Intelligent multi-robot grinding tool cooperative control system

Abstract

The application relates to the technical field of multi-robot control, and discloses an intelligent multi-robot grinding tool cooperative control system. The system comprises a data acquisition module, a state analysis module, a cooperative calculation module and a control optimization module. The data acquisition module acquires real-time operation data of multiple robots in the grinding tool process, and the real-time operation data comprises position data, force data and tool wear data; the state analysis module identifies a stable working mode and a change working mode according to the change mode of the real-time operation data of each robot over time; the cooperative calculation module calculates the cooperative performance index of each robot based on the real-time operation data in the stable working mode and the change working mode; and the control optimization module generates a cooperative grinding tool control strategy of the multiple robots according to the cooperative performance index and the real-time operation data. The system can realize effective cooperative control of the grinding tool operation of the multiple robots and is suitable for the multi-robot cooperative grinding tool scene in the mechanical processing field.

Description

Technical Field

[0001] This invention relates to the field of multi-manipulator control technology, specifically to an intelligent multi-manipulator collaborative control system for sharpening tools. Background Technology

[0002] In the field of machining, tool sharpening is a crucial step in ensuring the accuracy and efficiency of subsequent processing. With the development of automation technology, multi-robot collaborative tool sharpening has gradually become the mainstream application in the industry. Currently, multi-robot tool sharpening systems face the challenge of synchronously acquiring operational data from multiple devices during actual operation. Most systems can only collect some operating parameters of a single robot, making it difficult to comprehensively grasp the positional changes, force output, and tool wear of multiple robots during the sharpening process, resulting in blind spots in the monitoring of the entire collaborative operation process.

[0003] Existing systems lack an effective mechanism for recognizing the working modes of robotic arms, and cannot distinguish between stable and changing working states based on the changes in real-time operation data over time. When the working mode of a robotic arm changes, the system cannot detect this change in time, and therefore cannot adjust subsequent collaborative control strategies, which can easily lead to uncoordinated movements among multiple robotic arms.

[0004] In the collaborative control of multiple robotic arms, existing systems struggle to quantitatively evaluate the collaborative performance of each arm based on operational data from different working modes. The lack of scientific standards for measuring collaborative performance hinders the development of reliable control strategies. While some systems attempt to generate collaborative control strategies, they rely solely on single-dimensional parameters, failing to fully integrate real-time operational data and collaborative performance of multiple robotic arms. This results in control strategies with poor applicability, unable to meet the collaborative control requirements of complex grinding scenarios, impacting the overall quality and efficiency of grinding operations, and restricting the further application of multi-robotic arm grinding systems in high-precision, high-efficiency machining fields. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent multi-robotic arm sharpening collaborative control system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent multi-robotic arm sharpening collaborative control system, the system comprising:

[0007] The data acquisition module is used to acquire real-time operation data of multiple robotic arms during the tool sharpening process. The real-time operation data includes position data, force data, and tool wear data.

[0008] The status analysis module is used to identify stable and variable working modes based on the changes in real-time operation data of each robot over time.

[0009] The collaborative computing module is used to calculate the collaborative performance index of each robot based on real-time operation data under stable and variable working modes.

[0010] The control optimization module is used to generate a collaborative grinding control strategy for multiple robotic arms based on collaborative performance indicators and real-time operation data.

[0011] Preferably, the real-time operation data also includes the start time, end time, and operation duration of each sharpening operation.

[0012] Preferably, the state analysis module is specifically used for:

[0013] For each robotic arm, extract the real-time operation data sequence, calculate the first-order difference value of the sequence, and select the moments with positive difference values ​​as growth moments;

[0014] The growth moments are sorted by the difference value to form a difference sequence. The second difference value of the difference sequence is calculated to determine the critical moments. The time period before the critical moment corresponds to the stable working mode, and the time period after the critical moment corresponds to the changing working mode.

[0015] Preferably, the collaborative computing module is specifically used for:

[0016] In stable working mode, the robot arm with real-time operation data is marked as the reference robot arm; in variable working mode, the robot arm with real-time operation data is marked as the target robot arm.

[0017] For each reference robot and target robot, the control influence of the reference robot relative to the target robot is calculated based on the similarity analysis of real-time operation data and combined with time information.

[0018] By aggregating the control influence of all target robotic arms, the collaborative performance index of each reference robotic arm is obtained.

[0019] Preferably, the calculation of the control influence includes:

[0020] Based on the start time and operation duration of the reference robot and the target robot in each sharpening operation, a similarity metric is derived.

[0021] The ratio of the time spent on the target robot to that of the reference robot during operation is used as the first factor, and the ratio of the similarity metric to the operation duration is used as the second factor.

[0022] The product of the first and second factors is calculated as the control influence.

[0023] Preferably, the control optimization module is further used to calculate the decision-making ability index of each robot based on the tool wear data and operation duration in the real-time operation data;

[0024] By combining collaborative performance indicators and decision-making ability indicators, the control priority of each robotic arm is obtained;

[0025] Based on control priority, a dominant robotic arm is selected, and a collaborative control hierarchy is constructed.

[0026] Preferably, the calculation of the decision-making ability index includes:

[0027] Calculate the characteristic ratio of operation duration to interval time for each sharpening operation, and take the product of the average characteristic ratio of all operations and the number of tool wears as the decision-making ability index.

