Robotic component processing method with multimodal ai fault self-diagnosis and fault-tolerant control
Patent Information
- Application Number
- CN202610763060.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-29
- Publication Date
- 2026-08-28
Smart Images
Figure CN122653016A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of robot component processing technology, specifically to a robot component processing method with multimodal AI fault self-diagnosis and fault-tolerant control. Background Technology
[0002] In the field of robotic component manufacturing, machining centers inevitably experience various malfunctions during prolonged high-load operation, such as tool wear, bearing abnormalities, and cooling system failures. Failure to identify and address these malfunctions in a timely manner can not only lead to decreased machining accuracy and workpiece scrap, but in severe cases, it can also cause equipment damage and significant economic losses.
[0003] Currently, the main methods for anomaly detection and fault handling in the machining process are as follows: setting a single sensor threshold, such as determining the equipment status by setting fixed thresholds for physical quantities such as vibration and current. However, this method has poor adaptability, and the background of sensor signals varies greatly under different machining conditions of the machine tool, which can easily lead to false alarms or missed alarms. Relying on a single sensor signal for analysis, such as using only vibration signals to monitor tool status, but this method has a single information source, and for early and weak fault characteristics such as crack initiation, the signal is easily drowned out by background noise, resulting in insufficient detection sensitivity.
[0004] In existing technologies, data collected from different sensors, such as vibration, sound, and temperature sensors, are typically processed independently, failing to achieve effective fusion and cross-validation of multimodal sensor data. Therefore, when a sensor is affected by noise or its signal characteristics are insignificant, existing systems struggle to verify and confirm the fault using information from other sensors, limiting the reliability and accuracy of diagnostic results. In the post-fault handling phase, most systems only have alarm functions and cannot provide differentiated responses based on the severity of the fault. Once an anomaly is detected, a direct shutdown is usually implemented, leading to production interruptions. This simplistic approach lacks flexibility and causes unnecessary production losses in some minor fault scenarios.
[0005] In conclusion, how to establish an effective correlation analysis mechanism among multi-sensor data to achieve fault self-diagnosis is an urgent problem to be solved. Summary of the Invention
[0006] In view of this, this application provides a robot component processing method with multimodal AI fault self-diagnosis and fault-tolerant control, which can establish an effective correlation analysis mechanism between multi-sensor data to achieve fault self-diagnosis.
[0007] In a first aspect, this application provides a method for machining robot parts using multimodal AI fault self-diagnosis and fault-tolerant control, comprising: pre-deploying an AI database of the current machining target; acquiring sensor data of multiple modalities during the machining process, and constructing multiple data pools according to the modal type; if an abnormal state is detected in a certain sensor data, analyzing to obtain the corresponding fault type; retrieving multiple sensor data associated with the fault type from other data pools through the AI database analysis; based on the AI database analysis, if multiple sensor data all point to the abnormal state, diagnosing and generating the corresponding fault type, and summarizing the corresponding multiple sensor data to generate a fusion feature dataset; evaluating and submitting the corresponding fault rating based on the fusion feature dataset; obtaining a correction strategy for the fusion feature dataset; simulating the correction strategy, and if the fault probability of the simulation result is reduced to the fault-tolerant range, then the correction strategy is certified as passed.
[0008] In conjunction with the first aspect, in one possible implementation, the step of retrieving multiple sensor data associated with the fault type from other data pools through the AI database analysis includes: analyzing and obtaining a first fault feature matching the abnormal state based on the abnormal state-triggered attention mechanism; triggering a guidance mechanism and a corroboration mechanism based on the first fault feature to match at least two associated data pools; extracting the verification timestamp of the first fault feature; and extracting the sensor data from each of the associated data pools according to the verification timestamp.
[0009] In conjunction with the first aspect, in one possible implementation, based on the AI database analysis, if multiple sensor data points to the abnormal state, a corresponding fault type is diagnosed and generated. The process of aggregating the multiple sensor data to generate a fusion feature dataset includes: analyzing the matching degree between each sensor data point and the first fault feature based on the multiple sensor data points; if each matching degree meets a preset matching degree, it is determined that multiple sensor data points to the abnormal state; and packaging the multiple sensor data points into the fusion feature dataset.
