Industrial robot real-time maintenance system and method combined with edge calculation

Abstract

The invention relates to the field of industrial robots, and discloses an industrial robot real-time maintenance system and method in combination with edge computing, and the method comprises the steps: obtaining the multi-source operation state data of an industrial robot, and constructing an operation feature data set of key parts of the robot in combination with a state coding mechanism of an edge end and a feature coupling analysis method; carrying out rapid distributed processing on the operation characteristic data set, constructing an equipment health state model based on a lightweight time sequence modeling algorithm, and introducing a multi-dimensional correlation analysis mechanism to carry out incremental learning on the model; judging whether the edge side state recognition result is stable or not based on the change trend of the trigger frequency; according to the corrected state mapping relation, performing response level division on the potential fault trend by applying a multi-scale fault prediction mechanism, and extracting matched maintenance plan parameters; and based on the maintenance scheduling plan, in combination with a preset fault handling knowledge base, performing automatic evaluation and optimization on the current maintenance strategy. The method has the advantage of improving the response speed.

Description

Technical Field

[0001] The present invention relates to the field of industrial robots, and particularly to a real-time maintenance system and method for industrial robots combined with edge computing. Background Art

[0002] With the continuous improvement of industrial automation levels, industrial robots have been widely used in fields such as manufacturing, assembly, and inspection. To ensure their efficient and stable operation, it is necessary to monitor and maintain their key components, sensors, and control systems in real time. Currently, most industrial robot maintenance systems rely on cloud computing platforms, where operation data is uploaded to the cloud for analysis, diagnosis, and maintenance decision-making. However, this architecture has certain limitations in practical applications, especially in scenarios with high requirements for the timeliness of fault handling. Since data needs to be transmitted to remote cloud servers, the system often fails to immediately identify and respond to critical abnormal states when facing problems such as insufficient network bandwidth, latency fluctuations, or disconnections. For example, when there is a slight abnormality in the execution mechanism of a robot, since the diagnostic logic is deployed in the cloud, the system may not be able to generate warning information in a timely manner, leading to the expansion of faults and even equipment damage. Therefore, it is necessary to design a real-time maintenance system and method for industrial robots combined with edge computing to improve the response speed. Summary of the Invention

[0003] (1) Technical Problems to be Solved

[0004] Aiming at the deficiencies of the prior art, the present invention provides a real-time maintenance system and method for industrial robots combined with edge computing, which has the advantage of improving the response speed and solves the problems in the above background art.

[0005] (2) Technical Solutions

[0006] To achieve the above purpose of improving the response speed, the present invention provides the following technical solutions: A real-time maintenance method for industrial robots combined with edge computing, comprising the following steps:

[0007] Obtain multi-source operation status data of the industrial robot, and construct an operation feature data set of the robot's key components by combining the status encoding mechanism at the edge and the feature coupling analysis method;

[0008] Perform fast distributed processing on the operation feature data set, construct a device health status model based on a lightweight time series modeling algorithm, introduce a multi-dimensional correlation analysis mechanism for incremental learning of the model, and monitor the change in the triggering frequency of abnormal features in real time during the model update process;

[0009] Based on the changing trend of the triggering frequency, determine whether the edge - side status recognition result is stable. If it is stable, record the changes in the key dimension status under the current working condition of the robot, and in combination with the working condition perception mechanism, perform real - time correction on the maintenance status mapping relationship;

[0010] According to the corrected status mapping relationship, apply a multi - scale fault prediction mechanism to divide the potential fault trend into response levels, extract the matching maintenance plan parameters, and dynamically adjust the response strategy to form a maintenance scheduling plan for the edge side;

[0011] Based on the maintenance scheduling plan, in combination with the preset fault handling knowledge base, automatically evaluate and optimize the current maintenance strategy, and generate equipment maintenance suggestions in real - time.