[0028] Preferably, the control optimization module is specifically used for:

[0029] The robot arm with a control priority higher than the set value is designated as the dominant robot arm;

[0030] For each dominant robot arm, the following degree index of the target robot arm relative to the dominant robot arm is calculated. The following degree index is based on the control influence and decision-making ability index.

[0031] The collaborative control hierarchy is arranged according to the magnitude of the follow-through index, and nodes at the same level in the hierarchy have the same follow-through index.

[0032] Preferably, the control optimization module is further used to divide the control stage hierarchy according to the collaborative control hierarchy and apply differentiated control strategies for different stage levels.

[0033] Preferably, the system further includes a verification module for checking the validity of real-time operation data, verifying local operation data based on global control parameters, and adjusting global control parameters after the verification is passed.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] This intelligent multi-robot knife-grinding collaborative control system, through the inclusion of a data acquisition module, can comprehensively acquire real-time operational data from multiple robots during the knife-grinding process. This data includes position data, force data, and tool wear data, allowing the system to fully grasp the operational status of multiple robots and clearly understand the specific operational status of each robot in the knife-grinding operation. This provides a comprehensive information foundation for subsequent work mode recognition and collaborative control strategy formulation, avoiding the problem of incomplete monitoring of the collaborative operation process due to data loss.

[0036] The status analysis module can accurately identify stable and variable working modes based on the real-time operation data of each robot over time. This allows the system to promptly detect changes in the working status of the robots and clearly understand the working characteristics of multiple robots at different times. When a robot switches from a stable working mode to a variable working mode, the system can quickly capture this change, providing accurate pattern information for subsequent collaborative calculations and control optimization. This ensures that the system has more precise control over the working status of multiple robots.

[0037] The collaborative computing module calculates the collaborative performance indicators of each robot based on real-time operation data under stable and variable working modes, so that the performance of each robot in collaborative operation is presented in a quantitative way. The system can clearly know the differences in collaborative capabilities of different robots under different working modes, clarify the role of each robot in collaborative operation, and provide a scientific performance reference for the differentiated formulation of subsequent control strategies, making collaborative control more targeted.

[0038] The control optimization module generates a collaborative grinding control strategy for multiple robotic arms based on collaborative performance indicators and real-time operation data. This strategy fully integrates the actual operation and collaborative performance of the multiple robotic arms, ensuring that the generated control strategy is highly matched with the actual grinding operation requirements. It adapts to the collaborative control needs of multiple robotic arms under different working modes, making the movements of multiple robotic arms more coordinated during the grinding process. This improves the orderliness of the entire grinding process, ensures that the grinding operation can proceed at the expected pace, and meets the requirements for collaborative control of multiple robotic arms in complex grinding operation scenarios. This will help the multi-robotic arm grinding system play a role in more processing scenarios and promote the improvement of the level of automated collaborative operation in the machining field.

[0039] The various modules of the entire system work together to form a complete multi-robot sharpening collaborative control system. From data acquisition to pattern recognition, and then to performance calculation and strategy generation, each link is closely connected to ensure that the system can continuously and stably control the multi-robot sharpening operation, reduce the uncertainty in the multi-robot collaborative operation process, make the sharpening operation process more controllable, and thus improve the overall effect of the sharpening operation, meeting the industry's higher application expectations for multi-robot sharpening systems. Attached Figure Description

[0040] Figure 1 This is a timing diagram of the intelligent multi-robotic arm sharpening collaborative control system described in this invention;

[0041] Figure 2 This is a schematic diagram of the working principle of the status analysis module;

[0042] Figure 3 This is a schematic diagram illustrating the working principle of the collaborative computing module. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 This invention provides an intelligent multi-robot collaborative sharpening control system. The system integrates multiple modules to achieve intelligent monitoring and optimized control of the robotic sharpening process. A data acquisition module is responsible for acquiring real-time position data, force data, and tool wear data of multiple robots during the sharpening operation; these data constitute the system's basic input. A state analysis module receives real-time operation data and analyzes the data change patterns of each robot over time, thereby identifying stable and variable operating modes. The stable operating mode corresponds to a smooth operation phase with small data fluctuations, while the variable operating mode reflects an adjustment phase with significant data changes. A collaborative calculation module, based on the state analysis results, calculates the collaborative performance index of each robot using data from both stable and variable modes. This index quantifies the degree of mutual influence between the robots. A control optimization module generates a collaborative sharpening control strategy for the multiple robots based on the collaborative performance index and real-time data, achieving overall efficiency improvement by dynamically adjusting the robot's operating parameters. The system's various modules are connected via a data bus to ensure real-time information exchange. The data acquisition module uses a high-precision sensor array installed in key parts of the robot arm, the status analysis module runs on an embedded processor, and the collaborative computing and control optimization module is deployed on a central server, communicating with the robot arm controller via a network protocol.

[0045] Example 1: See Figure 2 The data acquisition module continuously acquires multi-dimensional real-time operational data from the sensor units and control system interfaces of each robot arm. This module not only captures the spatial coordinates of the robot arm's end effector and the normal and tangential forces acting on the cutting tool, but also records the morphological changes of the cutting edge through high-frequency sampling. All these data streams are assigned precise timestamps. The start time marks the instant when the grinding action is activated, and the end time corresponds to the signal indicating the action is completed. The duration of the operation is automatically calculated from the difference between these two time points. The introduction of time data creates conditions for subsequent analysis of the temporal correlation and sequence dependency of robot arm behavior. The state analysis module takes over the preprocessed and time-aligned data streams. Its core task is to interpret the hidden working state modes behind the data sequences generated by each robot arm operating independently.

