborative nodes to monitor and analyze the results in multiple rounds of collaborative processing. It shards and aggregates the quality inspection data that continuously meets the stability conditions into a consensus quality inspection shard set, and distributes it to each robot node along with the updated trajectory identifier to optimize the quality inspection strategy for subsequent tasks.

[0016] The beneficial effects of this invention are as follows: By constructing a distributed collaborative quality inspection mechanism based on space-time mapping, this invention achieves standardized fragment management and cross-node consistency constraints for quality inspection data from multiple robot nodes, effectively reducing the deviation of quality inspection results caused by differences in node acquisition, abnormal processing links, and mismatched time slices. Through a dual-indicator collaborative judgment mechanism of feature difference degree and processing sequence integrity, abnormal fragments can be accurately identified and trigger fine-grained re-segmentation processing, thereby avoiding the transmissive amplification of local anomalies on the overall quality inspection conclusion and improving the stability and reliability of the quality inspection results.

[0017] Meanwhile, by constructing a consensus quality inspection shard set under multi-round collaborative processing, the system can continuously accumulate and reuse highly stable and consistent quality inspection results, forming an evolvable quality inspection knowledge template for dynamically optimizing the quality inspection strategy of subsequent tasks, thus significantly improving the system's adaptability and long-term operational accuracy.

[0018] Therefore, this invention not only improves the robustness and reliability of multi-robot collaborative quality inspection, but also enhances the continuous evolution capability and engineering applicability of the quality inspection system in complex production environments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of the multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution shown in this invention; Figure 2 This is a structural diagram of the multi-robot quality inspection optimization system based on distributed collaboration and knowledge evolution, as shown in this invention. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution, comprising: S1: Multiple robot nodes collect raw data of the same quality inspection object in parallel; the raw data is raw point cloud data.

[0024] Each robot node is equipped with a high-precision 3D scanning device, including but not limited to a combination of line laser scanner, structured light 3D camera, time-of-flight depth camera or lidar.

[0025] Each robot node forms a star or mesh network topology with the collaborating nodes through a wireless communication module. The communication protocol adopts Ethernet or 5G NR-U technology based on time-sensitive networking to ensure real-time data transmission and low latency.

[0026] Each robot node has a built-in synchronization clock module that supports the IEEE 1588 Precision Time Protocol (PTP), enabling sub-millisecond time synchronization between nodes.

[0027] Each robot node autonomously plans the optimal observation pose based on the assigned acquisition sub-region, ensuring that the angle between the sensor optical axis and the normal of the measured surface is less than the maximum allowable angle (e.g., <45°).

[0028] Upon receiving a global trigger signal from the coordinating node, all robot nodes synchronously begin data acquisition. During the acquisition process, each robot node continuously acquires point cloud frames, with each frame of point cloud data appended with an acquisition timestamp, node identifier, and sensor intrinsic and extrinsic parameter calibration matrix. The synchronization error of the timestamp is less than 1 / 10 of the length of a single time slice.

[0029] It should be noted that by using multiple robot nodes for parallel data acquisition, the time required for traditional single-robot sequential acquisition is reduced to 1 / N of the original time (where N is the number of robots). This is especially effective for large and complex components, where the acquisition efficiency can be improved by 5-10 times. Furthermore, multi-view collaborative coverage effectively reduces blind spots caused by occlusion under a single viewpoint, enabling the point cloud coverage of the measured object surface to reach over 99%. Synchronous acquisition avoids multi-frame point cloud registration errors caused by object displacement or deformation, thus improving data consistency.

[0030] S2: Based on the preloaded space-time mapping table, each robot node divides the raw data into quality inspection data slices, extracts the quality feature vector of each quality inspection data slice, and generates slice processing trajectory identifiers that include spatial unit identifiers, spatial boundaries, time slice numbers, processing sequence numbers within nodes, and quality feature vectors, thus forming a slice set and a trajectory identifier set.

[0031] The generation of the preloaded space-time mapping table includes the following steps: S2.1: In the initial stage, each robot node retrieves historical quality inspection data fragments of similar quality inspection objects from storage.