[0012] Preferably, the process of constructing the operation feature dataset of the key components of the robot is as follows:

[0013] Perform data pre - processing on the obtained multi - source operation status data;

[0014] Perform normalization processing on continuous operation indicators;

[0015] Convert discrete state features using mechanisms such as one - hot encoding or label encoding to form a structured state feature set;

[0016] Use the state encoding deployed at the edge side to perform structural embedding expression on the pre - processed feature set to improve its adaptability to edge computing resources;

[0017] Based on the feature coupling analysis method, identify the multiple coupling relationships between operation indicators, and present the dynamic association between key variables in the form of a coupling strength matrix;

[0018] Combined with the results of the coupling matrix, screen out the high - coupling - degree feature combinations, and finally construct an operation feature dataset reflecting the operation health status of the key components of the robot.

[0019] Preferably, the process of fast distributed processing and lightweight time - series modeling algorithm is as follows:

[0020] Divide the constructed operation feature dataset into multiple edge subtasks and adopt a multi - thread concurrent processing mechanism;

[0021] Introduce a lightweight edge computing framework and deploy it on the terminal nodes;

[0022] For the processed operation data sequence, use a local modeling method based on a sliding window to construct a short - time - series health trend model;

[0023] Adopt a gated recurrent unit to model the feature fluctuations in the state sequence and capture potential abnormal change patterns;

[0024] The outputs of multiple source sub-models are aggregated and a unified device health status model is constructed through the feature aggregation module on the edge side.

[0025] Preferably, the process of introducing a multi-dimensional correlation analysis mechanism to perform incremental learning on the equipment health status model is as follows:

[0026] After the initial training is completed, each newly acquired running data segment is updated online to achieve adaptive evolution of the model;

[0027] Combining the historical synergistic change relationships between dimensions, we construct a multi-dimensional interaction map and explore the potential causal relationships between key components.

[0028] The attention mechanism is introduced to assign higher weights to feature dimensions with higher correlation strength, thus achieving focused modeling of device status features.

[0029] During the incremental learning process, a dynamic batch update strategy is used to control the weight update amplitude of each model learning.

[0030] Preferably, the trigger frequency change process of real-time monitoring abnormal characteristics is:

[0031] During the incremental learning training cycle, feature dimensions identified as abnormal in the model input features are extracted in real time;

[0032] Based on the time window mechanism, statistical analysis is performed on the triggering frequency of abnormal features within the window period;

[0033] A dynamic change threshold strategy is introduced to identify whether there is a frequency offset in the current cycle based on the abnormal frequency baseline under the system's historical stable operation cycle;

[0034] When the frequency deviates, the corresponding characteristics and timestamps are automatically recorded and included in the maintenance plan as a potential failure sign;

[0035] Combined with the distribution range and aggregation trend of abnormal features, it helps to evaluate the fluctuation degree of the robot's current operating status.

[0036] Preferably, the process of judging whether the edge side state recognition result is stable is:

[0037] Smooth the abnormal characteristic frequency change curve and extract the change rate index;

[0038] If the absolute value of the abnormal frequency change rate is less than the preset threshold, the edge side state recognition result is determined to be stable;

[0039] If the absolute value of the abnormal frequency change rate is greater than or equal to a preset threshold, it is determined that the edge side state recognition result is unstable.

[0040] Preferably, the process of real-time correction of the maintenance status mapping relationship in combination with the working condition perception mechanism is as follows:

[0041] Record the key operating parameters and corresponding equipment status labels in the stable state to form a phased state perception segment;

[0042] In each perception segment, collect the working condition background information at that time;

[0043] Construct a working condition perception factor matrix, and use the principal component analysis or clustering algorithm to identify typical working condition patterns;

[0044] Combine the corresponding relationship between each working condition pattern and the status label, and dynamically adjust the original status-maintenance behavior mapping strategy.

[0045] Preferably, the process of dividing the response levels for potential fault trends by applying the multi-scale fault prediction mechanism is as follows:

[0046] Mine the fault trend from the corrected state mapping results, and use the multi-scale sliding window mechanism for time-domain resampling;

[0047] At each time scale, train an independent state evolution prediction sub-model, and use the attention fusion mechanism to integrate the prediction results of each scale;

[0048] Analyze the change rate and amplitude of the predicted potential fault trend, and divide the risk level according to the set threshold;

[0049] Extract the maintenance parameter templates corresponding to different response levels, including the maintenance scope, required resources, and response time.

