A robot arm automation operation and maintenance management system integrated with state monitoring

Abstract

The application discloses a kind of integrated state monitoring's robot mechanical arm automation operation management system, belong to industrial robot operation management technical field.It includes six layers of data acquisition, edge computing, state evaluation, fault prediction, operation decision and visual interaction.Data acquisition layer is deployed multiple sensors to collect vibration, current, temperature, position and other multi-source signals;Edge computing layer completes signal pre-processing and time-frequency domain feature extraction nearby;State evaluation layer is based on multidimensional features using principal component analysis and Mahalanobis distance to calculate joint health degree score;Fault prediction layer uses three-layer LSTM network to predict remaining life based on health degree time series;Operation decision layer automatically generates hierarchical maintenance strategy and dispatches work order according to the prediction result;Visualization layer displays health status through three-dimensional digital twin model in heat map.The application integrates multi-source data, uses edge computing to reduce latency, forms an automated closed loop from monitoring to work order, and improves operation real-time and intelligent level.

Description

Technical Field

[0001] This invention relates to the field of industrial robot operation and maintenance management technology, specifically to an automated operation and maintenance management system for robotic arms with integrated status monitoring. Background Technology

[0002] Industrial robotic arms are widely used in automated production lines in automobile manufacturing, electronic assembly, and metal processing. Their operational reliability and stability directly affect production efficiency and product quality. As robots age, critical components such as reducers, servo motors, and bearings in the robotic arms inevitably experience wear and performance degradation. Traditional operation and maintenance management methods mainly rely on periodic inspections and reactive repairs, which suffer from problems such as untimely fault detection, lack of data support for maintenance plans, and unreasonable spare parts replacement cycles, leading to frequent unplanned downtime and significant production losses.

[0003] In recent years, condition monitoring and predictive maintenance technologies have developed rapidly in the field of industrial equipment. Existing technologies include methods that collect operational data from industrial robots and combine it with machine learning models for condition monitoring and fault diagnosis. For example, Chinese patent CN116572256B discloses a method for industrial robot condition monitoring and fault diagnosis, which achieves online monitoring by dividing operational data into stable and non-stable segments and converting them into monitoring status codes. Another approach uses digital twin technology to construct a real-time dynamic virtual model of the industrial robot for condition monitoring. Still other approaches establish robot joint torque prediction models and mechanical parameter identification models to determine the robot's operational and health status in real time.

[0004] However, the aforementioned existing technologies still have significant shortcomings. First, most solutions only analyze data from a single type of sensor, lacking the comprehensive integration and utilization of multi-source heterogeneous data such as vibration, current, temperature, and position, resulting in insufficient comprehensiveness and accuracy in condition assessment. Second, existing solutions generally adopt a centralized architecture of "data acquisition and cloud processing," requiring a large amount of raw data to be transmitted from the robot to a remote server for processing, leading to significant data transmission latency and bandwidth consumption issues. Third, most systems only provide condition monitoring and alarm functions, lacking complete closed-loop operation and maintenance management capabilities from data acquisition, condition assessment, fault prediction to maintenance work order dispatch. Fourth, the visualization interaction methods of existing systems are relatively simple, making it difficult to intuitively present the real-time health status distribution of each joint of the robotic arm.

[0005] Therefore, there is an urgent need in this field for an automated operation and maintenance management system for robotic arms that can integrate multi-source sensor data, has real-time edge processing capabilities, and covers the entire chain of "monitoring, evaluation, prediction, and decision-making". Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an automated operation and maintenance management system for robotic arms with integrated status monitoring. This system solves the problems of insufficient fusion of multi-source sensor data, high data transmission latency, lack of complete closed-loop operation and maintenance management capabilities, and unintuitive visualization interaction in existing robotic arm operation and maintenance management systems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An automated operation and maintenance management system for robotic arms with integrated condition monitoring includes: The data acquisition layer, deployed at each joint and end effector of the robot arm, includes vibration sensors, current sensors, temperature sensors and position sensors, used to collect multi-source status signals during the operation of the robot arm in real time; The edge computing layer is communicatively connected to the data acquisition layer and has a built-in signal preprocessing unit and feature extraction unit, which are used to perform noise reduction, filtering and time-frequency domain feature extraction on the acquired multi-source state signals. The status assessment layer, which is communicatively connected to the edge computing layer, includes a health assessment unit and an anomaly detection unit, used to calculate the health indicators of each joint of the robotic arm based on the extracted feature data and detect abnormal states. The fault prediction layer is communicatively connected to the state assessment layer and has a built-in remaining life prediction model based on a long short-term memory network (LSTM) to predict the remaining life of key components of the robotic arm based on the time series of health indicators. The operation and maintenance decision layer is connected to the fault prediction layer and includes a maintenance strategy generation unit and a work order management unit, which are used to generate maintenance suggestions and automatically dispatch maintenance work orders based on the remaining service life prediction results. The visualization and interaction layer communicates with the above layers and is used to display the robotic arm's operating status, health score, and prediction results in real time in the form of a 3D model.