[0032] Specifically, in the initial stage of the system, each robot node accesses the historical quality inspection database through a distributed file system or cloud storage service. Historical quality inspection data is stored in shards within a centralized or distributed storage system. Each record contains the following fields: a unique shard identifier, shard spatial boundary coordinates, shard acquisition timestamp, shard source node identifier, and associated quality inspection object model identifier. Historical data undergoes data cleaning and normalization preprocessing to ensure consistency between the spatiotemporal coordinate system and the current task's baseline coordinate system. The system employs a version management mechanism to ensure the timeliness and consistency of historical data.

[0033] S2.2: Based on the baseline 3D model of the object to be inspected, the inspection space is divided into non-overlapping basic spatial units using a cubic mesh with a side length of L millimeters, and each basic spatial unit is assigned a unique spatial unit identifier.

[0034] The baseline 3D model is obtained through any of the following methods: importing the standard 3D model file of the object to be inspected from a computer-aided design (CAD) system, obtaining a reference point cloud model by performing a high-precision 3D scan of the standard sample in advance, or generating a statistical average model based on similar objects.

[0035] The value of L satisfies the following condition: L is the product of the average point spacing of the point cloud and the scaling factor. The scaling factor ranges from 5 to 20, which can ensure that each basic spatial unit contains at least the preset minimum number of point cloud points (such as 1000 points).

[0036] It should be noted that the starting point of the cube mesh is aligned with the lower left corner of the smallest outer cube of the baseline 3D model, and each basic spatial unit is assigned a unique spatial unit identifier.

[0037] S2.3: Calculate the frequency of each basic spatial unit being covered by data shards in the historical quality inspection data sharding records, and generate a spatial region sharding frequency statistics table.

[0038] The spatial overlap determination uses the following algorithm: calculate the intersection of the historical fragmentation spatial boundary and the axis-aligned bounding box of the basic spatial unit. If the intersection volume is greater than the volume threshold (e.g., set to 10% of the unit volume), it is determined to be an overlap.

[0039] S2.4: Set a primary threshold and a secondary threshold for segmentation frequency. The primary threshold is the quantile value of the top 30% of historical frequency rankings, and the secondary threshold is the quantile value of the bottom 30%. Basic spatial units with frequencies higher than the primary threshold are marked as Class I spatial units, those lower than the secondary threshold are marked as Class II spatial units, and otherwise as Class III spatial units (mid-frequency region). A default time slice length is used, determined by calculating the median of all historical segmentation time intervals. The classification results and the basic spatial unit set are preloaded to each robot node.

[0040] S2.5: Each robot node analyzes the historical quality inspection data fragment records. For the first and second spatial units that have been classified, the timestamp sequence of all historical data fragments falling into them is extracted respectively.

[0041] S2.6: Calculate the arithmetic mean of the intervals between adjacent timestamps in each category of timestamp sequence, filter out outliers that exceed the average interval n (optional 2.0-3.0, preferably 2.5) times, and recalculate to obtain the average data collection time intervals Δt1 and Δt2 respectively.

[0042] The outlier filtering uses the Z-score method.

[0043] S2.7: Set Δt1 to the time slice length L1 corresponding to the first type of spatial unit, set Δt2 to the time slice length L2 corresponding to the second type of spatial unit, and use the default time slice length for the third type of spatial unit; form a preloaded space-time mapping table consisting of spatial unit identifier, spatial unit category and corresponding time slice length, and send it to each robot node.

[0044] Furthermore, after receiving the raw point cloud data, each robot node performs the following sharding operation: Based on the point cloud timestamp sequence, the continuous time stream is divided into non-overlapping time slice windows according to the time slice length of the corresponding spatial unit in the mapping table. If the point cloud time span cannot be divided by the time slice length, the last time slice adopts the actual time span.

[0045] For the point cloud within each time slice, it is divided based on the spatial cell grid: each point in the point cloud is mapped to the corresponding basic spatial cell; point clouds belonging to the same spatial cell are aggregated to form quality inspection data slices.

[0046] In this process, the spatial boundaries of each quality inspection data segment are obtained by calculating the bounding box based on axis alignment.

[0047] Furthermore, the quality feature vector includes: geometric features (such as surface roughness, curvature distribution, flatness, etc.); statistical features (point cloud uniformity, outlier ratio, etc.); and topological features (number of connected components and boundary complexity). Each feature dimension is z-score standardized to ensure that the numerical range of each dimension is consistent.

