A machine learning method and system

Abstract

This invention provides a machine learning method and system, relating to the field of data processing technology. The method includes: Step 1, acquiring real-time sensor data streams from industrial equipment at an edge computing node, the data streams including the end-effector temperature value of a welding robot, the drive motor current value, and the robotic arm vibration frequency value; Step 2, based on the real-time sensor data streams, extracting feature values ​​of the starting positioning point, the midpoint of the path, and the ending positioning point in the welding trajectory through an edge intelligent processing unit, and dynamically generating anomaly judgment thresholds based on the feature values. This invention can quickly respond to instantaneous anomalies, improve the ability to identify and predict progressive faults, reduce equipment failure risks and maintenance costs, and ensure production safety and quality.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a machine learning method and system. Background Technology

[0002] In edge computing monitoring of industrial equipment (such as welding robots), it is common practice to use real-time sensor data (temperature, current, vibration, etc.) for anomaly detection and immediate control (power outage, speed adjustment, etc.). However, existing edge-side solutions have certain shortcomings when dealing with progressive faults that indicate slow deterioration of equipment health (such as the gradual change in vibration characteristics caused by wear on robotic arm bearings).

[0003] While most systems can record sensor data and control commands, there is room for improvement in integrating and analyzing the long-term correlation between real-time control operations (such as speed regulation) and equipment status (such as vibration changes). In some cases, historical data generated by control commands that respond to instantaneous anomalies and their impact on equipment status (especially vibration modes) are not stored in the system.

[0004] This may make it difficult for the diagnostic system to fully utilize the state evolution information of the equipment during long-term operation (especially after control intervention), and the accuracy and predictive ability of early identification of progressive problems may be limited, affecting the effectiveness of predictive maintenance. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a machine learning method and system to reduce the risk of equipment failure and ensure the safety of industrial production.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] In a first aspect, a machine learning method, the method comprising:

[0008] Step 1: Collect real-time sensor data streams from industrial equipment at the edge computing node. The data streams include the end temperature value of the welding robot, the current value of the drive motor, and the vibration frequency value of the robotic arm.

[0009] Step 2: Based on the real-time sensor data stream, the feature values ​​of the starting point, midpoint, and ending point in the welding trajectory are extracted by the edge intelligent processing unit, and an anomaly judgment threshold is dynamically generated based on the feature values.

[0010] Step 3: Based on the abnormal judgment threshold, when the current value of the drive motor exceeds the threshold, a power cut-off command is triggered, or when the vibration frequency value of the robotic arm deviates from the normal range, a welding speed adjustment command is generated.

[0011] Step 4: Aggregate the sensor data stream, anomaly detection threshold, and welding speed adjustment command into time-series simulated units according to the equipment entity and store them.

[0012] Step 5: Based on the historical vibration frequency value sequence and corresponding control command records in the simulated unit, construct a progressive fault diagnosis model and update the decision parameters of the progressive fault diagnosis model to obtain an updated progressive fault diagnosis model.

[0013] Step 6: Based on the updated progressive fault diagnosis model, output a report on the evolution sequence of the robot arm's vibration characteristics.

[0014] Secondly, a machine learning system includes:

[0015] The acquisition module is used to acquire real-time sensor data streams of industrial equipment at the edge computing node. The data streams include the end temperature value of the welding robot, the current value of the drive motor, and the vibration frequency value of the robotic arm.

[0016] The threshold generation module is used to extract the feature values ​​of the starting point, midpoint and ending point of the welding trajectory based on the real-time sensor data stream through the edge intelligent processing unit, and dynamically generate anomaly judgment thresholds based on the feature values.

[0017] The control module is used to trigger a power cut-off command when the current value of the drive motor exceeds the threshold, or to generate a welding speed adjustment command when the vibration frequency value of the robotic arm deviates from the normal range, based on an anomaly detection threshold.

[0018] The storage module is used to aggregate sensor data streams, anomaly detection thresholds, and welding speed adjustment commands into time-series-based virtual units for storage, based on the device entity.