[0008] Preferably, in the data acquisition layer, vibration sensors are installed on the reducer housing of each joint of the robotic arm, with a sampling frequency of 10-20kHz; current sensors are connected in series in the power supply line of the servo motor of each joint, with a sampling frequency of 1-5kHz; temperature sensors are attached to the inner wall of the joint housing, with a sampling frequency of 0.5-1Hz; and position sensors are encoders built into each joint, used to collect joint angles and angular velocities.

[0009] Preferably, the signal preprocessing unit of the edge computing layer includes: a noise suppression module based on wavelet transform for denoising the vibration signal; a power frequency interference elimination module based on adaptive filtering for eliminating 50Hz power frequency interference in the current signal; and a data alignment module based on a sliding window for aligning multi-source signals with different sampling frequencies on the time axis.

[0010] Preferably, the time-domain features extracted by the feature extraction unit of the edge computing layer include peak value, root mean square value, kurtosis and waveform factor; the frequency-domain features include fundamental frequency amplitude, second harmonic amplitude and harmonic energy ratio extracted by fast Fourier transform.

[0011] Preferably, the health assessment unit of the state assessment layer calculates the comprehensive health score of each joint of the robotic arm based on the multidimensional feature data of each joint, after dimensionality reduction by principal component analysis algorithm, and the score range is 0 to 100 points.

[0012] Preferably, the anomaly detection unit of the state evaluation layer adopts an anomaly detection model based on the isolated forest algorithm, which maps feature data to an anomaly score space, and triggers an alarm when the anomaly score exceeds a preset threshold.

[0013] Preferably, the remaining lifetime prediction model of the fault prediction layer is a three-layer LSTM network structure. The input layer receives time series data of health indicators, the hidden layer has 128 nodes, and the output layer outputs the remaining lifetime prediction value. The model is trained using historical health degradation trajectory data as the training set and mean squared error as the loss function.

[0014] Preferably, the maintenance strategy generation unit of the operation and maintenance decision layer includes: determining the maintenance priority based on the predicted remaining service life and in combination with the preset service life threshold range; the maintenance priority is divided into three levels: emergency maintenance, planned maintenance, and status observation.

[0015] Preferably, the visualization interaction layer includes: a three-dimensional digital twin model of the robotic arm built based on Unity3D or Three.js, which is synchronized with the joint angles of the physical robotic arm in real time; and a health score superimposed on each joint of the three-dimensional model in the form of a heat map.

[0016] Preferably, the system further includes a data storage layer, which is communicatively connected to the above layers, for storing raw status data, feature data, health score history and maintenance records, and supports retrieval by time range and device number.

[0017] This invention provides an automated operation and maintenance management system for robotic arms with integrated status monitoring. It offers the following advantages: 1. Vibration signals reflect the wear and impact conditions of mechanical transmission components; current signals reflect the load and drive conditions of the servo motor; temperature signals reflect thermal balance and lubrication conditions; and position signals reflect motion accuracy and trajectory deviation. These four types of signals characterize the health status of the robotic arm from different dimensions. Compared with existing solutions that rely on a single type of sensor, the condition assessment results of this invention are more comprehensive and accurate.

[0018] 2. This invention employs an edge computing architecture, moving signal preprocessing and feature extraction from the cloud to edge nodes closer to the data source. High-sampling-rate data such as vibration and current undergo noise reduction, filtering, and feature extraction at the edge, with only low-dimensional feature data transmitted to the cloud for status assessment and fault prediction. This architecture effectively reduces data transmission bandwidth consumption and cloud computing pressure, while also minimizing data transmission latency, meeting the real-time requirements of industrial environments.