[0050] Preferably, the process of real-time generating equipment maintenance suggestions includes:

[0051] Based on the maintenance plan parameter set, match the maintenance rules and case base in the fault handling knowledge base;

[0052] Call the expert system module, compare the current system state with historical cases, and identify the optimal maintenance path;

[0053] Combine the current operation plan of the robot, the available time window, and the maintenance resource configuration, and automatically generate a maintenance scheduling plan;

[0054] On the basis of generating the scheduling plan, analyze the impact of the scheduling task on the overall production rhythm and put forward optimization suggestions;

[0055] Finally, automatically output structured equipment maintenance suggestions.

[0056] An industrial robot real-time maintenance system combined with edge computing, including:

[0057] Feature construction module: Collect and fuse multi-source status data, and construct a structured operation feature dataset of key components of the robot through status encoding and feature coupling analysis;

[0058] Incremental learning module: Establish a device health model based on distributed processing and lightweight time series modeling algorithms, and continuously optimize the model performance and anomaly monitoring capabilities through incremental learning;

[0059] Status recognition module: Analyze the change trend of the abnormal feature trigger frequency, judge the stability of status recognition, and dynamically correct the maintenance behavior mapping relationship in combination with the working condition perception mechanism;

[0060] Fault prediction module: Perform multi-scale fault trend prediction according to the corrected status information and automatically divide the response levels;

[0061] Suggestion generation module: Rely on the maintenance knowledge base to automatically evaluate and adjust the current strategy, and generate intelligent maintenance suggestions that match the actual working conditions in real time.

[0062] (3) Beneficial effects

[0063] Compared with the prior art, the present invention provides an industrial robot real-time maintenance system and method combined with edge computing, having the following beneficial effects:

[0064] By deploying a status encoding and feature coupling analysis mechanism at the edge side, the present invention improves the real-time performance and accuracy of the extraction of key component operation features; introduces a lightweight time series modeling and multi-dimensional incremental learning strategy to realize the dynamic evolution modeling of the device health status and the instant response to abnormal behaviors; effectively avoids misjudgment and waste of maintenance resources by identifying the stability of status recognition through the frequency change trend; dynamically corrects the maintenance mapping relationship in combination with the working condition perception, making the maintenance behavior more suitable for the current operating environment; further enhances the system's early warning and maintenance scheduling capabilities for potential faults through multi-scale fault prediction and dynamic strategy adjustment; and finally improves the rationality and execution efficiency of maintenance suggestions through a knowledge-driven maintenance optimization mechanism. It has the advantages of fast response, flexible deployment, strong scalability, and good edge intelligence adaptability, significantly improving the autonomy, intelligence, and precision levels of industrial robot maintenance. Description of the drawings

[0065] Figure 1 It is a schematic diagram of the method of the present invention;

[0066] Figure 2 It is a schematic diagram of the structure of the present invention. Detailed implementation manners

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0068] Embodiment 1: Please refer to Figure 1 As shown, a real-time maintenance method for an industrial robot combined with edge computing according to an embodiment of the present invention includes the following steps:

[0069] S1: Obtain multi-source operation status data of the industrial robot, and construct an operation feature dataset of key components of the robot by combining the status encoding mechanism at the edge side and the feature coupling analysis method.

[0070] The process of constructing the operation feature dataset of key components of the robot in S1 is as follows:

[0071] Perform data preprocessing on the obtained multi-source operation status data, covering operations such as noise filtering, outlier detection, and missing data filling;

[0072] Perform normalization processing on continuous operation indicators to ensure a unified scale standard for different feature dimensions;

[0073] Convert discrete state features using mechanisms such as one-hot encoding or label encoding to form a structured state feature set; for discrete state features with fixed classification labels, select one-hot encoding or label encoding methods for processing according to the number of categories and model adaptability; the converted discrete features form a structured binary or integer feature vector;