[0019] The update module is used to construct a progressive fault diagnosis model and update the decision parameters of the progressive fault diagnosis model based on the historical vibration frequency value sequence and corresponding control command records in the simulated unit, so as to obtain the updated progressive fault diagnosis model.

[0020] The report output module is used to output a report on the evolution sequence of robot arm vibration characteristics based on the updated progressive fault diagnosis model.

[0021] Thirdly, a computing device includes:

[0022] One or more processors;

[0023] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0024] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0025] The above-described solution of the present invention has at least the following beneficial effects:

[0026] By collecting real-time sensor data through edge computing nodes and dynamically generating anomaly judgment thresholds, it can quickly respond to instantaneous anomalies (such as overcurrent in drive motors) and promptly trigger safety commands such as power cut-off, reducing the risk of equipment failure and ensuring industrial production safety. In case of deviation of the robotic arm vibration frequency, it generates welding speed adjustment commands that match the degree of anomaly, realizing fine-tuning of equipment operation status and reducing product quality problems caused by abnormal vibration.

[0027] By aggregating sensor data, thresholds, and control commands into time-series simulated units based on the physical equipment, structured data management is achieved, facilitating the tracing of historical equipment status. Based on historical data in the simulated units, a progressive fault diagnosis model is constructed and dynamically updated, fully integrating the correlation information between control intervention and status evolution during long-term equipment operation. This improves the early identification accuracy and prediction capability of slow deterioration faults such as wear on robotic arm bearings, providing strong support for predictive maintenance. The output vibration characteristic evolution sequence report includes information such as characteristic changes, intervention correlations, and wear predictions, reducing equipment downtime and maintenance costs. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a machine learning method provided by an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of a machine learning system provided by an embodiment of the present invention. Detailed Implementation

[0030] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0031] like Figure 1 As shown, an embodiment of the present invention proposes a machine learning method, the method comprising the following steps:

[0032] Step 1: Collect real-time sensor data streams from industrial equipment at the edge computing node. The data streams include the end temperature value of the welding robot, the current value of the drive motor, and the vibration frequency value of the robotic arm.

[0033] Step 2: Based on the real-time sensor data stream, the feature values ​​of the starting point, midpoint, and ending point in the welding trajectory are extracted by the edge intelligent processing unit, and an anomaly judgment threshold is dynamically generated based on the feature values.

[0034] Step 3: Based on the abnormal judgment threshold, when the current value of the drive motor exceeds the threshold, a power cut-off command is triggered, or when the vibration frequency value of the robotic arm deviates from the normal range, a welding speed adjustment command is generated.

[0035] Step 4: Aggregate the sensor data stream, anomaly detection threshold, and welding speed adjustment command into time-series simulated units according to the equipment entity and store them.

[0036] Step 5: Based on the historical vibration frequency value sequence and corresponding control command records in the simulated unit, construct a progressive fault diagnosis model and update the decision parameters of the progressive fault diagnosis model to obtain an updated progressive fault diagnosis model.

[0037] Step 6: Based on the updated progressive fault diagnosis model, output a report on the evolution sequence of the robot arm's vibration characteristics.

[0038] In this embodiment of the invention, by collecting real-time sensor data through edge computing nodes and dynamically generating anomaly judgment thresholds, it can quickly respond to instantaneous anomalies (such as overcurrent in the drive motor) and promptly trigger safety commands such as power cut-off, thereby reducing the risk of equipment failure and ensuring industrial production safety. In the case of deviation of the vibration frequency of the robotic arm, a welding speed adjustment command matching the degree of anomaly is generated to achieve fine adjustment of the equipment operating status and reduce product quality problems caused by abnormal vibration.

[0039] By aggregating sensor data, thresholds, and control commands into time-series simulated units based on the physical equipment, structured data management is achieved, facilitating the tracing of historical equipment status. Based on historical data within these simulated units, a progressive fault diagnosis model is built and dynamically updated. This fully integrates the correlation information between control interventions and status evolution during long-term equipment operation, improving the early identification accuracy and predictive capability of slow-deterioration faults such as robotic arm bearing wear, providing strong support for predictive maintenance. The output vibration characteristic evolution sequence report includes information on characteristic changes, intervention correlations, and wear predictions, reducing equipment downtime and maintenance costs.