[0046] This module maintains a dynamically updated data window for each robotic arm, containing data points arranged chronologically, including position, force, wear level, and corresponding timestamps. The analysis begins with a first-order difference operation on the sequence, calculating the numerical change between each data point and its predecessor. A positive value indicates that the physical quantity is increasing at the current moment. The module's built-in filtering logic filters out minor fluctuations caused by sensor noise, accurately capturing those significant growth moments. These marked growth moments and their corresponding positive difference values ​​are temporarily stored. The system sorts these growth moments in descending order based on the magnitude of the difference values, forming a new sequence that reflects the order of growth intensity. For this sequence focusing on the intensity of change, the module further applies a second-order difference operation to investigate the changes in the growth rate itself. When the second-order difference value shows an inflection point from positive to negative, the moment in the original data sequence corresponding to that point is determined to be a critical moment. This inflection point signifies a fundamental change in the robotic arm's operating mode. In the period leading up to the critical moment, the fluctuation range of various operational data is limited to a narrow range. The data changes are gradual and show a certain regularity. This state is defined as the stable working mode, which usually corresponds to the stage when the robot arm performs standardized and highly repetitive grinding tasks.

[0047] After the critical moment, the data sequence begins to exhibit significant dispersion, trend changes, or abrupt changes, indicating that the robot has entered a changing working mode, possibly due to increased tool wear, changing workpieces, or adjustments in response to external commands. The entire state recognition process is continuous and automated. The module constantly receives new data points, updates the data window, and re-executes the aforementioned difference calculation and critical moment judgment process. This dynamic recognition mechanism enables the system to sensitively perceive any transition of the robot from one stable state to another, or from a stable state to an adjustment state. The identified working mode information, along with its time boundaries, is transmitted in real time to the subsequent collaborative calculation module, serving as important contextual basis for assessing the mutual influence between robots. The smoothing of the data sequence employs a recursive filtering algorithm to suppress high-frequency interference, ensuring the stability of the difference calculation. The selection of growth moments does not simply rely on a fixed threshold but is adaptively adjusted by combining the statistical characteristics of the robot's historical data. The critical moment judgment algorithm comprehensively considers the local extrema of the second-order difference sequence and its persistence, avoiding misjudgments due to a single noise point. The pattern segmentation results include not only labels but also a confidence score, reflecting the reliability of the current judgment.

[0048] All calculations are performed independently and in parallel for each robot. An internal state machine tracks the migration history of each robot across different modes, providing information for understanding its long-term behavioral patterns. When a robot's data stream experiences a brief interruption or anomaly, the module can interpolate based on recent historical data or maintain the last known state, ensuring the continuity of system operation. This analysis method based on temporal differential change patterns effectively transforms continuous, complex data streams into semantically clear work phase identifiers. The integrated processing of time data requires a high-precision clock synchronization mechanism. The clock signals of all robot controllers and sensors are unified to the master clock source, with errors controlled to below the millisecond level, thus ensuring accurate time alignment across robots. The detection of start and end times relies on digital signals emitted by the robot controllers or preset sensor thresholds, ensuring consistent start and end point judgments for each grinding operation cycle. The calculated operation duration is compared to a preset reasonable range. If an anomaly occurs (such as a duration that is too short or too long), the operation data is marked for subsequent verification module processing. Within the state analysis module, real-time operational data sequences are managed using a circular buffer, satisfying real-time requirements while saving memory overhead. The calculation of the first-order difference is accompanied by validity checks. For invalid data points caused by data loss or communication delays, the module initiates a data repair procedure or skips the point, waiting for the next valid data point. The logic for filtering growth moments is not isolated; it references information from other data channels of the same robotic arm. When a growth moment occurs in the force data, it simultaneously checks whether the position data also changes accordingly.

[0049] The sorting process by difference value employs an efficient heap sort algorithm, ensuring fast response even under high-density data sampling. Second-order difference calculation targets this sorted growth intensity sequence, aiming to identify inflection points where growth momentum accelerates or decelerates. The final determination of critical moments requires certain persistence conditions; that is, the inflection point feature needs to be confirmed to a certain extent on several subsequent data points to prevent excessively frequent mode switching due to transient interference. Stable and changing operating modes are labeled with time attributes, recording their start time and duration, forming a working mode log for the robot. The module's output includes not only the current operating mode but may also contain predictions of upcoming mode switching, such as when the second-order differences of multiple consecutive data points show a certain trend. This in-depth analysis method, penetrating the internal characteristics of time-series data, enables the system not only to identify the current state but also to perceive the trend of state evolution. All processing logic is designed to be configurable; relevant parameters (such as filter coefficients and thresholds) can be adjusted through the configuration interface to adapt to different robot models or different grinding process requirements. The entire implementation process emphasizes computational efficiency and reliability to ensure stable operation in the real-time control environment of the industrial site.