[0074] Use the status encoding deployed at the edge side to perform structural embedding expression on the preprocessed feature set to improve its adaptability to edge computing resources; input the feature set that has been encoded and preprocessed into the edge-side status encoder, and use a low-dimensional embedding algorithm to compress and express the features; the embedding expression form takes into account both computational efficiency and feature fidelity, and effectively adapts to the operation environment with limited edge computing resources;

[0075] Based on the feature coupling analysis method, identify the multiple coupling relationships between operation indicators, and present the dynamic association between key variables in the form of a coupling strength matrix; use correlation calculation to quantify the degree of dynamic association between features; construct a coupling strength matrix, and present the coupling strength and change trend between key operation indicators in the form of a two-dimensional heat map or a sparse matrix;

[0076] Combined with the coupling matrix results, high-coupling feature combinations are screened out, and finally an operation feature dataset reflecting the running health status of the robot's key components is constructed; a coupling strength threshold is set, feature pairs higher than the threshold are screened out, and the feature combination that can best reflect the health status of the robot's key components is identified; the screened high-correlation feature set is aggregated to form the final operation feature dataset.

[0077] S2: Perform fast distributed processing on the operation feature dataset, construct a device health status model based on a lightweight time series modeling algorithm, introduce a multi-dimensional correlation analysis mechanism for incremental learning of the model, and monitor the change in the triggering frequency of abnormal features in real time during the model update process.

[0078] The processes of fast distributed processing and lightweight time series modeling algorithm in S2 are as follows:

[0079] Divide the constructed operation feature dataset into multiple edge subtasks, and use a multi-threaded concurrent processing mechanism to achieve efficient distributed processing of feature data;

[0080] Introduce a lightweight edge computing framework and deploy it on the terminal node to reduce the model calculation complexity;

[0081] For the processed operation data sequence, use a local modeling method based on a sliding window to construct a short time series health trend model;

[0082] Adopt a gated recurrent unit to model the feature fluctuations in the state sequence and capture potential abnormal change patterns;

[0083] Summarize the outputs of multi-source sub-models, and form a unified device health status model through a feature aggregation module on the edge side, which serves as the basis for subsequent incremental learning.

[0084] The process of introducing a multi-dimensional correlation analysis mechanism for incremental learning of the device health status model in S2 is as follows:

[0085] After the initial training is completed, perform an online update operation on each newly obtained operation data segment to achieve the adaptive evolution of the model; set the initial training stage of the model as a complete training cycle, obtain a representative historical operation dataset, and complete the basic state modeling; during actual operation, the edge side periodically receives the latest operation data segment, converts it into a standardized feature format, and inputs it as incremental data; use an online learning algorithm to continuously optimize the model parameters to make the model adapt to the dynamic changes of the operation state; during the update process, monitor the change trend of the model loss function to determine whether the model has effectively learned the new data features and prevent overfitting or drift phenomena from occurring;

[0086] By combining historical synergistic change relationships between dimensions, a multidimensional interaction graph is constructed to explore potential causal relationships between key components. Based on historical operating data, the synergistic change trends between multiple operating dimensions are statistically analyzed, and potential correlations are identified through methods such as time series cross-correlation analysis and Granger causality tests. A graph structure is constructed with key operating indicators as nodes and interaction relationships as edges, using directed edges to represent potential causal paths. The multidimensional interaction graph provides structured information input, which helps capture the coupled evolution characteristics between key components and improves the causal interpretability and foresight of state modeling.

[0087] The attention mechanism is introduced to assign higher weights to feature dimensions with higher correlation strength, enabling focused modeling of device state features. The correlation strength score is calculated for each feature dimension. Key dimensions with high correlation strength scores are given higher attention weights during the model input phase to enhance their contribution to state prediction results. The attention mechanism automatically updates the weight distribution during training to ensure that the model focuses on feature dimensions that play a decisive role in state fluctuations.

[0088] During the incremental learning process, a dynamic batch update strategy is used to control the weight update amplitude of each model learning to ensure the stability and responsiveness of the training process. The sample batch size of each round of incremental update is adaptively adjusted according to the current system load and model learning status. A sliding learning rate mechanism is used for model weight updates, and a learning rate decay strategy or triggered adjustment rules are set to prevent violent oscillations during the training process. Combined with the real-time computing capabilities of the edge side and the acceptable delay range, the training accuracy and computing resource consumption are dynamically balanced to ensure the real-time and stability of model evolution.