[0040] In a preferred embodiment of the present invention, step 1 involves acquiring real-time sensor data streams from industrial equipment at an edge computing node, the data streams including the welding robot end-effector temperature value, the drive motor current value, and the robotic arm vibration frequency value; step 2 includes:

[0041] Step 200: Extract the temperature change rate characteristics of the starting point of the welding trajectory, the current distribution statistics of the midpoint of the welding path, and the vibration frequency domain envelope characteristics of the welding termination point based on the real-time sensor data stream.

[0042] Step 201: Combine the temperature change rate feature with the real-time welding speed parameter and perform reverse weight adjustment to generate temperature change weighted feature; combine the current distribution statistical feature with the real-time welding material thickness parameter and perform forward weight adjustment to generate current distribution weighted feature; combine the vibration frequency domain envelope feature with the equipment cumulative working time parameter and perform forward weight adjustment to generate vibration frequency domain weighted feature.

[0043] Step 202: Dynamically generate anomaly judgment threshold by integrating temperature change weighted features, current distribution weighted features, and vibration frequency domain weighted features.

[0044] In this embodiment of the invention, the edge computing node (such as an industrial-grade edge gateway or embedded controller) establishes a real-time communication link with the sensor system of the welding robot and collects three types of core data according to preset rules. The specific process is as follows:

[0045] Temperature acquisition at the welding robot end effector: A high-precision thermocouple sensor (measurement range -50℃ to 1000℃) is installed at the end effector of the robot's welding torch. The sensor is connected to the edge node via an industrial bus (such as Modbus RTU). The edge node triggers acquisition at a sampling period of 50 milliseconds. Each acquisition first performs sliding window filtering on the raw temperature signal (the window size is 3 sampling points, and the average value is taken) to remove instantaneous jump values ​​caused by arc interference (if a single jump exceeds 50℃, it is judged as interference and replaced with the valid value of the previous moment). After acquisition, each temperature value is marked with a precise timestamp (accurate to milliseconds) and temporarily stored in the circular buffer of the edge node in the format of "acquisition time-temperature value".

[0046] Drive motor current acquisition: A Hall current sensor (measurement range 0~50A) is connected in series in the power supply circuit of the robot drive motor. The analog signal output by the sensor is converted into a digital signal by a 16-bit analog-to-digital converter and then transmitted to the edge node. The sampling period is synchronized with the temperature acquisition (50 milliseconds / time). During acquisition, the motor operation stage is recorded synchronously (the "arc ignition", "welding", and "arc termination" status signals output by the robot controller). For example, the current value is only included in the valid data when the status signal is "welding". For the acquired current values, the edge node performs outlier pruning (retaining data within the range of 0~40A, recording the excess as 40A to avoid extreme values ​​caused by sensor failure) and stores it with a timestamp.

[0047] Robotic arm vibration frequency acquisition: A three-axis accelerometer (measurement range ±16g, sampling rate 1000Hz) is attached to each of the upper and lower arm joints of the robotic arm. The sensor sends vibration time-domain data (1000 acceleration values ​​per second) to the edge node via a wireless transmission module (such as Wi-Fi 6). After receiving the data, the edge node first performs preprocessing: removing the gravitational acceleration component (through high-pass filtering, cutoff frequency 0.5Hz), then dividing the preprocessed time-domain signal into frames (each frame contains 1024 sampling points, frame overlap rate 50%), performing a Fast Fourier Transform (FFT) on each frame of data to convert it to the frequency domain, obtaining the frequency-amplitude distribution in the range of 0~500Hz, and finally extracting the top 5 frequency values ​​with the highest amplitude from the frequency domain results (such as 20Hz, 50Hz, 100Hz, etc.) as the robotic arm vibration frequency values ​​for that sampling period, and storing them with a timestamp.

[0048] After the three types of data are collected, the edge nodes integrate them into a real-time sensing data stream according to the structure of "device ID-timestamp-temperature-current-vibration frequency", and continuously push it to the input buffer of the edge intelligent processing unit.