[0010] In conjunction with the first aspect, in one possible implementation, the types of the sensing data include vibration data, sound data, temperature data, visual data, and current data; the step of triggering the attention mechanism based on the abnormal state and analyzing the first fault feature matching the abnormal state includes: when the sensing data corresponding to the abnormal state is vibration data, then triggering the attention mechanism based on the vibration data; analyzing the first fault feature corresponding to the vibration data based on the attention mechanism; the step of triggering the guidance mechanism and the corroboration mechanism based on the first fault feature and matching at least two associated data pools includes: if the first fault feature is a cutting type, then triggering the guidance mechanism based on the vibration data and calling the... The sound data is part of an associated data pool; the verification mechanism is triggered based on the sound data, and the current data is called as an associated data pool; the step of extracting the sensor data from each associated data pool according to the verification timestamp includes: marking the vibration data at the verification timestamp as an abnormal vibration parameter; extracting the abnormal sound parameter and abnormal current parameter corresponding to the timestamp from the sound data and the current data according to the verification timestamp; the step of analyzing the matching degree between each sensor data and the first fault feature based on multiple sensor data includes: analyzing the matching degree between the abnormal vibration parameter, the abnormal sound parameter, and the abnormal current parameter and multiple matching degrees with the first fault feature.
[0011] In conjunction with the first aspect, in one possible implementation, the types of the sensing data include vibration data, sound data, temperature data, visual data, and current data; the step of triggering the attention mechanism based on the abnormal state and analyzing the first fault feature matching the abnormal state includes: when the sensing data corresponding to the abnormal state is vibration data, then triggering the attention mechanism based on the vibration data; analyzing the first fault feature corresponding to the vibration data based on the attention mechanism; the step of triggering the guidance mechanism and the corroboration mechanism based on the first fault feature and matching at least two associated data pools includes: if the first fault feature is a bearing type, then triggering the guidance mechanism based on the vibration data and calling the... Temperature data forms an associated data pool; the verification mechanism is triggered based on the temperature data, and the visual data forms an associated data pool; extracting the sensor data from each associated data pool according to the verification timestamp includes: marking the vibration data at the verification timestamp as an abnormal vibration parameter; extracting the abnormal temperature parameter and abnormal visual parameter corresponding to the timestamp from the temperature data and the visual data according to the verification timestamp; analyzing the matching degree between each sensor data and the first fault feature based on multiple sensor data includes: analyzing the matching degree between the abnormal vibration parameter, the abnormal temperature parameter, and the abnormal visual parameter and multiple matching degrees with the first fault feature.
[0012] In conjunction with the first aspect, in one possible implementation, the types of the sensing data include vibration data, sound data, temperature data, visual data, and current data; the step of triggering the attention mechanism based on the abnormal state and analyzing to obtain the first fault feature matching the abnormal state includes: when the sensing data corresponding to the abnormal state is the visual data, then triggering the attention mechanism based on the vibration data; analyzing the first fault feature corresponding to the visual data based on the attention mechanism; the step of triggering the guidance mechanism and the corroboration mechanism based on the first fault feature and matching at least two associated data pools includes: if the first fault feature is coolant-related, then triggering the guidance mechanism based on the visual data and calling the... The vibration data is used as an associated data pool; the verification mechanism is triggered based on the vibration data, and the current data is used as an associated data pool; the step of extracting the sensor data from each associated data pool according to the verification timestamp includes: marking the visual data at the verification timestamp as abnormal visual parameters; extracting the abnormal vibration parameters and abnormal current parameters corresponding to the timestamp from the vibration data and the current data according to the verification timestamp; the step of analyzing the matching degree between each sensor data and the first fault feature based on multiple sensor data includes: analyzing the matching degree between the abnormal visual parameters, the abnormal vibration parameters, and the abnormal current parameters and multiple matching degrees with the first fault feature.
[0013] In conjunction with the first aspect, one possible implementation further includes: performing a preset number of simulations on the fused feature dataset to obtain the probability of occurrence of the first fault feature; and generating a corroborating confidence level based on the probability of occurrence.
[0014] In conjunction with the first aspect, one possible implementation further includes: pre-configuring association mechanisms for various types of the first fault features; mapping the associated data pools to the corresponding first fault features based on the association mechanisms; and periodically updating the association mechanisms according to an external knowledge base and / or the AI database.
[0015] In conjunction with the first aspect, one possible implementation also includes: acquiring an updated external knowledge base; and training the AI database based on the external knowledge base.
[0016] In conjunction with the first aspect, in one possible implementation, the step of simulating the correction strategy and, if the failure probability of the simulation result decreases to a tolerance range, then the correction strategy is certified. This includes: determining a tolerance threshold range corresponding to the current processing target; inputting the correction strategy into a digital twin model corresponding to the processing equipment; executing the correction strategy in the simulation environment and outputting the corresponding failure probability; if the failure probability is within the tolerance threshold range, then the correction strategy is certified; if the failure probability is not within the tolerance threshold range, then the correction strategy is deemed not certified and a shutdown command is triggered.