[0019] 3. This invention constructs a complete closed-loop management chain from data acquisition, edge processing, status assessment, fault prediction to operation and maintenance decision-making. The status assessment layer outputs health scores and anomaly alarms for each component, the fault prediction layer predicts the remaining lifespan based on the health degradation trajectory, and the operation and maintenance decision-making layer automatically generates tiered maintenance strategies and dispatches work orders based on the prediction results. The layers communicate through standardized interfaces, forming an automated operation and maintenance closed loop of "monitoring, assessment, prediction, decision-making, and execution," completing the entire process from anomaly detection to maintenance task dispatch without manual intervention.

[0020] 4. This invention uses a three-dimensional digital twin model to visualize the operating status and health of a robotic arm. The joint angles of the digital twin model are synchronized with the physical robotic arm in real time, and the health scores of each joint are superimposed on the corresponding parts of the three-dimensional model in the form of a heatmap. Maintenance personnel can intuitively grasp the health distribution of each joint of the robotic arm through a visual interface, quickly locate abnormal joints, and reduce the technical threshold and training costs for maintenance personnel. Attached Figure Description

[0021] Figure 1 A layered architecture diagram of the system; Figure 2 This is the main flowchart for data processing; Figure 3 This is a flowchart of the operation and maintenance closed loop. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figure 1-3 This invention provides an integrated status monitoring automated operation and maintenance management system for robotic arms, comprising the following components: Data Acquisition Layer The data acquisition layer is deployed at each joint of the robot arm and at the end effector, including vibration sensors, current sensors, temperature sensors and position sensors.

[0024] The vibration sensor is a piezoelectric accelerometer, mounted on the flat surface of the reducer housing at each joint of the robotic arm, and fixed by a magnetic base or adhesive. The sampling frequency of the vibration sensor is set to 10–20 kHz, which can effectively cover the gear meshing frequency and its harmonics of the robotic arm joint reducer, and is used to capture vibration characteristic changes caused by gear wear, bearing failure, etc.

[0025] The current sensor is a closed-loop Hall effect current sensor, connected in series in the U-phase power supply line of each joint servo motor, with a sampling frequency set to 1–5 kHz. The current signal can reflect the load torque change of the servo motor. When the efficiency of the joint reducer decreases or jamming occurs, the motor drive current will show characteristic fluctuations.

[0026] The temperature sensor used is a PT100 platinum resistance temperature sensor, which is mounted on the inner wall of each joint housing near the reducer bearing. The sampling frequency is set to 0.5–1 Hz. The temperature signal is used to monitor the thermal balance of the joint; abnormal temperature rise often indicates lubrication failure or increased friction.

[0027] The position sensors are incremental or absolute encoders built into each joint, which are read through the communication interface of the servo driver to collect joint angles and angular velocities. The position and angular velocity data are used to determine the deviation of the robotic arm's motion trajectory and speed fluctuations.

[0028] The four types of sensors are connected to the data acquisition card of the edge computing node via shielded cables. The data acquisition card has a multi-channel synchronous sampling function to ensure the time consistency of data from each sensor.

[0029] Edge computing layer The edge computing layer is deployed on computing nodes (such as industrial control computers or embedded edge computing devices) in industrial sites, and communicates with the data acquisition layer. It has built-in signal preprocessing units and feature extraction units.

[0030] The signal preprocessing unit consists of three sub-modules: The noise suppression module based on wavelet transform uses the Daubechies wavelet basis (db4) to perform four-level wavelet decomposition on the vibration signal, and then reconstructs the signal after removing high-frequency noise components using a soft thresholding method. Wavelet transform can effectively preserve the transient impact components in the vibration signal while suppressing background noise.

[0031] The power frequency interference cancellation module based on adaptive filtering uses an adaptive notch filter to eliminate 50Hz power frequency interference and its harmonic components in the current signal. The center frequency and bandwidth of the notch filter are adaptively adjusted in real time according to the signal spectrum.