[0050] Example 2: See Figure 3 The collaborative computing module operates based on the precise differentiation of the robotic arm's working modes. It receives real-time output from the state analysis module, which identifies whether each robotic arm is currently in a stable or changing working mode. Internally, the module maintains a dynamic role mapping table, identifying robotic arms in stable working modes with valid real-time operational data as reference robotic arms. These robotic arms are considered relatively reliable and predictable in their operations. Meanwhile, robotic arms in changing working modes with operational data are identified as target robotic arms; their operations are in an adjustment or transition period. This role assignment is dynamic and parallel; a robotic arm may act as a reference at some times and as a target at others, entirely depending on the real-time evolution of its operational state. For each pair of potentially interacting reference and target robotic arms, the module initiates a similarity analysis process. The core of this process is comparing their real-time operational data during overlapping or adjacent time periods. Similarity analysis is not simply a comparison of data points, but a comprehensive examination of the trajectory matching of position data sequences, the consistency of the shape of force data curves, and the correlation of tool wear trends. Time information is deeply integrated into the analysis process. For example, deviations caused by time asynchrony can be corrected by aligning the start points of operations or comparing the proportion of operation durations.

[0051] Based on this in-depth analysis combining spatiotemporal elements, the module begins to calculate the control influence of the reference robot relative to the target robot. This quantified value aims to characterize the strength of the guiding or constraining effect of the stable operational behavior of the reference robot on the changing operations of the target robot. The specific calculation process of control influence follows a structured method. First, it is necessary to extract the detailed temporal attributes of each sharpening operation of the reference and target robots within a specific time window, including the start time and operation duration. Based on this temporal data, the overlap relationship between the two robots on the operation timeline can be deduced. The calculated time overlap ratio serves as the first factor, reflecting the potential opportunity for the two robots to work collaboratively in the time dimension. The module calculates a similarity metric based on real-time operational data (such as position and force). This metric, after normalization, is ratioed to the current operation duration to form the second factor. The second factor focuses on assessing the consistency of the operational behavior of the two robots per unit time. Multiplying the first factor (time ratio) by the second factor (unit time similarity) yields the product, which is defined as the control influence of the reference robot on the target robot in this operation. This calculation is performed one by one for all paired operations with temporal overlap. After completing the calculation of the micro-level control influence of all valid pairs, the module enters the aggregation phase, which aims to integrate the dispersed influence of a reference manipulator on all its associated target manipulators into a unified collaborative performance index.

[0052] The aggregation process is not a simple arithmetic addition, but rather considers the importance weights of different target robots and the time decay effect of their influence. For example, recently occurring influential events may be given higher weights. The aggregation algorithm traverses all units marked as target robots, weighting and fusing their control influence values ​​on the same reference robot according to time series and correlation strength, ultimately generating a comprehensive scalar value, namely the collaborative performance index of the reference robot. The level of this index directly reflects the robot's ability to coordinate and drive its surrounding companions in changing states within the current system environment. The entire collaborative calculation process is a continuously running loop. The module constantly obtains the latest pattern labels from the state analysis module and the latest operational data from the data acquisition module, then refreshes the role mapping table and re-executes the complete process of pairing, similarity analysis, influence calculation, and index aggregation. The calculation process fully considers the temporality, noise, and uncertainty of industrial field data. The built-in fault tolerance mechanism can handle occasional data missing or outliers, ensuring that the output collaborative performance index has sufficient robustness and reference value. The module maintains a dynamic data cache for each robot arm to store recent operation data and timestamps of sufficient duration to ensure adequate historical context when performing similarity comparisons.

[0053] The role mapping table is updated in strict synchronization with the output of the state analysis module. Any switch in the robot's working mode will immediately trigger an update of the mapping relationship, thus ensuring the real-time nature of collaborative analysis. The similarity analysis process employs multiple measurement algorithms to adapt to different types of data characteristics. For position trajectories, a dynamic time warping method may be used to address time scaling issues, while for force curves, the Pearson correlation coefficient may be used to assess morphological similarity. Time alignment is a crucial preprocessing step before similarity analysis. The system uses interpolation or sliding window matching techniques to compensate for potential time synchronization errors between different robots. The time overlap ratio calculation in the control influence calculation is accurate to the millisecond level, requiring consideration of various complex situations such as partial overlap, complete inclusion, or sequential adjacency in the operation cycle. The algorithm can accurately identify and calculate the effective overlap ratio under various time relationships. The similarity measurement results undergo a standardization process, mapping them to a numerical range between zero and one to eliminate the influence of different physical dimensions, allowing data from different dimensions (such as position and force) to participate in comprehensive calculations. In the calculation of the second factor, the operation duration is used as the denominator, which means that even if the absolute similarity is high in long-term operations, the cooperative efficiency per unit time may be diluted. This guides the system to pay more attention to efficient and compact cooperative behavior.

[0054] When integrating the influence of multiple target robotic arms, the aggregation algorithm can adopt different strategies based on the system design objectives. For example, summing and averaging emphasizes the overall scope of influence, while taking the maximum value focuses on the most important collaborative relationships. The calculation results of the collaborative performance index are accompanied by a time validity label, indicating the representative time window of the index. After exceeding this window, the index needs to be recalculated. The module has a monitoring subroutine that continuously tracks various intermediate variables and final indicators during the calculation process. Once a sharp fluctuation or deviation from a reasonable range is detected in the calculation results, a self-check process is triggered to investigate whether there are any anomalies in the data source or calculation logic. All parameters of the calculation steps, such as the selection of similarity measures, the setting of weight factors, and the aggregation algorithm, are designed as configurable items, allowing system integrators to fine-tune them according to the specific characteristics of the grinding process and the on-site environment. This highly modular and parameterized design makes the collaborative computing module highly adaptable and scalable, capable of handling multi-robotic arm collaborative grinding scenarios of different scales and complexities. The entire implementation emphasizes the effectiveness of the calculation and the stability of the system, ensuring that all calculation tasks can be completed within a real industrial control cycle and output reliable results for high-level decision-making.