[0017] This application, when applied, does not rely on isolated anomaly judgments from a single sensor. Instead, after confirming a single modal anomaly, it further calls upon an AI database to analyze and extract sensor data from other related modalities for cross-validation. Finally, it determines the fault type indicated by the abnormal state and generates a fusion feature dataset. This process avoids misjudgments caused by noise interference or short-term failure of a single sensor, achieving multi-dimensional, high-confidence confirmation of the fault state and significantly improving the accuracy of fault diagnosis. Before issuing the correction strategy to the physical processing equipment, simulation and effect evaluation are performed in the established digital model. Only when the simulation results show that the fault rating can be reduced to a preset tolerance range is the strategy certified and executed, thus avoiding secondary accidents and equipment damage that may be caused by incorrect strategies or improper adjustments. This application enables the AI database to accumulate relevant data and strategy correlation knowledge after processing a fault, continuously optimizing its ability to diagnose and correct similar faults in the future. Therefore, while possessing self-diagnosis and fault-tolerant control, the system's diagnostic performance has the characteristic of self-enhancing over time. Attached Figure Description
[0018] Figure 1 The diagram shows the steps of a robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to an embodiment of this application.
[0019] Figure 2 The diagram shows the steps involved in extracting the associated data pool.
[0020] Figure 3 The diagram shows the steps involved in analyzing multimodal data.
[0021] Figure 4 The diagram shown illustrates the first type of data association process.
[0022] Figure 5 The diagram shown illustrates the second type of data association process.
[0023] Figure 6The diagram shown illustrates the third type of data association process.
[0024] Figure 7 The diagram shows the steps involved in evaluating and correcting the strategy. Detailed Implementation
[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0026] Figure 1 The diagram shown is a schematic representation of the method steps for a robot component manufacturing method with multimodal AI fault self-diagnosis and fault-tolerant control according to an embodiment of this application. This application provides a robot component manufacturing method with multimodal AI fault self-diagnosis and fault-tolerant control, such as... Figure 1 As shown, in one embodiment, the method includes: Step 101: Pre-deploy the AI database for the current processing target.
[0027] Step 102: Acquire sensor data for multiple modes during the processing and construct multiple data pools according to the mode type.
[0028] Step 103: If an abnormal state is detected in a certain sensor data, the corresponding fault type is obtained through analysis.
[0029] Step 104: Use AI database analysis to retrieve multiple sensor data related to the fault type from other data pools.
[0030] Step 105: Based on AI database analysis, if multiple sensor data all point to an abnormal state, the corresponding fault type is diagnosed and generated, and the corresponding multiple sensor data are aggregated to generate a fusion feature dataset.
[0031] Step 106: Evaluate and submit the corresponding fault rating based on the fused feature dataset.
[0032] Step 107: Obtain the correction strategy for the fused feature dataset.
[0033] Step 108: Simulate the correction strategy. If the failure probability of the simulation results is reduced to the fault tolerance range, the correction strategy is certified.
[0034] In this embodiment, steps 103 to 105 do not rely on isolated anomaly judgments from a single sensor. Instead, after confirming a single modal anomaly, it further calls the AI database to analyze and extract sensor data from other related modalities for cross-validation, ultimately determining the fault type indicated by the abnormal state, and then generating a fusion feature dataset. This process avoids misjudgments caused by noise interference or short-term failure of a single sensor, achieving multi-dimensional, high-confidence confirmation of the fault state and significantly improving the accuracy of fault diagnosis. Step 106 uses the fusion feature dataset generated in step 105 to perform fault rating to quantify the severity of the fault state. Fault ratings corresponding to various fault types can be pre-set and called in step 106. In step 107, a correction strategy for resolving the fault type can be provided through the AI database, or a correction strategy can be provided by human decision-making. In step 108, before any correction strategy is issued to the physical processing equipment, it is first simulated and evaluated in the established digital model. The strategy is only certified and executed when the simulation results show that the fault rating can be reduced to a preset fault tolerance range, thereby avoiding secondary accidents and equipment damage that may be caused by incorrect strategies or improper adjustments. This embodiment forms a dynamic closed loop of monitoring-diagnosis-strategy generation-simulation verification-execution-feedback, enabling the AI database to accumulate relevant data and strategy correlation knowledge after handling a fault, continuously optimizing its ability to diagnose and correct similar faults in the future. As a result, while the system has self-diagnosis and fault-tolerant control, its diagnostic performance has the characteristic of self-enhancing over time.
[0035] Figure 2 The diagram illustrates the steps involved in extracting a correlated data pool. Specifically, as shown... Figure 2 As shown, step 104 includes: Step 201: Based on the abnormal state triggering attention mechanism, analyze and obtain the first fault feature matching the abnormal state.
[0036] In this step, when step 103 detects an abnormal state in a certain sensor data, the system does not blindly search other data pools directly. Instead, it first performs in-depth analysis of the abnormal signal through an attention mechanism, analyzing the most prominent and identifiable features of the abnormal signal. These features are the first fault features. For example, if an abnormal increase in energy is identified in the frequency band corresponding to the vibration of the cutting teeth from the vibration signal, the first fault feature can be defined as a fault type related to cutting teeth.