[0032] The sliding window-based data alignment module aligns multi-source signals with different sampling frequencies (vibration 10–20 kHz, current 1–5 kHz, temperature 0.5–1 Hz) on the time axis. The alignment method is as follows: using the sampling time of the vibration signal as a reference, the current signal and temperature signal are resampled to the same timestamp as the vibration signal through linear interpolation.

[0033] The feature extraction unit extracts time-domain and frequency-domain features from the preprocessed signal. Time-domain features include: peak value (the difference between the maximum and minimum signal values), root mean square (RMS) value (reflecting the signal energy level), kurtosis (reflecting the signal's impulse characteristics), and waveform factor (the ratio of the RMS value to the rectified average value). Frequency-domain features are extracted after converting the time-domain signal to the frequency domain using a Fast Fourier Transform (FFT), including the fundamental frequency amplitude (the spectral amplitude corresponding to the servo motor's rotation frequency), the second harmonic amplitude (the spectral amplitude corresponding to twice the rotation frequency), and the harmonic energy ratio (the ratio of the sum of the squares of the first five harmonic amplitudes to the total spectral energy). For each type of sensor signal at each joint, the above features are extracted to form a multi-dimensional feature vector.

[0034] State evaluation layer The status assessment layer is deployed on a cloud server or a local server and communicates with the edge computing layer via industrial Ethernet or MQTT protocol. It includes a health assessment unit and an anomaly detection unit.

[0035] The health assessment unit uses Principal Component Analysis (PCA) to reduce dimensionality based on the multidimensional feature data of each joint. PCA projects the original high-dimensional feature space onto the principal component space, selecting the top principal components with a cumulative variance contribution rate of 85% or higher as the dimensionality-reduced features. Then, using the feature data of the robotic arm in a healthy state (newly manufactured or recently overhauled) as a benchmark, the Mahalanobis distance between the current feature vector and the benchmark feature vector is calculated. Mahalanobis distance considers the correlation between different feature dimensions and can more accurately measure the deviation between the current state and the health benchmark. The Mahalanobis distance is mapped to a health score range of 0-100, with higher scores indicating better health. The mapping formula is: Health Score = 100 × exp(λ × Mahalanobis Distance), where λ is a scaling factor determined based on historical data.

[0036] The anomaly detection unit employs an anomaly detection model based on the Isolation Forest algorithm. The Isolation Forest constructs multiple isolation trees by randomly selecting features and randomly selecting split values, mapping the feature data to an anomaly score space. An anomaly score closer to 1 indicates that the sample is more likely to be isolated, i.e., the higher the degree of anomaly. When the anomaly score exceeds a preset threshold (e.g., 0.65), the system triggers an alert and marks that the joint has an abnormal state. The alert information includes the abnormal joint number, anomaly occurrence time, and abnormal feature description, and is pushed to the visual interaction layer and the operation and maintenance decision-making layer through the message queue.

[0037] Fault prediction layer The fault prediction layer is communicatively connected with the state assessment layer, and has a built-in remaining useful life prediction model based on the Long Short-Term Memory (LSTM) network.

[0038] The LSTM model has a three-layer network structure: the input layer receives time series data of health indicators, with a time window length of 30 consecutive sampling points (each sampling point corresponds to one operation cycle, and the cycle length is determined according to the working rhythm of the manipulator); the first hidden layer has 128 nodes and adopts LSTM units for capturing the long-term dependency relationship of health degradation; the second hidden layer has 64 nodes; the output layer is a single node, which outputs the predicted remaining useful life (in units of operating hours or working cycles).

[0039] The model training process is as follows: collect complete health degradation trajectory data of multiple manipulators of the same model from the healthy state to fault shutdown as the training set, take the health time series as input features and the corresponding actual remaining useful life as labels, use Mean Squared Error (MSE) as the loss function, use the Adam optimizer for training, set the learning rate to 0.001, the batch size to 32, and the number of training epochs to 200. After the training is completed, the model is deployed to the cloud inference service, receives the real-time health sequence from the state assessment layer, and outputs the remaining useful life prediction result.

[0040] Operation and maintenance decision-making layer The operation and maintenance decision-making layer is communicatively connected with the fault prediction layer, and includes a maintenance strategy generation unit and a work order management unit.