[0048] It should be noted that this invention achieves adaptive partitioning of the detection space based on a space-time mapping table derived from historical data analysis. High-frequency regions utilize short time slices to capture dynamic changes, while low-frequency regions employ longer time slices to reduce processing overhead and improve resource utilization efficiency. The slice processing trajectory identifier fully records the spatiotemporal origin and processing sequence of the slices, supporting end-to-end quality traceability and anomaly localization, thus enhancing the interpretability of the quality inspection process.

[0049] S3: The collaborative node receives all sets, groups them according to spatial unit identifier and time slice number, performs consistency comparison on the quality feature vectors within the same group, calculates the feature difference degree, and checks the continuity of processing sequence numbers within different robot nodes to determine the integrity of the processing sequence and generate the slice consistency analysis results.

[0050] S3.1: The calculation of feature differences includes: collaborative nodes group quality inspection data with the same spatial unit identifier and time slice number into the same comparison group. This grouping mechanism ensures that the analysis is conducted between data collected by different nodes within the same time period at the same detection location, forming the basis for horizontal comparability.

[0051] For each comparison group, extract the quality feature vectors from all segmented processing trajectory identifiers within the group, calculate the Euclidean distance between any two feature vectors, and take the median of all distances as the feature difference within the group.

[0052] Specifically, to quantify the dispersion of features among different segments within a group, the system executes the following calculation process: First, all feature vectors within the group are standardized to eliminate differences in the units and numerical ranges of different feature dimensions. Next, the Euclidean distance between feature vectors of any two different segments within the group is calculated; a larger value indicates a greater difference in quality features between the two segments. After obtaining the Euclidean distance values ​​for all paired segments, instead of simply averaging, the median of all distance values ​​is taken as the feature dissimilarity of the comparison group.

[0053] It should be noted that using the median as the statistic, compared to the arithmetic mean, is less sensitive to a few abnormally large distance values ​​(which may be caused by noise or serious defects), thus allowing the feature variance to more robustly reflect the consistency level of the main body of the data. Simultaneously, the system identifies and records the feature dimension that contributes the most to the overall variance by analyzing the mean absolute deviation across each feature dimension.

[0054] S3.2: Determining the integrity of the processed sequence includes: For each comparison group, the quality inspection data is grouped according to the robot node identifier. For each subgroup corresponding to a robot node: the processing sequence number within the node of all quality inspection data subgroups is extracted and sorted in ascending order to obtain sequence S; continuity analysis is performed on sequence S: if the number of sequence numbers in sequence S is less than 2, the sequence integrity is recorded as 1, that is, if there is only one sequence number in the sequence, the integrity is considered impossible to determine and is recorded as complete; otherwise, the interval between adjacent sequence numbers is calculated. If all intervals are 1 (the sequence numbers in the sequence increase continuously, that is, the difference between adjacent sequence numbers is always 1), the sequence integrity is 1; if there are intervals greater than 1, the abnormal interval coefficient is calculated. The numerator of the abnormal interval coefficient is the sum of all interval values ​​greater than one, and the denominator is the theoretical span of the entire sequence number (i.e., the difference between the largest and smallest sequence numbers plus one), reflecting the proportion of missing sequence numbers. The sequence integrity is 1 minus the abnormal interval coefficient, and the minimum value is 0. Ultimately, the overall sequence completeness of an alignment group is determined by the lowest calculated subgroup completeness among all participating nodes, following the "weakest link" principle to ensure rigorous evaluation. The system also precisely records which node is missing and which specific sequence numbers are missing.

[0055] It should be noted that determining the integrity of the processing sequence aims to check whether the internal processing flow of each robot node is continuous and uninterrupted when generating data fragments belonging to the same alignment group.

[0056] The minimum sequence completeness among all robot nodes is taken as the sequence completeness of the comparison group.

[0057] S3.3: The results of the fragment consistency analysis include: alignment group identifier, feature difference degree, sequence integrity, feature dimension identifier with the largest difference, node identifier of the processed sequence number within the missing node, and list of missing sequence numbers.