[0037] Step 202: Based on the first fault characteristic, trigger the guidance mechanism and the verification mechanism to match at least two related data pools.
[0038] In this step, the first fault feature extracted in step 201 is used as a guiding signal to guide the query of sensor data from other data pools that are synchronized in time and whose features can corroborate the first fault feature. The system requires that at least two other data pools contain feature data that may be associated with it in order to perform matching verification.
[0039] Step 203: Extract the verification timestamp of the first fault feature.
[0040] Step 204: Extract sensor data from each associated data pool based on the verification timestamp.
[0041] Figure 3 The diagram illustrates the steps of a method for analyzing multimodal data. In one embodiment, as shown... Figure 3 As shown, step 105 includes: Step 301: Analyze the matching degree between each sensor data and the first fault characteristic based on multiple sensor data.
[0042] This step is the first step of cross-validation. In step 204, synchronized sensor data fragments were extracted from multiple relevant data pools. This step performs quantitative analysis on each data fragment, using AI to analyze the matching probability or likelihood between the sensor data of each modality and the first fault characteristic.
[0043] Step 302: Determine whether each matching degree meets the preset matching degree. If so, proceed to step 303.
[0044] Step 303: It is determined that multiple sensor data points to an abnormal state.
[0045] In this step, after obtaining the independent matching degree of each mode, the system introduces a preset matching degree threshold. For example, if the matching degree is the matching probability, then the preset matching degree is the probability threshold. This step, through full-modal consistency verification, eliminates the possibility of false alarms from a single sensor or random noise interference.
[0046] Step 304: Package multiple sensor data into a fusion feature dataset.
[0047] Figure 4 The diagram illustrates the first type of data association process. In one embodiment, the types of sensor data include vibration data, sound data, temperature data, visual data, and current data. Figure 4 As shown, step 201 includes: Step 401: When the sensing data corresponding to the abnormal state is vibration data, the attention mechanism is triggered based on the vibration data.
[0048] Step 402: Analyze the first fault characteristics corresponding to the vibration data based on the attention mechanism.
[0049] Step 202 includes: Step 403: Determine whether the first fault characteristic is a cutting type. If so, proceed to step 404.
[0050] Step 404: Based on the vibration data triggering guidance mechanism, call the sound data as the associated data pool.
[0051] Step 405: Based on the sound data, trigger the evidence mechanism and call the current data as the associated data pool.
[0052] Step 204 includes: Step 406: Mark the vibration data of the verification timestamp as abnormal vibration parameters.
[0053] Step 407: Based on the verification timestamp, extract the abnormal sound parameters and abnormal current parameters corresponding to the timestamp from the sound data and current data, respectively.
[0054] Step 301 includes: Step 408: Analyze the abnormal vibration parameters, abnormal sound parameters, and abnormal current parameters respectively, and their matching degree with multiple characteristics of the first fault.
[0055] In this embodiment, in steps 401-402, when the system detects abnormal vibration data, the vibration data is used as the priority mode for fault source perception. After abnormal vibration is detected, an attention mechanism is first triggered, and the vibration signal is analyzed in depth using AI to extract the first fault feature for subsequent correlation.
[0056] In steps 403-405, the system determines the type of fault based on the extracted first fault feature. If the feature belongs to the cutting category, i.e., related to the tool cutting process, the guidance mechanism is activated, prioritizing sound data as the first associated data pool. This is because the cutting process is inevitably accompanied by specific acoustic features, such as cutting sounds, and the two are physically strongly correlated. Subsequently, the system activates the corroboration mechanism, selecting current data as the second associated data pool. This is because cutting-related anomalies such as tool wear or breakage will inevitably lead to changes in the spindle motor load, which will be reflected in the current waveform.
[0057] Steps 406-407 involve data extraction based on timestamps. Using the verification timestamp extracted from the vibration signal as a benchmark, abnormal sound parameters and abnormal current parameters synchronized with the vibration anomaly timestamp are extracted from the associated sound data pool and current data pool, respectively, and the three are aligned in the time dimension.
[0058] Step 408 re-analyzes the abnormal vibration parameters, abnormal sound parameters, and abnormal current parameters, respectively, and their matching degree with the previously extracted first fault features. For example, through AI analysis, it determines the probability that the vibration data at that timestamp belongs to abnormal vibration, the probability that the sound segment at the same timestamp belongs to an abnormal sound wave frequency, and the probability that the current waveform at the same timestamp belongs to an abnormal power fluctuation.
[0059] This embodiment enhances the ability to identify complex fault types. Taking tool wear in robot component processing as an example, a single vibration signal may be difficult to identify early, subtle changes due to background noise. This embodiment, by forcing sound and current data to be synchronized in time and features, can capture weak cross-modal correlation signals that cannot be detected by a single mode, thereby improving the ability to identify hidden faults.