[0041] The maintenance strategy generation unit determines the maintenance priority based on the predicted Remaining Useful Life (RUL) in combination with the preset life threshold intervals. The thresholds are set as follows: when RUL ≤ 50 hours, the priority is determined as "emergency maintenance", and the system automatically triggers a shutdown suggestion and notifies on-site maintenance personnel to handle it immediately; when 50 hours < RUL ≤ 200 hours, the priority is determined as "planned maintenance", and the system generates maintenance suggestions and incorporates them into the recent maintenance plan; when RUL > 200 hours, the priority is determined as "status observation", and the system only records the status information without triggering maintenance actions.

[0042] The work order management unit automatically generates electronic maintenance work orders when "emergency repair" and "planned repair" priorities are triggered. The work order includes: robotic arm number, abnormal joint number, health score, predicted remaining lifespan, suggested repair measures (such as reducer replacement, lubrication maintenance, parameter calibration, etc.), and suggested repair window time. Work orders are automatically dispatched to the corresponding maintenance personnel's mobile terminals via the enterprise operations and maintenance system interface, and their execution status is tracked until the process is closed.

[0043] Visual interaction layer The visualization and interaction layer communicates with the above layers, including a 3D digital twin model of the robotic arm built based on Unity3D or Three.js.

[0044] The construction process of the digital twin model is as follows: First, the three-dimensional geometric models of each joint are imported according to the CAD model of the robotic arm to establish the kinematic chain structure between the joints; then, real-time joint angle data from the edge computing layer is received through the WebSocket protocol to drive the rotation of each joint of the three-dimensional model, so as to realize the synchronous movement of the digital model and the physical robotic arm.

[0045] Health scores are overlaid as heatmaps on various joints of the 3D model: scores of 90-100 are displayed in green, 70-90 in yellow, 50-70 in orange, and below 50 in red. Anomaly alerts are displayed as pop-ups next to the model; clicking the pop-up displays detailed anomaly data and prediction results. The visualization interface also provides auxiliary information panels such as historical health trend charts, remaining life prediction curves, and maintenance work order lists.

[0046] Data storage layer The system also includes a data storage layer that communicates with the layers mentioned above. It uses a time-series database (such as InfluxDB) to store raw status data, feature data, and historical health scores, and a relational database (such as MySQL) to store equipment ledger information, maintenance records, and work order information. The data storage layer supports retrieval based on multiple dimensions such as time range, equipment number, and joint number, providing a data foundation for subsequent data analysis and model iteration.

[0047] Example 1: System Deployment and Operation This embodiment uses a six-axis industrial robot arm in the welding workshop of an automobile manufacturing plant as an example to deploy the system described in this invention.

[0048] Data acquisition layer: A piezoelectric accelerometer (sampling frequency 12.8kHz) is installed on the reducer housing of each of the six joints J1 to J6 of the robotic arm. A Hall current sensor (sampling frequency 2.56kHz) is connected in series in the U-phase circuit of the servo motor of each joint. A PT100 temperature sensor (sampling frequency 1Hz) is attached to the inner wall of the housing of each joint. The encoder position data is read through the servo driver communication interface.

[0049] Edge Computing Layer: An industrial-grade edge computing gateway (configured with an Intel Core i7 processor, 16GB of RAM, and a 512GB SSD) is deployed on-site in the workshop, running Ubuntu 20.04. A data acquisition card connects to the edge computing gateway via a PCIe interface, acquiring and buffering data from each channel at a 1-second interval. The signal preprocessing and feature extraction program is written in C++, completing a full data processing and feature extraction every 10 seconds, generating a multi-dimensional feature vector for each joint (8 features extracted per signal type per joint, totaling 6×4×8=192 features).

[0050] The status assessment layer and fault prediction layer are deployed on a local server at the factory (configured with an NVIDIA Tesla T4 GPU). The edge computing gateway uploads feature vectors to the local server every 10 seconds via the MQTT protocol. After receiving the feature data, the status assessment layer calculates the health score and anomaly score for each joint. In testing, after 30 days of continuous operation, the health score of joint J3 gradually decreased from an initial 98 points to 72 points, while the anomaly score increased from 0.12 to 0.58 (threshold 0.65), without triggering any alarms.