[0055] Example 3: Before generating a collaborative strategy, the control optimization module needs to evaluate the comprehensive capabilities of each robot. This evaluation relies not only on collaborative performance indicators reflecting collaborative relationships but also introduces a decision-making capability indicator characterizing the robot's own operational characteristics and durability. This indicator is calculated based on statistical analysis of the robot's historical operational data, focusing on two core elements: first, the ratio of the operation duration of each sharpening operation to the subsequent interval; and second, the total number of tool wear cycles experienced by the robot. The operation duration is directly derived from the start-end time difference of each operation recorded by the data acquisition module, while the interval is defined as the idle time between the end of the current operation and the start of the next operation. For a given robot, its decision-making capability indicator... Determined by the following relationship:

[0056] ;

[0057] in: This represents the characteristic ratio of all sharpening operations performed by the robotic arm within the statistical period. The arithmetic mean, , It is the duration of the i-th operation. It is the interval after the i-th operation ends. It is the total number of tool wear events recorded by the robot within the statistical period.

[0058] Calculate the characteristic ratio The process requires precise time data. The system extracts the robot's operation sequence from the data log in chronological order and calculates the time of each operation. and the corresponding Characteristic ratio The value reflects the intensity of the operation; a higher ratio indicates a larger operation time occupancy, meaning the robotic arm is in a relatively busy working state. Subsequently, for all calculated... Calculate the arithmetic mean of the values. This average value characterizes the typical operation-intermittent rhythm of the robotic arm within the statistical period. Simultaneously, the system accumulates the number of times the tools the robotic arm is responsible for grinding are determined to require maintenance or replacement from the tool wear data stream; that is, the wear count. Finally, the average characteristic ratio will be... With wear count Multiply them to obtain the decision-making ability index. This indicator has a dual meaning: the average characteristic ratio. High-performance robotic arms typically exhibit higher work efficiency or task continuity, while experiencing fewer wear cycles. More robotic arms mean more accumulated experience in tool handling, and the product of these two factors comprehensively evaluates the robotic arm's operational efficiency and experience level.

[0059] After obtaining the decision-making capability index, the control optimization module fuses it with the collaborative performance index from the collaborative computing module to determine the control priority of each robot. The fusion process is not a simple weighted sum, but rather employs a multi-attribute decision-making method, assigning appropriate weight coefficients to the two indicators based on the management objectives of different production stages. For example, in stages emphasizing production cycle time, the weight of the decision-making capability index may be increased; while in stages focusing on system coordination, the weight of the collaborative performance index will dominate. Through a weighted fusion algorithm, a scalar value representing the overall priority of each robot is calculated. The level of the control priority directly determines the robot's position in the subsequent collaborative control structure.

[0060] Based on the calculated set of control priorities, the module begins to construct a hierarchical structure for collaborative control. First, a priority threshold needs to be set. This threshold can be the median, average, or a value dynamically adjusted based on system load. All robots with a control priority higher than this threshold are identified as dominant robots, which will become key nodes or leaders in the control network. The remaining robots are considered followers or ordinary nodes. The selection process for dominant robots ensures that the control core is handled by the unit with the strongest overall capabilities. The hierarchical structure is built with these dominant robots as the root or top-level nodes. The remaining robots are assigned to different levels based on their correlation strength with the highest-priority dominant robot (usually derived from the influence relationship reflected by collaborative performance indicators), forming a tree-like or network-like control topology. This hierarchical structure clearly defines the flow of control commands and the scope of responsibility, with higher-level robots providing guidance or constraints to lower-level robots.

[0061] The calculation and prioritization of decision-making capability indicators is a periodic process. The module sets a statistical window, for example, every certain period of time (e.g., ten minutes) or after completing a certain number of sharpening operations, to recalculate the indicators and update the priority ranking and control hierarchy with the latest data. This dynamic update mechanism allows the system to adapt to gradual changes in robot performance (such as efficiency changes caused by tool wear) or sudden adjustments in task requirements. The module internally maintains a complete operation history for each robot to calculate the average characteristic ratio, while an independent counter continuously tracks the number of tool wear events associated with each robot. All calculations consider the validity and completeness of the data; missing or obviously abnormal data points are removed or smoothed to ensure the accuracy of indicator calculations. The list of control priorities and the collaborative control hierarchy, as key outputs, are transmitted in real time to the strategy generation unit for the final synthesis of a specific multi-robot collaborative sharpening control instruction set.

[0062] Characteristic ratio The calculations imply a deep understanding of the robot's work cycle, resulting in a higher... A high value indicates that the robotic arm quickly begins the next operation after completing one, with a tight work rhythm. This may mean that the robotic arm is undertaking a heavy workload or has high work efficiency. Conversely, a lower value indicates a lower efficiency. The value indicates a relatively long idle time between operations, possibly due to waiting for upstream processes, system scheduling, or maintenance. Calculate the average value. This is to eliminate the random fluctuations of a single operation, thereby capturing the typical time characteristics of the robot's stable operating mode. (Tool wear count) The statistics are not simply counts. The system will confirm a valid wear event based on a preset wear threshold (such as the change in the cutting edge radius) to avoid false counts caused by sensor noise.