[0060] Figure 5 The diagram illustrates the second type of data association process. In one embodiment, the types of sensor data include vibration data, sound data, temperature data, visual data, and current data. Figure 5 As shown, step 201 includes: Step 501: When the sensing data corresponding to the abnormal state is vibration data, the attention mechanism is triggered based on the vibration data.
[0061] Step 502: Analyze the first fault characteristics corresponding to the vibration data based on the attention mechanism.
[0062] Step 202 includes: Step 503: Determine whether the first fault characteristic is a bearing-related fault. If so, proceed to step 504.
[0063] Step 504: Based on the vibration data triggering guidance mechanism, call the temperature data as the associated data pool.
[0064] Step 505: Trigger the evidence mechanism based on temperature data and call the visual data as the associated data pool.
[0065] Step 204 includes: Step 506: Mark the vibration data of the verification timestamp as abnormal vibration parameters.
[0066] Step 507: Based on the verification timestamp, extract the abnormal temperature parameters and abnormal visual parameters corresponding to the timestamp from the temperature data and visual data, respectively.
[0067] Step 301 includes: Step 508: Analyze the abnormal vibration parameters, abnormal temperature parameters, and abnormal visual parameters respectively, and their matching degree with multiple characteristics of the first fault.
[0068] In this embodiment, steps 501-502 use vibration data as the priority mode for fault source perception. After abnormal vibration is detected, an attention mechanism is first triggered. Based on AI, the vibration signal is analyzed in depth to extract the first fault feature for subsequent correlation, such as a periodic pulse signal that matches the frequency of bearing fault characteristics.
[0069] In steps 503-505, the system determines the type of fault based on the extracted first fault feature. If the feature belongs to the bearing category, i.e., related to bearing damage or wear, the guiding mechanism is activated, prioritizing temperature data as the first associated data pool. This is because abnormal friction or wear in the bearing is inevitably accompanied by localized temperature rise, and the two have a strong causal relationship in physics. Subsequently, the system activates the corroboration mechanism, selecting visual data (such as infrared thermography) as the second associated data pool. Visual data can provide spatial temperature distribution information in the bearing area, thus providing more direct evidence for the phenomenon of localized overheating.
[0070] Steps 506-507 extract data based on timestamps. Using the verification timestamp extracted from the vibration signal as a benchmark, abnormal temperature parameters and abnormal visual parameters synchronized with the vibration anomaly timestamp are extracted from the associated temperature data pool and visual data pool, respectively, thereby aligning the vibration phenomenon with the thermal phenomenon in the time dimension.
[0071] Step 508 involves another multimodal matching quantification evaluation. The system calculates the matching degree between the abnormal vibration parameters, abnormal temperature parameters, and abnormal visual parameters and the previously extracted first fault features. For example, through AI analysis, it determines the probability that the vibration parameters at that time stamp belong to abnormal vibration, the probability that the temperature parameters at the same time stamp belong to abnormal changes, and the probability that the visual image at the same time stamp belongs to an abnormal heat diffusion pattern.
[0072] Compared to point-based sensing that relies solely on vibration and temperature, the visual data introduced in this embodiment provides spatial distribution information. It not only identifies temperature increases but also the location of the temperature-increased area, allowing for further precision down to specific locations on the outer ring of the bearing where localized overheating exists, thus achieving refined fault location.
[0073] Figure 6 The diagram illustrates the third type of data association process. In one embodiment, the types of sensor data include vibration data, sound data, temperature data, visual data, and current data. Figure 6 As shown, step 201 includes: Step 601: When the sensor data corresponding to the abnormal state is visual data, the attention mechanism is triggered based on the vibration data.
[0074] Step 602: Analyze the first fault features corresponding to the visual data based on the attention mechanism.
[0075] Step 202 includes: Step 603: Determine whether the first fault characteristic is related to coolant. If so, proceed to step 604.
[0076] Step 604: Based on the visual data triggering guidance mechanism, call the vibration data as the associated data pool.
[0077] Step 605: Based on the vibration data, trigger the verification mechanism and call the current data as the associated data pool.
[0078] Step 204 includes: Step 606: Mark the visual data of the verification timestamp as an anomalous visual parameter.
[0079] Step 607: Based on the verification timestamp, extract the abnormal vibration parameters and abnormal current parameters corresponding to the timestamp from the vibration data and current data, respectively.
[0080] Step 301 includes: Step 608: Analyze the abnormal visual parameters, abnormal vibration parameters, and abnormal current parameters respectively, and their matching degree with the first fault characteristics.
[0081] In this embodiment, steps 601-602 use visual data as the priority mode for fault source perception. When a visual anomaly is detected, the attention mechanism is first triggered. Based on AI, the visual signal is deeply analyzed to extract the first fault feature for subsequent association. For example, the blue-purple abnormal spark feature in the image, the uneven color feature on the workpiece surface, or the built-up edge feature caused by poor chip removal.