[0051] The fault prediction layer predicts the remaining service life (RUL) of the J3 joint based on its health time series (30 consecutive sampling points). The model outputs a RUL prediction value of 180 hours, falling within the "planned maintenance" range (50-200 hours). The maintenance decision layer automatically generates a maintenance work order, recommending that the J3 joint reducer lubrication maintenance and bearing inspection be scheduled within 180 hours. The work order is pushed to the maintenance team leader's mobile terminal via the enterprise MES system interface.

[0052] The visualization and interaction layer displays the 3D model of the robotic arm in real time on a large screen in the workshop control room. The color of the J3 joint gradually changes from green to yellow, while the other joints remain green, allowing maintenance personnel to intuitively locate the joints that need attention.

[0053] Example 2: Verification of Anomaly Alarms This embodiment simulates a mechanical arm joint bearing failure in a laboratory environment to verify the system's anomaly detection capability.

[0054] The input shaft bearing of the J2 joint reducer of a six-axis robotic arm was replaced with a bearing pre-installed with an outer ring pitting fault (fault dimensions: diameter 2mm, depth 0.5mm). The robotic arm was then operated continuously under standard conditions (load rate 60%, operating speed 50%).

[0055] The data acquisition layer and edge computing layer operated according to the configuration of Example 1. During the second hour of operation, the kurtosis value of joint J2, extracted by the edge computing layer, increased from 3.2 during normal operation to 6.8, and the second harmonic amplitude increased from 0.15g to 0.42g. The anomaly detection unit of the status assessment layer calculated an anomaly score of 0.71 for joint J2, exceeding the preset threshold of 0.65, triggering an alarm. The health assessment unit calculated a health score of 68 for joint J2, a decrease of 27 points from the normal operating score of 95.

[0056] The alarm information is pushed to the visualization interaction layer via the message queue. The J2 joint is displayed in orange in the 3D model, and an alarm pop-up window appears with the content: "J2 joint abnormal alarm: abnormal score 0.71, health score 68, it is recommended to check the reducer bearing." At the same time, it is pushed to the operation and maintenance decision layer to generate an emergency repair work order.

[0057] Following the work order's recommendations, maintenance personnel disassembled and inspected joint J2, confirming a pitting fault in the reducer's input shaft bearing. After replacing the bearing, the system returned to normal, and the health score rebounded to 93 points. From the occurrence of the anomaly to fault confirmation, the system achieved fully automated monitoring, alarming, and work order dispatch, requiring no manual intervention.

[0058] Comparative experiment To verify the technical effects of the present invention, the following comparative examples are provided: Comparative Example 1: The difference from Example 1 is that only vibration signals are collected, and current, temperature, and position signals are not collected. The condition assessment is based solely on vibration characteristics.

[0059] Comparative Example 2: The difference from Example 1 is that an edge computing architecture is not used. All raw data (vibration 12.8kHz × 6 channels) are transmitted to a local server for processing via industrial Ethernet.

[0060] Comparative Example 3: The difference from Example 1 is that it does not include a fault prediction layer and an operation and maintenance decision layer, but only provides status monitoring and alarm functions and does not generate maintenance work orders.

[0061] After operating under the same conditions for 30 days, the comparison results are as follows:

[0062] Comparative Example 1 shows that the anomaly detection accuracy relying solely on vibration signals is significantly lower than the multi-source signal fusion scheme of this invention. In the early stage of bearing failure, the vibration signal changes are not obvious while the current signal has already shown characteristic fluctuations, and a single signal source is prone to causing missed detections.

[0063] Comparative Example 2 shows that a centralized architecture that uploads all raw data to the cloud leads to a sharp increase in data transmission bandwidth requirements (approximately 21 times) and a significant increase in end-to-end latency (approximately 3.5 times), making it difficult to meet real-time requirements when network conditions in industrial settings are limited.

[0064] Comparative Example 3 shows that systems lacking fault prediction and automatic work order dispatch functions still require manual judgment and work order dispatch after an anomaly is detected. The average time from fault detection to maintenance task issuance is 45 minutes, while the present invention achieves a second-level automated closed-loop response.