[0063] Decision-making ability indicators The design philosophy is that an ideal robotic arm with high decision-making capabilities should exhibit a highly efficient work rhythm (high efficiency). ) and experience in handling complex wear conditions (high) Multiplying the two is based on the assumption that they may mutually reinforce each other, for example, an experienced robotic arm (high... ) Idle time can be shortened by optimizing operation paths, thereby improving efficiency. The fusion calculation for control priorities employs standardized processing. First, collaborative performance indicators and decision-making ability indicators from different dimensions are normalized to the same numerical range before weighted summation. This prevents any single indicator from dominating the result due to excessively large absolute values. The weighting coefficients are typically configurable system parameters, allowing operators to flexibly adjust them according to actual production strategies.

[0064] The selection threshold for the dominant robot arm is not static. The system can incorporate adaptive algorithms, such as dynamically setting the threshold based on the dispersion of the current robot arm priorities, to ensure that a reasonable number of dominant robots are selected under any working condition. The algorithm for constructing the cooperative control hierarchy may employ community detection or hierarchical clustering methods from graph theory, grouping closely related robots into the same subtree or cluster to optimize internal communication efficiency. The entire evaluation and construction process is designed with low computational complexity to ensure it can be completed within the real-time scheduling cycle of the industrial controller and will not become a performance bottleneck for the system. The module also outputs a construction log, recording the reasons for each hierarchy update, changes in the dominant robot arm, and other information for system behavior analysis and optimization. This dynamic, multi-dimensional index-based method for control priority allocation and hierarchy construction provides a core decision-making framework for achieving efficient and adaptive large-scale multi-robot cooperative tool sharpening.

[0065] Example 4: After determining the control priorities of all robotic arms, the control optimization module enters the stage of constructing a specific collaborative control structure. The core task of this stage is to identify the dominant robotic arm and clarify the follower relationships. The module internally presets a priority threshold, which may be an absolute value or a relative value dynamically calculated based on the statistical distribution of the priorities of all robotic arms in the current system (such as the upper quartile). The control priority of each robotic arm is compared with this preset value. Individuals with priorities higher than the preset value are marked as dominant robotic arms, and they will assume the roles of planning, coordinating, or issuing reference commands in subsequent collaborative control. The remaining robotic arms with priorities lower than or equal to the preset value are initially classified as potential followers. To accurately characterize the affiliation between these potential followers and specific dominant robotic arms, the module needs to calculate a quantitative value called the follower index. The follower index is not an isolated value; its calculation deeply integrates control influence data from the collaborative calculation module and the currently evaluated decision-making ability index.

[0066] The calculation process is typically represented by a weighted function. Control influence reflects the strength of the dominant robot's influence on the target robot's behavior in historical operations, while the decision-making ability index modifies the reliability of this relationship: a target robot with higher decision-making ability may have a better understanding and execution of the dominant robot's instructions, and its following behavior may be more stable. The follow-through index is jointly determined by the control influence value and the decision-making ability index. Its calculation method ensures that a target robot with higher decision-making ability will exhibit higher follow-through when subjected to stronger control influence. After calculating the follow-through index of each potential follower relative to each dominant robot, the module begins to construct a collaborative control hierarchy. The construction process is based on the magnitude of the follow-through index, with the core principle being to organize robots with similar follow-through levels into the same level. Specifically, for each dominant robot, the system arranges all its potential followers in descending order of the calculated follow-through. Using clustering or setting follow-through intervals, followers whose follow-through values ​​fall within a specific range are grouped into the same level. For a dominant robot, a tree-like substructure is formed with it as the root node, containing several follower levels. Nodes within the same level (i.e., follower robots) have the same or very similar follower performance values, while there are significant differences in follower performance between different levels. The overall cooperative control hierarchy of the system is composed of all these subtrees centered on the dominant robot. It clearly defines the flow path and coordination scope of control information, and higher-level nodes provide guidance to lower-level nodes.

[0067] Based on the established collaborative control hierarchy, the control optimization module further divides the control phases. Each phase corresponds to a different stage of the operational process, such as the system startup and initialization phase, the stable grinding production phase, and the tool change or process parameter adjustment phase. Different control strategies are applied to different phases. In the initialization phase, the control strategy might focus on having the higher-level dominant robot perform baseline calibration actions and distribute calibration parameters to lower-level followers. In the stable production phase, the strategy might emphasize synchronization and load balancing between levels, with the dominant robot generating the baseline motion trajectory and the follower robots making fine adjustments to compensate for individual differences. In the adjustment phase, the strategy might allow robots at specific levels to gain greater autonomy to respond quickly to changes. This hierarchical, phased, differentiated control allows multi-robot systems to flexibly adapt to different task requirements and changes in external conditions. Assuming a system with five robots (numbered M1 to M5), the control optimization module determines the dominant robot based on the calculated control priority (assuming a threshold of 7.0).