[0058] As can be seen, this invention, by verifying the continuity of processing sequence numbers within nodes, can effectively detect anomalies such as frame drops and interruptions in the data acquisition, transmission, or processing stages, ensuring the traceability and reliability of the data chain and guaranteeing the quality of the final data from a process perspective. Furthermore, the analysis results not only provide the overall consistency level but also clearly indicate which quality characteristic dimensions the differences mainly originate from, and which specific robot node's data sequence is missing. This greatly facilitates maintenance personnel in troubleshooting and targeted repairs.

[0059] S4: Collaborative nodes filter quality inspection data fragments with feature differences higher than the difference threshold or sequence integrity lower than the integrity threshold, perform local re-division based on spatial boundaries, generate fine-grained quality inspection data sub-fragments, and update the trajectory identifier set.

[0060] S4.1: Based on the sharding consistency analysis results, the cooperating nodes select shards to be further divided.

[0061] The difference threshold is defined as follows: it is based on the statistical distribution of feature differences in historical quality inspection data (e.g., taking the top 20% quantile) or a threshold value set through domain knowledge, used to identify boundaries where feature consistency significantly deviates from normal levels. The integrity threshold is set based on the system's minimum requirements for data integrity (e.g., 0.7), used to identify sequence missing issues that have reached a level affecting reliability. If either condition is met, the comparison group is determined to have a data consistency problem, and all quality inspection data fragments within the group are marked as needing further fragmentation. This screening mechanism ensures that the optimization objective is precisely focused on the problematic data.

[0062] S4.2: Send a re-division request to the robot node that is collecting the data to be re-divided into segments; wherein, the re-division request includes: the trajectory identifier of the segment to be re-divided, and the spatial boundaries of all other quality inspection data segments within the same time slice number.

[0063] Optionally, the spatial boundaries can be recalculated here. It should be noted that the initial calculation (in S2) occurs when the robot node generates trajectory identifiers. The spatial boundaries are calculated based on the preprocessed point cloud (which may have been filtered and downsampled), primarily to generate compact descriptors for transmission and initial grouping. The spatial boundary calculation here occurs after the robot node receives the re-division request, recalculated based on the locally stored raw point cloud data (uncompressed or downsampled, retaining complete details). The purpose is to provide the most accurate spatial range basis for subsequent high-precision spatial relationship determination and segmentation operations.

[0064] S4.3: Perform the following spatial relationship determinations in sequence: If the overlap volume between the spatial boundary and the spatial boundary of any other quality inspection data segment exceeds an overlap threshold (e.g., 50%), it is considered data redundancy, and a data merging operation is performed. The data merging operation first uses the Iterative Closest Point (ICP) algorithm to precisely register the two highly overlapping point cloud segments, ensuring spatial alignment. Then, a voxel grid downsampling fusion algorithm is used to sample the two point clouds into a common voxel grid. Points falling within the same voxel are weighted and averaged to generate a new fused point cloud segment with uniform point distribution and eliminated duplicate information. This new segment inherits the trajectory identifier of the main segment and adds a merging marker; its quality feature vector needs to be recalculated.

[0065] If the minimum distance between the spatial boundary and the spatial boundaries of all other quality inspection data segments is greater than the gap threshold (e.g., 20 mm, or dynamically set according to the point cloud resolution), it is determined to be an isolated segment and marked as an acquisition anomaly.

[0066] This typically means that the segment is spatially disconnected from the acquisition areas of other nodes, possibly due to uneven task allocation or obstructed sensor field of view. The marking information includes the anomaly type (spatial isolation), isolation distance, and the associated spatial unit identifier and time slice number. This anomaly marker is returned to the coordinating node, which drives the task planning module to reschedule or supplement the acquisition of the corresponding spatial region in subsequent acquisition cycles.

[0067] If the minimum distance between the spatial boundary and at least one other quality inspection data segment spatial boundary is greater than 0 and less than or equal to the gap threshold, or the overlap volume ratio is greater than 0 but less than the overlap threshold, then it is determined that further subdivision is required.

[0068] Specifically, if neither of the above two conditions occurs—that is, the fragment is spatially related to other fragments but is neither highly redundant nor completely isolated—specifically, if the minimum spatial distance to at least one other fragment's boundary is greater than 0 but less than or equal to the gap threshold (there is a bridging gap), or the overlap volume percentage is greater than 0 but less than the overlap threshold (there is partial overlap but it does not meet the fusion standard), then the fragment is ultimately determined to need further subdivision. Its core purpose is to divide a large fragment that is spatially poorly matched with other fragments into several fine-grained sub-fragments with clearer boundaries and better alignment with adjacent fragments.