[0089] The trigger frequency change process of real-time monitoring of abnormal characteristics in S2 is as follows:

[0090] During the incremental learning training cycle, feature dimensions identified as "abnormal" from the model input features are extracted in real time. Before each round of model update, anomaly detection is performed on the input operating status features, and anomalies are identified using statistical deviations, threshold crossings, or a discrimination mechanism based on model residuals. Feature dimensions that deviate from the normal operating range within the current time period are marked, and their anomaly type and degree of numerical deviation are recorded. Feature dimensions identified as abnormal are temporarily cached in the edge status monitoring module as objects for subsequent frequency statistics.

[0091] Based on the time window mechanism, statistical analysis is performed on the triggering frequency of abnormal features within the window period. A fixed or sliding time window is set to count the number of occurrences of abnormal features of each dimension within the window. The statistical results are normalized into the abnormal triggering frequency per unit time to reflect the level of abnormal activity in the current operation phase. If multiple abnormal feature dimensions are repeatedly triggered within the same window period, it can be considered that there is a potential risk of short-term fluctuation or instability in the system.

[0092] Introduce a dynamic change threshold strategy. Based on the abnormal frequency baseline under the historical stable operation period of the system, identify whether there is a frequency offset in the current period; use the historical abnormal frequency data collected during the stable operation stage of the system to construct a statistical baseline for the normal fluctuation range; design a dynamic threshold calculation rule, and combine the recent frequency mean, fluctuation amplitude and distribution characteristics to dynamically generate the judgment threshold for the current window period; when the abnormal frequency of a certain feature in the current window period is significantly higher than the historical baseline, it is determined that a frequency offset has occurred, indicating that the health status of the equipment is deteriorating;

[0093] When the frequency offset is significant, automatically record the corresponding feature and timestamp, and include it as a potential failure symptom in the consideration scope of the maintenance plan;

[0094] At the same time, combine the distribution range and aggregation trend of abnormal features to assist in evaluating the fluctuation degree of the current operating state of the robot.

[0095] Example 2: As Figure 1 shown, a real-time maintenance method for an industrial robot combined with edge computing further includes the following steps:

[0096] S3: Based on the change trend of the trigger frequency, judge whether the status recognition result on the edge side is stable. If it is stable, record the key dimension status changes under the current working condition of the robot, and combine the working condition perception mechanism to perform real-time correction on the maintenance status mapping relationship.

[0097] In the above S3, the process of judging whether the status recognition result on the edge side is stable is as follows:

[0098] Smooth the abnormal feature frequency change curve and extract the change rate index;

[0099] If the absolute value of the abnormal frequency change rate is less than the preset threshold, it is determined that the status recognition result on the edge side is stable;

[0100] If the absolute value of the abnormal frequency change rate is greater than or equal to the preset threshold, it is determined that the status recognition result on the edge side is unstable.

[0101] In the above S3, the process of performing real-time correction on the maintenance status mapping relationship by combining the working condition perception mechanism is as follows:

[0102] Record the key operating parameters and corresponding equipment status labels in the stable state to form a stage state perception segment; during the time period when the equipment operating state recognition result is stable, extract the key operating parameters in the current period, such as spindle temperature, current fluctuation, joint torque, etc.; combine the maintenance label or health score of the system during this period to construct an operating parameter - equipment status comparison relationship, and mark it as a stage state sample; each state sample records the corresponding timestamp, running duration and status annotation to form a traceable perception segment;

[0103] In each perception segment, collect the background information of the working conditions at that time, including various context information such as environmental temperature, equipment load, task type, etc.; use edge sensors and task scheduling systems to synchronously record in real time the working condition information such as external environmental factors, current job task types, and job beats; bind the collected working condition information to each perception segment one by one to construct a working condition perception sample, ensuring that each set of state data has complete context information; the collected working condition variables need to be standardized for subsequent model analysis and pattern recognition;