[0068] Table 1. Control Priority and Dominant State Determination Table for Robotic Arm:

[0069]

[0070] Referring to Table 1, robotic arms M1, M2, and M5 are identified as the dominant robotic arms. The module needs to calculate the following degree index of the non-dominant robotic arms M3 and M4 relative to each dominant robotic arm. Assume the control influence data of M3 and M4 on each dominant robotic arm obtained from the collaborative computing module are as follows: M3's influence from M1 is 0.7, from M2 is 0.4, and from M5 is 0.6; M4's influence from M1 is 0.5, from M2 is 0.8, and from M5 is 0.3. The following degree index calculation will combine these influence values ​​with the robotic arm's own decision-making ability index (3.0 for M3, 2.8 for M4). A simple calculation method is to multiply the control influence by the decision-making ability index (the specific algorithm can be adjusted). The following degree of M3 relative to M1 is approximately... The following degree relative to M2 is approximately The follow-through rate relative to M5 is approximately Similarly, the following degree of M4 relative to M1 is approximately The following degree relative to M2 is approximately The follow-through rate relative to M5 is approximately Based on these calculations, the module assigns followers to each dominant robot. For the dominant robot M1, its potential followers are M3 (following degree 2.1) and M4 (following degree 1.4), arranged in descending order of following degree as M3 and M4.

[0071] Assuming the system defines a following degree of 2.0 or higher as the first following level, and between 1.0 and 2.0 as the second following level, then M3 is assigned to the first following level of M1, and M4 is assigned to the second following level of M1. For the dominant robot M2, its potential followers are M3 (following degree 1.2) and M4 (following degree 2.24), arranged as M4, M3. M4's following degree 2.24 > 2.0, so it is assigned to the first following level of M2; M3's following degree 1.2 is between 1.0 and 2.0, so it is assigned to the second following level of M2. For the dominant robot M5, its potential followers are M3 (following degree 1.8) and M4 (following degree 0.84), arranged as M3, M4. M3 is classified into the second follower level of M5 (because its follower degree of 1.8 is between 1.0 and 2.0), while M4 may not be included in the direct jurisdiction of M5 because its follower degree of 0.84 is lower than the set minimum level threshold (e.g., 1.0), or may be classified into a more lenient level.

[0072] The resulting collaborative control hierarchy is a network rather than a simple tree structure because a follower robot (such as M3) can simultaneously belong to different follower levels of multiple dominant robots. The module needs to manage this complex attribution relationship and determine when to prioritize which dominant relationship at different control stages. For example, during the stable grinding stage, the hierarchy centered on M5 might be activated, with M3 acting as its second-level follower; while when rapid parameter adjustments are needed, the hierarchy might switch to one centered on M1, leveraging the rapid response of M3 as its first-level follower. The division of control stage levels and the application of differentiated strategies are based on a deep understanding and dynamic scheduling of this complex hierarchy. The entire construction process is periodic; as the robot's priority, influence, and decision-making capabilities change, the hierarchy is dynamically updated to ensure the control system always adapts to the needs of the actual production environment.

[0073] Example 5: The verification module, acting as the gatekeeper of system data flow and control logic, is crucial for ensuring that system decisions are based on reliable data. This module continuously monitors the real-time operational data flow from the data acquisition module, checking its validity in three main aspects: data range verification, logical consistency verification, and outlier detection. Data range verification judges the physical rationality of each data type, such as whether the robot's position coordinates exceed its workspace limits, whether the applied force is within the motor's rated output range, and whether the tool wear shows a non-negative monotonically increasing trend. Logical consistency verification focuses on the reasonableness of the correlation between different data fields, such as whether the end time of a sharpening operation is later than the start time, and whether the operation duration matches the collected position change trajectory. Outlier detection uses statistical methods, such as calculating the mean and standard deviation of the data based on a moving window, marking data points that significantly deviate from the main distribution (e.g., exceeding three times the standard deviation) as suspicious. After completing the initial screening of the local operational data's validity, the verification module initiates a verification process based on global control parameters.

[0074] Global control parameters are standards or thresholds pre-set at the system level or learned historically. They represent the rules or statistical characteristics that the system as a whole should follow under normal operating conditions. Global parameters may include the allowable fluctuation range of the average operation duration of all robots, the expected value of the overall tool wear rate of the system, or the maximum allowable time synchronization error of collaborative operation between robots. The verification process compares the local operation data of a single robot with these global parameters. For example, it compares the operation duration of a robot in this instance with the global average duration and calculates the deviation rate; or it checks whether the tool wear data reported by the robot matches the trend predicted by the global wear model. The verification result directly determines subsequent operations. If the local data passes the validity check and the deviation from the global parameters is within an acceptable tolerance range, the verification is considered successful. After verification, the verification module does not simply release the data, but triggers an adjustment to the global control parameters. The adjustment is based on the current batch of valid data that has passed verification and reflects the latest state of the system. Adjustment algorithms typically employ a gradual update strategy, such as using a first-order low-pass filter or a moving average model, to slowly incorporate the information contained in new, valid data into the global parameters.