[0069] S4.4: If further subdivision is determined, perform the following operations: Based on the spatial relationship determination results, identify the specific spatial relationship that leads to the need for further subdivision. For example, if triggered by a gap with a segment in a specific direction, initially determine that the segmentation plane should be approximately perpendicular to the gap direction to generate a new boundary that can better connect with adjacent segments after segmentation; within the point cloud data of the segment to be further subdivided, use a RANSAC-based plane detection algorithm to find the optimal segmentation plane; divide the segment to be further subdivided into two or more sub-segments along the segmentation plane, and use a pruning algorithm to remove blurry points adjacent to the segmentation plane; perform point cloud pruning and re-encapsulation on the sub-segments to generate fine-grained quality inspection data sub-segments, and update the trajectory identifiers. Specifically, the spatial unit identifier is inherited from the original segment but with an added subdivision code, the time slice number remains unchanged, the processing sequence number within the node is derived from the original sequence number, the spatial boundary is recalculated based on the subset, and the quality feature vector also needs to be re-extracted. Simultaneously, establish an inheritance chain between the new identifier and the original identifier.

[0070] Optionally, the system may obtain multiple candidate planes that meet the criteria (e.g., gaps from different directions). The criteria for selecting the optimal segmentation plane are comprehensive, including: the number of interior points obtained by the plane (reflecting the degree of conformity with the point cloud structure), the balance of the volumes of the two sub-pieces generated after segmentation, and the coordination of the parallel or perpendicular relationship between the segmentation plane and the boundaries of adjacent pieces. Based on actual needs, the plane with the highest comprehensive score can be selected as the final segmentation plane through weighted scoring. This invention does not impose specific requirements here.

[0071] The segmented sub-shard data, updated trajectory identifiers, and re-segmentation analysis results are sent back to the collaborative node.

[0072] After receiving the results, the collaborative nodes update the global fragment set and trajectory identifier set accordingly: replacing merged or re-divided old fragments, adding newly generated sub-fragments or merged fragments, and recording anomaly information into the monitoring system. This achieves a closed loop between distributed edge processing and central collaboration.

[0073] S5: The collaborative nodes monitor and analyze the results in multiple rounds of collaborative processing, and collect the quality inspection data that continuously meets the stability conditions into a consensus quality inspection fragment set, and distribute it to each robot node along with the updated trajectory identifier to optimize the quality inspection strategy for subsequent tasks.

[0074] In a preferred embodiment of the present invention, the collaborative node serves as the knowledge aggregation and evolution hub of the distributed quality inspection system. Through a multi-round collaborative processing mechanism, it refines and solidifies the continuously stable and consistent data shards to form reliable consensus quality inspection knowledge, and feeds it back to each robot node, thereby completing the closed loop from data collection to knowledge feedback and realizing the continuous optimization and self-evolution of the system.

[0075] This invention employs an iterative collaborative processing model, defining a complete acquisition-processing-analysis-optimization cycle as a collaborative round. In each round, collaborative nodes execute steps S1 to S4 and maintain a global database of fragmented historical status records. This database records the fragmented consistency analysis results obtained by each quality inspection data fragment throughout its lifecycle and across all collaborative rounds, including time-series sequences of feature differences and sequence integrity.

[0076] The sustained stability condition refers to the following: For the same quality inspection data slice, the median of the feature difference values ​​calculated in multiple consecutive rounds (e.g., the most recent 10 rounds) of collaborative processing is taken as the feature difference stability threshold for that slice in the current round. Simultaneously, a preset fixed value of 0.9 is set as the sequence integrity stability threshold. If, in the most recent M consecutive rounds (M≥3, e.g., 3 rounds), the feature difference of that slice is consistently lower than the feature difference stability threshold for the current round, and the sequence integrity is consistently higher than the sequence integrity stability threshold, then that slice is considered to be continuously stable.

[0077] This dual dynamic threshold mechanism takes into account both the historical trend of the consistency of the fragment's own features (achieving self-adaptation by stabilizing the threshold through feature difference) and the absolute requirement for the integrity of the data acquisition and processing chain.