[0104] Construct a working condition perception factor matrix, and use principal component analysis to identify typical working condition patterns; extract eigenvectors and perform dimensionality reduction processing on all working condition perception samples, apply principal component analysis to identify the main influencing factors, and compress redundant information; generate a relationship matrix between working condition categories and key perception variables to identify the operating characteristic features under each type of typical working condition;

[0105] Combine the corresponding relationship between each working condition pattern and the state label, and dynamically adjust the original state-maintenance behavior mapping strategy; analyze the historical response data between the equipment state and the maintenance behavior under each working condition category, and construct a three-dimensional mapping relationship of working condition pattern-state category-maintenance strategy; compare the currently identified equipment state label with the maintenance response plan under the same historical working conditions to identify whether there are deviations or lags; if the identified strategy is inconsistent or the effect is not good, the system will automatically trigger a mapping correction mechanism to replace the strategy or fine-tune the parameters of the maintenance behavior recommendation in the current state;

[0106] Through the mapping correction mechanism, ensure that the maintenance behavior highly matches the current working condition, and improve the accuracy and timeliness of the maintenance response.

[0107] S4: According to the corrected state mapping relationship, apply a multi-scale fault prediction mechanism to divide the response levels of potential fault trends, extract the matching maintenance plan parameters, and dynamically adjust the response strategy to form a maintenance scheduling plan at the edge.

[0108] The process of applying the multi-scale fault prediction mechanism to divide the response levels of potential fault trends in S4 is as follows:

[0109] Mine the fault trends from the corrected state mapping results, and use the multi-scale sliding window mechanism for time-domain resampling; based on the state-maintenance mapping results corrected by the working conditions, extract the continuous sequence of key features changing over time; introduce the multi-scale sliding window mechanism to resample the original time series data to generate trend observation windows of different granularities; each time scale corresponds to an independent time perspective to capture short-term fluctuations, periodic changes, or long-term decline trends;

[0110] At each time scale, an independent state evolution prediction sub-model is trained, and an attention fusion mechanism is used to integrate the prediction results of each scale; for the feature sub-sequences at each scale, a lightweight time-series neural network model is constructed and trained respectively; each sub-model is responsible for predicting the evolution trend of the future state within its specific time range and outputting a sequence of predicted feature vectors; an attention fusion mechanism is introduced to weight and integrate the prediction results of sub-models at different scales according to the attention degree of the key features of the current prediction task, so as to obtain a comprehensive fault trend prediction output;

[0111] Analyze the change rate and amplitude of the predicted potential fault trend, and divide the response level of the risk degree according to the set threshold; perform numerical differentiation and statistical analysis on the fused predicted trend curve to extract key change indicators, including the change rate and the maximum fluctuation amplitude; set multi-level response risk thresholds and compare the trend indicators with the thresholds; if the trend curve meets the change rate and fluctuation amplitude conditions of a certain level, the potential fault risk level within that time period is defined to prepare for subsequent maintenance responses.

[0112] Extract the maintenance parameter templates corresponding to different response levels, including the maintenance scope, required resources and response time;

[0113] Form a set of maintenance plan parameters with high matching degree as the key input content for subsequent maintenance scheduling.

[0114] S5: Based on the maintenance scheduling plan, combined with the preset fault handling knowledge base, automatically evaluate and optimize the current maintenance strategy, and generate equipment maintenance suggestions in real time.

[0115] The process of generating equipment maintenance suggestions in real time in the above S5 includes:

[0116] Based on the above-mentioned set of maintenance plan parameters, match the maintenance rules and case base in the fault handling knowledge base;

[0117] Call the expert system module to compare the current system state with historical cases and identify the optimal maintenance path;

[0118] Combined with the current operation plan of the robot, available time window and maintenance resource allocation situation, automatically generate a maintenance scheduling plan;

[0119] Based on the generated scheduling plan, analyze the impact of the scheduling task on the overall production rhythm and put forward optimization suggestions;

[0120] Finally, automatically output structured equipment maintenance suggestions.