[0075] The verification module's workflow is a closed-loop process comprising several clearly defined sub-processes. The first sub-process is data reception and caching, where the module allocates a buffer to receive and temporarily store real-time data streams. The second sub-process is parallel checking, where data range verification, logical consistency checks, and outlier detection are often performed simultaneously to improve processing efficiency. The third sub-process is global comparison, comparing the initially filtered data with the current global parameter set read from the central parameter library. The fourth sub-process is decision-making and action, determining "verification passed" or "verification failed" based on the comparison results. The fifth sub-process is parameter adjustment and data routing. For data that passes verification, a smooth update of global parameters is triggered, and this "clean" data is marked as usable and sent to the status analysis module and collaborative computing module for further processing. For data that fails verification, it is marked, logged, and may trigger alarms or require the data acquisition module to resample. Assume a scenario where the system has five robotic arms, and one of the current global control parameters is an average force of 50 Newtons across all robotic arms in stable operating mode. The data acquisition module transmitted a batch of data, including a set of force data reported by the robotic arm M1, which was [52, 51, 80, 53, 49] Newtons. The verification module first performed a range check, assuming the upper limit of the force was 100 Newtons, and all data were within the range. Next, outlier detection was performed. The mean of this set of data was calculated to be 57 Newtons, with a standard deviation of approximately 12 Newtons. The value of 80 Newtons far exceeded the mean plus twice the standard deviation. Therefore, it was marked as a potential outlier. The module might remove the data at this point, interpolate using the mean of adjacent data, or directly mark this sampling as requiring caution. Then, a logical consistency check was performed, which revealed that the position data of M1 changed steadily during this time period, which might not match the force value of 80 Newtons, increasing suspicion about this outlier.

[0076] The module compares the processed data from M1 (recalculated after removing or correcting outliers, with an average force of approximately 51 Newtons) with the global parameter of 50 Newtons. The deviation is 1 Newton, within a preset tolerance (e.g., ±5 Newtons), thus the verification passes. The verification module then fine-tunes the global average force parameter, for example, slowly updating it from 50.0 Newtons to 50.1 Newtons to reflect the current slightly increasing force trend. Simultaneously, this verified data is released to downstream modules. The entire verification module design emphasizes a balance between real-time performance and robustness; the checking algorithm is optimized to complete within limited computational resources, avoiding becoming a system bottleneck. It maintains a global control parameter library and records the history of each parameter adjustment to track changes in system state. For manipulators that consistently fail verification or deviate significantly from the global parameters, the module increases their monitoring level and may send a prompt to the control optimization module, suggesting a more in-depth diagnosis of the manipulator or adjustment of its control priority. Through this continuous data verification and parameter adaptation mechanism, the verification module effectively ensures the quality of data input to the core control logic.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent multi-robotic arm collaborative control system for sharpening knives, characterized in that, The system includes: The data acquisition module is used to acquire real-time operation data of multiple robotic arms during the tool sharpening process. The real-time operation data includes position data, force data, and tool wear data. The status analysis module is used to identify stable and variable working modes based on the changes in real-time operation data of each robot over time. The collaborative computing module is used to calculate the collaborative performance index of each robot based on real-time operation data under stable and variable working modes. The control optimization module is used to generate a collaborative grinding control strategy for multiple robotic arms based on collaborative performance indicators and real-time operation data. The collaborative computing module is specifically used for: In a stable working mode, the robot with real-time operation data is marked as the reference robot. In a variable working mode, the robot with real-time operation data is marked as the target robot. For each reference robot and target robot, the control influence of the reference robot relative to the target robot is calculated based on the similarity analysis of the real-time operation data and combined with time information. The control influence of all target robots is aggregated to obtain the collaborative performance index of each reference robot. The calculation of the control influence includes: Based on the start time and operation duration of the reference robot and the target robot in each grinding operation, a similarity metric is derived; the ratio of the time spent by the target robot and the reference robot in the operation is used as the first factor, and the ratio of the similarity metric to the operation duration is used as the second factor; the product of the first factor and the second factor is calculated as the control influence.

2. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 1, characterized in that, The real-time operation data also includes the start time, end time, and duration of each sharpening operation.

3. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 2, characterized in that, The state analysis module is specifically used for: For each robotic arm, extract the real-time operation data sequence, calculate the first-order difference value of the sequence, and select the moments with positive difference values ​​as growth moments; The growth moments are sorted by the difference value to form a difference sequence. The second difference value of the difference sequence is calculated to determine the critical moments. The time period before the critical moment corresponds to the stable working mode, and the time period after the critical moment corresponds to the changing working mode.

4. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 1, characterized in that, The control optimization module is also used to calculate the decision-making ability index of each robot based on the tool wear data and operation duration in the real-time operation data; By combining collaborative performance indicators and decision-making ability indicators, the control priority of each robotic arm is obtained; Based on control priority, a dominant robotic arm is selected, and a collaborative control hierarchy is constructed.

5. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 4, characterized in that, The calculation of the decision-making ability index includes: Calculate the characteristic ratio of operation duration to interval time for each sharpening operation, and take the product of the average characteristic ratio of all operations and the number of tool wears as the decision-making ability index.

6. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 5, characterized in that, The control optimization module is specifically used for: The robot arm with a control priority higher than the set value is designated as the dominant robot arm; For each dominant robot arm, the following degree index of the target robot arm relative to the dominant robot arm is calculated. The following degree index is based on the control influence and decision-making ability index. The collaborative control hierarchy is arranged according to the magnitude of the follow-through index, and nodes at the same level in the hierarchy have the same follow-through index.

7. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 6, characterized in that, The control optimization module is also used to divide the control stage into levels according to the collaborative control hierarchy and apply differentiated control strategies for different stages.

8. The intelligent multi-robotic arm sharpening collaborative control system as described in claim 1, characterized in that, The system also includes a verification module, which checks the validity of real-time operation data, verifies local operation data based on global control parameters, and adjusts global control parameters after the verification is successful.

Images