[0078] The generation of the consensus quality inspection fragment set includes: collaborative nodes fragmenting all quality inspection data that meet the stability conditions and clustering them according to spatial unit identifiers; extracting quality feature vectors from the quality inspection data fragments within each spatial unit and calculating the arithmetic mean as the aggregated feature vector; calculating the Euclidean distance between each fragment feature and the aggregated feature, and grouping quality inspection data fragments with a distance less than the aggregation threshold into one class; wherein the aggregation threshold is 1.2 times the median of the aggregation distances calculated in the last 5 consensus formation processes; merging the fragments in each spatial candidate cluster that belong to the core consistency class, together with their optimized fragment processing trajectory identifiers (if they have undergone S4 re-division, the updated identifiers are used), to generate the final consensus quality inspection fragment set. This set is the material embodiment of the consistent understanding of a specific spatial region of the quality inspection object reached by the system after multiple collaborative verifications, constituting a high-quality reliable knowledge template library; and distributing it as a reliable knowledge template to each robot node.

[0079] As can be seen, through multiple rounds of verification and dynamic threshold determination, the system can automatically identify and aggregate highly consistent, stable, and reliable data, elevating it into reusable consensus knowledge, thus completing the transformation from data to knowledge. Furthermore, the generated consensus knowledge template is fed back to the front-end robot nodes, directly guiding subsequent collection, processing, and collaborative behaviors, forming a complete learning cycle of "collection-collaboration-consensus-optimization," enabling iterative improvements in system performance.

[0080] As can be seen, this invention, through multi-round collaborative processing, extracts reliable knowledge templates from multi-source data, dynamically updates the space-time mapping and quality inspection strategies, and feeds them back to each robot node. The system continuously learns and evolves in a closed loop of data-knowledge-optimization, enabling the distributed quality inspection capabilities to continuously improve and enhance themselves over time, achieving a qualitative leap from experience accumulation to intelligent adaptation.

[0081] On the other hand, such as Figure 2 As shown, the present invention also provides a multi-robot quality inspection optimization system based on distributed collaboration and knowledge evolution, comprising: The raw data acquisition module is used for multiple robot nodes to acquire raw data of the same quality inspection object in parallel. The slicing and mapping module is used by each robot node to divide the raw data into quality inspection data slices based on a preloaded space-time mapping table, extract the quality feature vector of each quality inspection data slice, and generate slicing processing trajectory identifiers containing spatial unit identifiers, time slice numbers, node processing sequence numbers and quality feature vectors, forming a slice set and a trajectory identifier set. The feature extraction module is used to receive all sets of data from the collaborative nodes, group them according to spatial unit identifiers and time slice numbers, perform consistency comparisons on the quality feature vectors within the same group, calculate the feature difference degree, and simultaneously check the continuity of processing sequence numbers within different robot nodes to determine the integrity of the processing sequence and generate slice consistency analysis results. The trajectory generation module is used to collaboratively filter quality inspection data fragments with feature differences higher than the first threshold or sequence integrity lower than the second threshold, perform local re-division based on spatial boundaries, generate fine-grained quality inspection data sub-fragments, and update the trajectory identifier set. The sharding and aggregation module is used by collaborative nodes to monitor and analyze the results in multiple rounds of collaborative processing. It shards and aggregates the quality inspection data that continuously meets the stability conditions into a consensus quality inspection shard set, and distributes it to each robot node along with the updated trajectory identifier to optimize the quality inspection strategy for subsequent tasks.

[0082] The system also includes one or more processors and memory.

[0083] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the multi-robot quality inspection optimization method based on distributed cooperation and knowledge evolution described in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0084] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the multi-robot quality inspection optimization method based on distributed cooperation and knowledge evolution of the foregoing embodiments, particularly... Figure 1 The flowchart of the method is shown.

[0085] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0086] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0087] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if required, the program can be implemented in assembly or machine language.

[0088] In any case, the language can be either compiled or interpreted.

[0089] Furthermore, for this purpose, the program can run on programmed application-specific integrated circuits.