[0065] As can be seen from the above comparative experiments, the multi-source sensor fusion, edge computing architecture and full-chain operation and maintenance closed-loop management of the present invention work together to form a complete technical solution.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automated operation and maintenance management system for robotic arms with integrated status monitoring, characterized in that, It includes a data acquisition layer, an edge computing layer, a status assessment layer, a fault prediction layer, an operation and maintenance decision-making layer, and a visualization and interaction layer. The data acquisition layer, deployed at each joint and end effector of the robot arm, includes vibration sensors, current sensors, temperature sensors and position sensors, used to collect multi-source status signals during the operation of the robot arm in real time; The edge computing layer is communicatively connected to the data acquisition layer and has a built-in signal preprocessing unit and feature extraction unit, which are used to perform noise reduction, filtering and time-frequency domain feature extraction on the acquired multi-source state signals. The status assessment layer, which is communicatively connected to the edge computing layer, includes a health assessment unit and an anomaly detection unit, used to calculate the health indicators of each joint of the robotic arm based on the extracted feature data and detect abnormal states. The fault prediction layer is communicatively connected to the state assessment layer and has a built-in remaining life prediction model based on a long short-term memory network. It is used to predict the remaining life of key components of the robotic arm based on the time series of health indicators. The operation and maintenance decision layer is connected to the fault prediction layer and includes a maintenance strategy generation unit and a work order management unit, which are used to generate maintenance suggestions and automatically dispatch maintenance work orders based on the remaining life prediction results. The visualization and interaction layer communicates with the above layers and is used to display the robotic arm's operating status, health score, and prediction results in real time in the form of a 3D model.

2. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, In the data acquisition layer, vibration sensors are installed on the reducer housings of each joint of the robotic arm, with a sampling frequency of 10-20kHz; current sensors are connected in series in the power supply lines of the servo motors of each joint, with a sampling frequency of 1-5kHz; temperature sensors are attached to the inner wall of the joint housing, with a sampling frequency of 0.5-1Hz; and position sensors are encoders built into each joint, used to collect joint angles and angular velocities.

3. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The signal preprocessing unit of the edge computing layer includes: a noise suppression module based on wavelet transform for denoising vibration signals; a power frequency interference elimination module based on adaptive filtering for eliminating 50Hz power frequency interference in current signals; and a data alignment module based on sliding window for aligning multi-source signals with different sampling frequencies on the time axis.

4. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The temporal features extracted by the feature extraction unit of the edge computing layer include peak value, root mean square value, kurtosis and waveform factor; Frequency domain characteristics include the fundamental frequency amplitude, second harmonic amplitude, and harmonic energy ratio extracted by fast Fourier transform.

5. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The health assessment unit of the state assessment layer calculates the comprehensive health score of each joint of the robotic arm based on the multidimensional feature data of each joint, after dimensionality reduction by principal component analysis algorithm, and then calculates the comprehensive health score of each joint of the robotic arm by Mahalanobis distance. The score range is 0 to 100 points.

6. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The anomaly detection unit of the state evaluation layer adopts an anomaly detection model based on the isolated forest algorithm, which maps feature data to anomaly score space and triggers an alarm when the anomaly score exceeds a preset threshold.

7. The robot arm automated operation and maintenance management system with integrated status monitoring according to claim 1, characterized in that, The remaining lifetime prediction model of the fault prediction layer is a three-layer LSTM network structure. The input layer receives time series data of health indicators, the hidden layer has 128 nodes, and the output layer outputs the remaining lifetime prediction value. The model is trained using historical health degradation trajectory data as the training set and mean squared error as the loss function.

8. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The maintenance strategy generation unit of the operation and maintenance decision layer includes: determining maintenance priority based on the predicted remaining service life and a preset service life threshold range; the maintenance priority is divided into three levels: emergency maintenance, planned maintenance, and status observation.

9. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The visualization and interaction layer includes: a 3D digital twin model of the robotic arm built based on Unity3D or Three.js, which is synchronized with the joint angles of the physical robotic arm in real time; and a health score superimposed on each joint of the 3D model in the form of a heat map.

10. The automated operation and maintenance management system for robotic arms with integrated status monitoring according to claim 1, characterized in that, The system also includes a data storage layer, which is communicatively connected to the above layers. It is used to store raw status data, feature data, health score history and maintenance records, and supports retrieval by time range and device number.

Images