[0082] Steps 603-605 determine the type of fault based on the extracted first fault feature. If the feature pertains to coolant, i.e., related to coolant deficiency, insufficiency, or nozzle blockage, insufficient coolant will lead to rapid tool wear and workpiece burning. In this case, a guiding mechanism is activated, prioritizing vibration data as the first associated data pool. This is because changes in dry friction or cutting force caused by coolant abnormalities are directly reflected in the acoustic emission components of the vibration signal or energy changes in specific frequency bands, and the two have a strong causal relationship in physics. Subsequently, the system activates a corroboration mechanism, selecting current data as the second associated data pool. Insufficient coolant leads to unstable cutting loads, which in turn causes fluctuations in the spindle motor current, thus providing more direct corroboration for the judgment of cooling abnormalities.
[0083] Steps 606-607 use the verification timestamp extracted from the visual signal as a reference to extract abnormal vibration parameters and abnormal current parameters synchronized with the visual anomaly timestamp from the associated vibration data pool and current data pool, thereby aligning the visual phenomenon with the vibration and electrical phenomena in the time dimension.
[0084] Step 608 recalculates the matching degree between the abnormal visual parameters, abnormal vibration parameters, and abnormal current parameters and the previously extracted first fault features. For example, through AI analysis, it determines the probability that the visual data at this timestamp belongs to abnormal vision, the probability that dry friction acoustic emission features appear in the vibration signal at the same timestamp, and the probability that the current waveform at the same timestamp shows load fluctuations.
[0085] Traditional coolant detection relies on level gauges, which are prone to false alarms due to bubbles, foam, or level fluctuations. This embodiment introduces visual data as the primary mode to directly monitor physical phenomena directly related to cooling performance, such as spark color and workpiece surface condition, thus avoiding the inherent limitations of level gauge detection. This embodiment correlates and verifies data from three different physical dimensions: visual, vibration, and current. The single fault source of coolant deficiency will simultaneously trigger visual anomalies (sparks / discoloration), vibration anomalies (dry friction), and current anomalies (unstable fluctuations). These three factors corroborate each other, resulting in high confidence in the diagnostic results and significantly reducing the false alarm rate caused by single-mode sensor failure or environmental interference.
[0086] In one embodiment, the multimodal AI fault self-diagnosis and fault-tolerant control robot component manufacturing method further includes: Step 701: Perform a preset number of simulations on the fused feature dataset to obtain the probability of the occurrence of the first fault feature.
[0087] Step 702: Generate supporting confidence based on the probability of occurrence.
[0088] In this embodiment, the fused feature dataset generated in step 105 is used as input, and a preset number of Monte Carlo simulations (e.g., 100 or 200) are performed using a virtual simulation engine preset in the AI model. For each simulation, the system uses the fused feature dataset as the simulation baseline parameter and counts how many times the first fault feature is triggered after the preset number of simulations, thus obtaining the probability of occurrence. Based on the number of occurrences of the first fault feature obtained in step 701 in multiple simulations, its probability of occurrence is calculated. For example, if it occurs 95 times in 100 simulations, the probability of occurrence is 95%. This probability of occurrence is directly used as the confidence level of the current diagnostic conclusion. If the probability value is high, for example, greater than the preset 95% threshold, it indicates that the currently diagnosed first fault feature can be stably reproduced under various operating condition disturbances, and its credibility is extremely high; conversely, if the probability value is low, it indicates that the feature may be affected by random noise or data anomalies, and its reliability is questionable.
[0089] In actual processing, sensors may generate occasional abnormal data spikes due to electromagnetic interference, instantaneous vibration and impact, etc. This embodiment can effectively distinguish between real physical fault characteristics and occasional data noise through multiple simulations and statistical analysis. Only features that can be stably reproduced in multiple simulations are given a high degree of corroboration confidence, thereby significantly reducing the probability of false alarms triggered by a single data anomaly.
[0090] In one embodiment, the multimodal AI fault self-diagnosis and fault-tolerant control robot component manufacturing method further includes: Step 801: Pre-configure the association mechanism for various first fault characteristics.
[0091] Step 802: Based on the association mechanism, map the associated data pool to the corresponding first fault feature.
[0092] Step 803: Regularly update the association mechanism based on external knowledge bases and / or AI databases.
[0093] In one embodiment, the multimodal AI fault self-diagnosis and fault-tolerant control robot component manufacturing method further includes: Step 901: Obtain the updated external knowledge base.
[0094] Step 902: Train the AI database based on an external knowledge base.