[0090] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0091] Furthermore, the method can be implemented in any suitable computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0092] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0093] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0094] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution, characterized in that: include: Multiple robot nodes collect raw data of the same quality inspection object in parallel; the raw data is raw point cloud data. Each robot node divides the raw data into quality inspection data slices based on a pre-loaded space-time mapping table, extracts the quality feature vector of each quality inspection data slice, and generates slice processing trajectory identifiers containing spatial unit identifiers, spatial boundaries, time slice numbers, processing sequence numbers within nodes, and quality feature vectors, forming a slice set and a trajectory identifier set; The collaborative node receives all sets, groups them according to spatial unit identifier and time slice number, performs consistency comparison on the quality feature vectors within the same group, calculates the feature difference degree, and checks the continuity of processing sequence numbers within different robot nodes to determine the integrity of the processing sequence and generate the piecewise consistency analysis results. Collaborative nodes filter quality inspection data fragments with feature differences higher than the difference threshold or sequence integrity lower than the integrity threshold, perform local re-division based on spatial boundaries, generate fine-grained quality inspection data sub-fragments, and update the trajectory identifier set; The collaborative nodes monitor and analyze the results in multiple rounds of collaborative processing, and collect the quality inspection data that continuously meets the stability conditions into a consensus quality inspection fragment set, and distribute it to each robot node along with the updated trajectory identifier to optimize the quality inspection strategy for subsequent tasks.

2. The multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in claim 1, characterized in that: The generation of the preloaded space-time mapping table includes: In the initial stage, each robot node retrieves fragmented records of historical quality inspection data for similar quality inspection objects from storage; Based on the baseline 3D model of the object to be inspected, the inspection space is divided into non-overlapping basic spatial units using a cubic mesh with a side length of L millimeters, and each basic spatial unit is assigned a unique spatial unit identifier. The frequency of each basic spatial unit being covered by data shards in the historical quality inspection data sharding records is counted, and a spatial region sharding frequency statistics table is generated. Set a primary threshold and a secondary threshold for fragmentation frequency; mark basic spatial units with frequencies higher than the primary threshold as first-class spatial units and those with frequencies lower than the secondary threshold as second-class spatial units; The classification results and the basic spatial unit set are preloaded to each robot node.

3. The multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in claim 2, characterized in that: The generation of the preloaded space-time mapping table also includes: Each robot node analyzes historical quality inspection data fragment records, and extracts the timestamp sequence of all historical data fragments falling into the classified first-class spatial unit and second-class spatial unit respectively; Calculate the arithmetic mean of the intervals between adjacent timestamps in each category of timestamp sequence, filter out outliers that are more than n times the average interval, and recalculate to obtain the average data collection time intervals Δt1 and Δt2 respectively. Set Δt1 to the time slice length L1 corresponding to the first type of spatial unit, and set Δt2 to the time slice length L2 corresponding to the second type of spatial unit. A preloaded space-time mapping table consisting of spatial unit identifiers, spatial unit categories, and corresponding time slice lengths is generated and distributed to each robot node.

4. The multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in claim 3, characterized in that: The calculated feature difference includes: Collaborating nodes will group quality inspection data with the same spatial unit identifier and time slice number into the same comparison group; For each comparison group, extract the quality feature vectors from all segmented processing trajectory identifiers within the group, calculate the Euclidean distance between any two feature vectors, and take the median of all Euclidean distances as the feature difference within the group. The determination of the integrity of the processed sequence includes: For each comparison group, the quality inspection data is segmented and grouped according to the robot node identifier. For each robot node, the segment subgroup is as follows: Extract the intra-node processing sequence number of all quality inspection data fragments in the fragment subgroup, and sort them in ascending order of value to obtain sequence S; Continuity analysis of sequence S: If the number of indexes in sequence S is less than 2, the sequence integrity is recorded as 1. Otherwise, calculate the interval between adjacent sequence numbers. If all intervals are 1, then the sequence integrity is 1. If there are intervals greater than 1, then calculate the abnormal interval coefficient. The numerator of the abnormal interval coefficient is the sum of all intervals greater than one, and the denominator is the theoretical span of the entire sequence number. The sequence integrity is 1 minus the abnormal interval coefficient, and the minimum value is 0. The minimum sequence completeness among all robot nodes is taken as the sequence completeness of the comparison group.

5. The multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in claim 4, characterized in that: The fragmentation consistency analysis results include: The comparison group identifier, feature difference degree, sequence completeness, identifier of the feature dimension with the greatest difference, node identifier of the processed sequence number within the missing node, and list of missing sequence numbers.