[0121] Embodiment 3: Please refer to Figure 2 As shown in the figure, an industrial robot real-time maintenance system combined with edge computing described in an embodiment of the present invention includes:

[0122] Feature construction module: Collect and fuse multi-source status data, and construct a structured operation feature dataset for key components of the robot through status encoding and feature coupling analysis;

[0123] Incremental learning module: Establish a device health model based on distributed processing and lightweight time series modeling algorithms, and continuously optimize the model performance and anomaly monitoring capabilities through incremental learning;

[0124] Status recognition module: Analyze the change trend of the abnormal feature trigger frequency, judge the stability of status recognition, and dynamically correct the maintenance behavior mapping relationship in combination with the working condition perception mechanism;

[0125] Fault prediction module: Perform multi-scale fault trend prediction according to the corrected status information and automatically divide the response levels;

[0126] Recommendation generation module: Rely on the maintenance knowledge base to automatically evaluate and adjust the current strategy, and generate intelligent maintenance recommendations that match the actual working conditions in real time.

[0127] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0128] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time maintenance method for industrial robots combined with edge computing, characterized in that, The following steps are involved: Acquire multi-source operating status data of industrial robots and, combined with the edge's state encoding mechanism and feature coupling analysis method, construct an operating feature dataset of key robot components. Rapid distributed processing of operational feature datasets is performed, and a device health status model is constructed based on a lightweight time series modeling algorithm. A multi-dimensional correlation analysis mechanism is introduced to perform incremental learning on the device health status model. Changes in the trigger frequency of abnormal features are monitored in real time during the model update process. Based on the changing trend of the trigger frequency, the system determines whether the edge-side state recognition result is stable. If it is stable, the system records the key dimension state changes under the current working condition of the robot and, combined with the working condition perception mechanism, makes real-time corrections to the maintenance state mapping relationship. Based on the revised state mapping relationship, a multi-scale fault prediction mechanism is applied to classify potential fault trends into response levels, extract matching maintenance plan parameters, and dynamically adjust the response strategy to form a maintenance scheduling plan at the edge. Based on the maintenance scheduling plan and combined with the preset fault handling knowledge base, the current maintenance strategy is automatically evaluated and optimized, and equipment maintenance recommendations are generated in real time.

2. The real-time maintenance method of an industrial robot combined with edge computing according to claim 1, wherein The process of constructing the operational feature dataset of the robot's key components is as follows: Perform data preprocessing on the acquired multi-source operating status data; Normalization is used for continuous operation indicators; The discrete state features are converted using mechanisms such as one-hot encoding or label encoding to form a structured state feature set; Using state encoding deployed at the edge, we can embed the pre-processed feature set into a structured representation, improving its adaptability to edge computing resources. Based on the characteristic coupling analysis method, multiple coupling relationships between operating indicators are identified, and the dynamic correlation between key variables is presented in the form of a coupling strength matrix; Combined with the coupling matrix results, high coupling feature combinations are screened out, and finally an operation feature dataset reflecting the operating health status of the robot's key components is constructed.

3. The real-time maintenance method of an industrial robot combined with edge computing according to claim 2, characterized in that, The fast distributed processing and lightweight time series modeling algorithm process is: Divide the constructed operational feature dataset into multiple edge subtasks and adopt a multi-threaded concurrent processing mechanism; Introducing a lightweight edge computing framework deployed on terminal nodes; For the processed operating data sequence, a local modeling method based on sliding windows is used to construct a short-time series health trend model; Adopting gated recurrent units to model feature fluctuations in state sequences and capture potential abnormal change patterns; The outputs of multiple source sub-models are aggregated and a unified device health status model is constructed through the feature aggregation module on the edge side.

4. The real-time maintenance method of an industrial robot combined with edge computing according to claim 3, characterized in that The process of incremental learning of the equipment health status model by introducing a multi-dimensional correlation analysis mechanism is as follows: After the initial training is completed, each newly acquired running data segment is updated online to achieve adaptive evolution of the model; Combining the historical synergistic change relationships between dimensions, we construct a multi-dimensional interaction map and explore the potential causal relationships between key components. The attention mechanism is introduced to assign higher weights to feature dimensions with higher correlation strength, thus achieving focused modeling of device status features. During the incremental learning process, a dynamic batch update strategy is used to control the weight update amplitude of each model learning.