[0095] Steps 801-803 pre-configure the correspondence between different fault characteristics (such as cutting, bearing, and coolant types) and data pools (sound, current, temperature, vision, etc.). When a fault characteristic is extracted, the system directly calls the corresponding data pool according to preset rules. Simultaneously, this correspondence can be periodically updated through an external knowledge base or the system's own learning results to adapt to new fault types or sensor configurations. Through preset association rules, the system does not need to dynamically infer the data pool correspondence during each diagnosis; it can directly and quickly call the data, reducing computation time. By periodically updating the association rules and training the model, the system can adapt to new fault types, extending the system's effective lifespan.
[0096] In steps 901-902, external knowledge bases can be acquired, such as publicly available industry fault data and maintenance records of the same model of equipment. This data is used to supplement the training of the AI diagnostic model, enhancing its ability to identify unknown or rare fault modes. By introducing external knowledge base training, the model can learn existing fault knowledge within the industry, improving its accuracy in identifying novel fault modes that it has never seen before.
[0097] Figure 7 The diagram illustrates the steps involved in evaluating a correction strategy. In one embodiment, as shown... Figure 7 As shown, step 108 includes: Step 1001: Determine the fault tolerance threshold range corresponding to the current processing target.
[0098] Step 1002: Input the correction strategy into a digital twin model corresponding to the processing equipment, execute the correction strategy in the simulation environment and output the corresponding failure probability.
[0099] Step 1003: Determine if the probability of failure is... exist If the error threshold is within the range, proceed to step 1004; otherwise, proceed to step 1005.
[0100] Step 1004: Determine if the correction strategy has passed authentication.
[0101] Step 1005: Determine that the correction strategy has failed authentication and trigger a shutdown command.
[0102] In this embodiment, before executing any correction strategy, the system first performs virtual verification using a digital twin model to ensure the security of the strategy. Step 1001 determines an acceptable fault tolerance threshold range based on the priority and process requirements of the current processing task, serving as a quantitative basis for subsequent judgments.
[0103] Step 1002 inputs the correction strategy (such as adjusting the spindle speed or modifying the feed rate) generated in Step 107 into a digital twin model consistent with the physical machining equipment. The correction strategy is simulated and executed in the virtual environment, and the failure probability of occurrence is output. Steps 1003-1005 compare the simulated failure probability with a preset fault tolerance threshold range. If the simulated failure probability falls within this range, the correction strategy is deemed safe and effective, and Step 1004 is executed to send it to the physical equipment. If it does not fall within this range, the correction strategy is deemed invalid or too risky, and Step 1005 is executed to trigger a safety shutdown command to avoid damage to the equipment or workpiece.
[0104] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.
[0105] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.
[0106] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.
[0107] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features of the invention herein.
[0108] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications or equivalent substitutions made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for machining robot parts using multimodal AI fault self-diagnosis and fault-tolerant control, characterized in that, include: Pre-deploy an AI database for the current processing target; Acquire sensor data from multiple modes during the processing and construct multiple data pools according to the mode type; If an abnormal state is detected in a certain sensor data, the corresponding fault type is obtained through analysis; The AI database analysis retrieves multiple sensor data related to the fault type from other data pools. Based on the AI database analysis, if multiple sensor data points to the abnormal state, a corresponding fault type is diagnosed and generated, and the corresponding multiple sensor data are aggregated to generate a fusion feature dataset. The corresponding fault rating is evaluated and submitted based on the fused feature dataset. Obtain a correction strategy for the fused feature dataset; The correction strategy is simulated. If the failure probability of the simulation results is reduced to the fault tolerance range, the correction strategy is certified.
2. The robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to claim 1, characterized in that, The step of retrieving multiple sensor data related to the fault type from other data pools through the AI database analysis includes: Based on the abnormal state triggering attention mechanism, the first fault feature matching the abnormal state is obtained through analysis; Based on the first fault characteristic triggering guidance mechanism and corroboration mechanism, at least two associated data pools are matched; Extract the verification timestamp of the first fault feature; Based on the verification timestamp, the sensing data is extracted from each of the associated data pools.
3. The robot component processing method with multimodal AI fault self-diagnosis and fault-tolerant control according to claim 2, characterized in that, Based on the AI database analysis, if multiple sensor data points to the abnormal state, a corresponding fault type is diagnosed and generated. The corresponding multiple sensor data are then aggregated to generate a fusion feature dataset, including: The matching degree between each of the sensor data and the first fault feature is analyzed based on multiple sensor data. If each of the matching degrees meets the preset matching degree, then it is determined that multiple of the sensor data points to the abnormal state; The multiple sensor data are packaged into the fused feature dataset.