6. The multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in claim 5, characterized in that: The local re-division based on spatial boundaries includes: Based on the sharding consistency analysis results, the collaborating nodes select shards to be further divided. Send a re-division request to the robot node that is collecting the data to be further divided into segments; wherein, the re-division request includes: the trajectory identifier of the segment to be further divided, and the spatial boundaries of all other quality inspection data segments within the same time slice number; After receiving the re-division request, the robot node retrieves the corresponding original point cloud data from local storage and uses a point cloud processing algorithm to calculate the spatial boundary of the original point cloud data. Perform the following spatial relationship determinations in sequence: If the overlap volume ratio between the spatial boundary and any other quality inspection data fragment spatial boundary exceeds the overlap threshold, it is determined to be data redundancy, and a data merging operation is performed. If the minimum distance between the spatial boundary and the spatial boundaries of all other quality inspection data segments is greater than the gap threshold, it is determined to be an isolated segment and marked as an abnormal acquisition. If the minimum distance between the spatial boundary and at least one other quality inspection data segment spatial boundary is greater than 0 and less than or equal to the gap threshold, or the overlap volume ratio is greater than 0 but less than the overlap threshold, then it is determined that further subdivision is required. If it is determined that further subdivision is necessary, the following operations are performed: within the point cloud data to be further subdivided, the optimal segmentation plane is found using a RANSAC-based plane detection algorithm; the segment to be further subdivided is divided into two or more sub-segments along the segmentation plane; the point cloud of the sub-segments is clipped and repackaged, and the trajectory labels are updated. The segmented sub-shard data, updated trajectory identifiers, and re-segmentation analysis results are sent back to the collaborative node.

7. The multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in claim 6, characterized in that: The continuous satisfaction of the stability condition means: In multiple rounds of collaborative processing, the feature difference of the same quality inspection data fragments was always lower than the feature difference stability threshold, and the sequence integrity was always higher than the sequence integrity stability threshold. The generation of the consensus quality inspection shard set includes: The collaborative nodes shard all quality inspection data that meet the stability conditions and cluster them according to spatial unit identifiers; The quality inspection data within each spatial unit is segmented, the quality feature vector is extracted, and the arithmetic mean is calculated as the aggregated feature vector. Calculate the Euclidean distance between each fragment feature and the aggregate feature, and group the quality inspection data fragments with a distance less than the aggregation threshold into one category; The quality inspection data fragments classified into one category and their corresponding optimized fragment processing trajectory identifiers are merged into the consensus quality inspection fragment set and distributed to each robot node as a reliable knowledge template.

8. A multi-robot quality inspection optimization system based on distributed collaboration and knowledge evolution, based on the multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in any one of claims 1 to 7, characterized in that: Also includes: The raw data acquisition module is used for multiple robot nodes to acquire raw data of the same quality inspection object in parallel. The slicing and mapping module is used by each robot node to divide the raw data into quality inspection data slices based on a preloaded space-time mapping table, extract the quality feature vector of each quality inspection data slice, and generate slicing processing trajectory identifiers containing spatial unit identifiers, time slice numbers, node processing sequence numbers and quality feature vectors, forming a slice set and a trajectory identifier set. The feature extraction module is used to receive all sets of data from the collaborative nodes, group them according to spatial unit identifiers and time slice numbers, perform consistency comparisons on the quality feature vectors within the same group, calculate the feature difference degree, and simultaneously check the continuity of processing sequence numbers within different robot nodes to determine the integrity of the processing sequence and generate slice consistency analysis results. The trajectory generation module is used to collaboratively filter quality inspection data fragments with feature differences higher than the first threshold or sequence integrity lower than the second threshold, perform local re-division based on spatial boundaries, generate fine-grained quality inspection data sub-fragments, and update the trajectory identifier set. The sharding and aggregation module is used by collaborative nodes to monitor and analyze the results in multiple rounds of collaborative processing. It shards and aggregates the quality inspection data that continuously meets the stability conditions into a consensus quality inspection shard set, and distributes it to each robot node along with the updated trajectory identifier to optimize the quality inspection strategy for subsequent tasks.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the multi-robot quality inspection optimization method based on distributed collaboration and knowledge evolution as described in any one of claims 1 to 7.

Images