5. A real-time maintenance method for an industrial robot combined with edge computing according to claim 4, characterized in that The process of real-time monitoring the change in the triggering frequency of abnormal features is as follows: During the incremental learning training cycle, the feature dimensions identified as abnormal states in the model input features are extracted in real time; Based on the time window mechanism, statistical analysis is performed on the triggering frequency of abnormal features within the window period; Introduce a dynamic change threshold strategy, and identify whether there is a frequency shift in the current cycle according to the abnormal frequency baseline under the historical stable operation cycle of the system; When there is a frequency shift, automatically record the corresponding features and timestamps, and include them as potential failure signs in the scope of consideration for the maintenance plan; Combined with the distribution range and aggregation trend of abnormal features, assist in evaluating the fluctuation degree of the current operating state of the robot.

6. The real-time maintenance method of an industrial robot combined with edge computing according to claim 5, characterized in that, The process of judging whether the edge-side state recognition result is stable is as follows: Smooth the abnormal feature frequency change curve and extract the change rate index; If the absolute value of the abnormal frequency change rate is less than the preset threshold, it is determined that the edge-side state recognition result is stable; If the absolute value of the abnormal frequency change rate is greater than or equal to the preset threshold, it is determined that the edge-side state recognition result is unstable.

7. The real-time maintenance method of an industrial robot combined with edge computing according to claim 6, characterized in that The process of real-time correction of the maintenance state mapping relationship in combination with the working condition perception mechanism is as follows: Record the key operating parameters and corresponding equipment state labels in the stable state to form a phased state perception segment; In each perception segment, collect the working condition background information at that time; Construct a working condition perception factor matrix, and use principal component analysis or clustering algorithms to identify typical working condition patterns; Combined with the corresponding relationship between each working condition pattern and the state label, dynamically adjust the original state-maintenance behavior mapping strategy.

8. The real-time maintenance method of an industrial robot combined with edge computing according to claim 7, wherein, The process of applying the multi-scale fault prediction mechanism to divide the response levels of potential fault trends is as follows: Mine the fault trend from the corrected state mapping results, and use the multi-scale sliding window mechanism for time-domain resampling; At each time scale, train an independent state evolution prediction sub-model, and use the attention fusion mechanism to integrate the prediction results of each scale; Analyze the change rate and amplitude of the predicted potential fault trend, and divide the risk level according to the set threshold; Extract the maintenance parameter templates corresponding to different response levels, including the maintenance scope, required resources, and response time.

9. The real-time maintenance method of an industrial robot combined with edge computing according to claim 8, characterized in that, The process of real-time generating equipment maintenance suggestions includes: Based on the maintenance plan parameter set, match the maintenance rules and case libraries in the fault handling knowledge base; Call the expert system module to compare the current system state with historical cases and identify the optimal maintenance path; Combined with the current operation plan of the robot, available time window, and maintenance resource configuration, automatically generate a maintenance scheduling plan; Based on the generated scheduling plan, analyze the impact of the scheduling task on the overall production rhythm and propose optimization suggestions; Finally, automatically output structured equipment maintenance suggestions.

10. An industrial robot real-time maintenance system combined with edge computing, which is applied to the method described in any one of claims 1-9, and is characterized in that, Including: Feature construction module: Collect and fuse multi-source state data, and construct a structured operation feature data set of key components of the robot through state coding and feature coupling analysis; Incremental learning module: Establish an equipment health model based on distributed processing and lightweight time series modeling algorithms, and continuously optimize the model performance and abnormal monitoring ability through incremental learning; Status recognition module: Analyze the changing trend of the triggering frequency of abnormal features, judge the stability of status recognition, and dynamically correct the mapping relationship of maintenance behaviors in combination with the working condition perception mechanism; Fault prediction module: Perform multi-scale fault trend prediction according to the corrected status information and automatically divide the response levels; Recommendation generation module: Rely on the maintenance knowledge base to automatically evaluate and adjust the current strategy, and generate intelligent maintenance recommendations that match the actual working conditions in real time.

Images