4. The robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to claim 3, characterized in that, The types of sensing data include vibration data, sound data, temperature data, visual data, and current data; the first fault feature obtained by analyzing the abnormal state matching based on the abnormal state-triggered attention mechanism includes: When the sensing data corresponding to the abnormal state is the vibration data, the attention mechanism is triggered based on the vibration data. The first fault characteristic corresponding to the vibration data is analyzed based on the attention mechanism. The triggering mechanism and corroboration mechanism based on the first fault characteristics, matching at least two associated data pools, include: If the first fault characteristic is a cutting type, then the guidance mechanism is triggered based on the vibration data, and the sound data is called as the associated data pool; The evidence mechanism is triggered based on the sound data, and the current data is used as the associated data pool. The step of extracting the sensing data from each of the associated data pools according to the verification timestamp includes: The vibration data at the verification timestamp is marked as abnormal vibration parameters; Based on the verification timestamp, abnormal sound parameters and abnormal current parameters corresponding to the timestamp are extracted from the sound data and the current data, respectively. The analysis of the matching degree between each of the multiple sensor data and the first fault feature includes: The abnormal vibration parameters, abnormal sound parameters, and abnormal current parameters are analyzed separately, and their matching degrees with multiple of the first fault characteristics are determined.
5. The robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to claim 3, characterized in that, The types of sensing data include vibration data, sound data, temperature data, visual data, and current data; the first fault feature obtained by analyzing the abnormal state matching based on the abnormal state-triggered attention mechanism includes: When the sensing data corresponding to the abnormal state is the vibration data, the attention mechanism is triggered based on the vibration data. The first fault characteristic corresponding to the vibration data is analyzed based on the attention mechanism. The triggering mechanism and corroboration mechanism based on the first fault characteristics, matching at least two associated data pools, include: If the first fault characteristic is a bearing-related fault, then the guidance mechanism is triggered based on the vibration data, and the temperature data is used as the associated data pool. The verification mechanism is triggered based on the temperature data, and the visual data is used as the associated data pool. The step of extracting the sensing data from each of the associated data pools according to the verification timestamp includes: The vibration data at the verification timestamp is marked as abnormal vibration parameters; Based on the verification timestamp, abnormal temperature parameters and abnormal visual parameters corresponding to the timestamp are extracted from the temperature data and the visual data, respectively. The analysis of the matching degree between each of the multiple sensor data and the first fault feature includes: The abnormal vibration parameters, abnormal temperature parameters, and abnormal visual parameters are analyzed separately, and their matching degrees with multiple of the first fault characteristics are determined.
6. The robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to claim 3, characterized in that, The types of sensing data include vibration data, sound data, temperature data, visual data, and current data; the first fault feature obtained by analyzing the abnormal state matching based on the abnormal state-triggered attention mechanism includes: When the sensing data corresponding to the abnormal state is the visual data, the attention mechanism is triggered based on the vibration data. The first fault feature corresponding to the visual data is analyzed based on the attention mechanism. The triggering mechanism and corroboration mechanism based on the first fault characteristics, matching at least two associated data pools, include: If the first fault characteristic is related to coolant, then the guidance mechanism is triggered based on the visual data, and the vibration data is used as the associated data pool; The verification mechanism is triggered based on the vibration data, and the current data is used as the associated data pool. The step of extracting the sensing data from each of the associated data pools according to the verification timestamp includes: The visual data of the verification timestamp is marked as an abnormal visual parameter; Based on the verification timestamp, abnormal vibration parameters and abnormal current parameters corresponding to the timestamp are extracted from the vibration data and the current data, respectively. The analysis of the matching degree between each of the multiple sensor data and the first fault feature includes: The abnormal visual parameters, abnormal vibration parameters, and abnormal current parameters are analyzed separately, and their matching degrees with multiple of the first fault characteristics are determined.
7. The robot component processing method with multimodal AI fault self-diagnosis and fault-tolerant control according to claim 2, characterized in that, Also includes: The probability of triggering the first fault feature is obtained by performing a preset number of simulations on the fused feature dataset. The corroborating confidence level is generated based on the probability of occurrence.
8. The robot component processing method with multimodal AI fault self-diagnosis and fault-tolerant control according to claim 2, characterized in that, Also includes: Pre-configure association mechanisms for various types of the first fault characteristics; Based on the association mechanism, the associated data pool is mapped to the corresponding first fault feature; The association mechanism is updated periodically based on an external knowledge base and / or the AI database.
9. The robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to claim 1, characterized in that, Also includes: Get updated external knowledge bases; The AI database is trained based on the external knowledge base.
10. The robot component processing method for multimodal AI fault self-diagnosis and fault-tolerant control according to claim 1, characterized in that, The step of simulating the correction strategy, and if the failure probability of the simulation results decreases to a fault-tolerant range, then the certification of the correction strategy includes: Determine the fault tolerance threshold range corresponding to the current processing target; The correction strategy is input into a digital twin model corresponding to the processing equipment, and the correction strategy is executed in a simulation environment to output the corresponding failure probability. If the failure probability is within the fault tolerance threshold, then the correction strategy is deemed to have passed authentication. If the failure probability is not within the fault tolerance threshold, the correction strategy is determined to have failed authentication, and a shutdown command is triggered.