[0049] Step 200: When the robot coordinates reach the starting point of the welding trajectory (preset coordinates X0, Y0, Z0), the starting positioning point marker is triggered. Temperature data (a total of 13 temperature values) from the 6 sampling periods (300 ms) before and after the marker time are selected and arranged in chronological order as T0, T1, ..., T12. The temperature difference between two adjacent moments is calculated: ΔT1 = T1 - T0, ΔT2 = T2 - T1, ..., ΔT12 = T12 - T11. Each difference corresponds to a time interval of 50 ms (0.05 seconds). Therefore, the instantaneous temperature change rate is v1 = ΔT1 / 0.05, v2 = ΔT2 / 0.05, ..., v12 = ΔT12 / 0.05. The two instantaneous change rates with the largest absolute values ​​are removed (to exclude extreme interference), and the arithmetic mean of the remaining 10 instantaneous temperature change rates is taken as the temperature change rate characteristic of the starting positioning point, reflecting the temperature rise trend in the initial stage of welding. When the robot's coordinates reach 50% of the total trajectory length (preset coordinates Xm, Ym, Zm), it is marked as the midpoint of the path. Current data (a total of 25 current values) are selected from the 12 sampling periods (600 ms) before and after the marked time, denoted as I0, I1, ..., I24. These current values ​​are statistically calculated as follows: Average = (I0 + I1 + ... + I24) / 25, reflecting the average motor load at the midpoint; Variance = [(I0 - Average)² + (I1 - Average)² + ... + (I24 - Average)²] / 25, reflecting the degree of current fluctuation; Range = max(I0, ..., I24) - min(I0, ..., I24), reflecting the range of current variation. These values ​​are combined in the order of "Average - Variance - Range" to represent the statistical characteristics of the current distribution at the midpoint of the path.

[0050] Vibration frequency domain envelope feature extraction at the termination positioning point: When the robot coordinates reach the end point of the welding trajectory (preset coordinate X... e Y e Z e When the time reaches 0, mark it as the termination positioning point; select vibration frequency domain data (a total of 17 sets of frequency-amplitude distributions) for the 8 sampling periods (400 ms) before and after the marking time, with each set of data containing amplitude information from 0 to 500 Hz. For each frequency point (e.g., 10 Hz, 20 Hz... 500 Hz), extract the maximum amplitude value from the 17 sets of data to form a "frequency-maximum amplitude" envelope. Set an amplitude benchmark value (1.2 times the average amplitude of the termination point during normal welding of the robot in the past 30 days), and filter all frequency points with amplitudes exceeding the benchmark value from the envelope, listing them in the format of "frequency value-amplitude value" as the vibration frequency domain envelope feature of the termination positioning point, reflecting the abnormal vibration distribution at the end of welding.

[0051] Step 201: Based on the basic features extracted in step 200, and combined with real-time process parameters and equipment status parameters, weighted features are generated through weight adjustment.

[0052] Temperature Change Weighted Feature (Reverse Weight Adjustment): Based on real-time welding speed parameters (provided by the robot control system, unit: mm / s, range: 1~10 mm / s), the adjustment adopts a reverse logic of "higher speed, lower weight" (high-speed welding results in rapid temperature changes and a higher proportion of non-fault-related fluctuations). A welding speed baseline of 5 mm / s is set, and the weighting coefficients are obtained: if the real-time speed V ≤ 5 mm / s, the weighting coefficient Kt = 1; if the real-time speed V > 5 mm / s, the weighting coefficient Kt = 5 / V (e.g., when V = 8 mm / s, Kt = 5 / 8). The temperature change rate feature extracted in step 200 is multiplied by Kt to obtain the temperature change weighted feature, thus weakening the influence of non-fault-related temperature fluctuations during high-speed welding.

[0053] Current distribution weighted feature (positive weight adjustment): Based on the real-time welding material thickness parameter (input from the production work order system, unit: mm, range: 0.5~10mm), the adjustment adopts the positive logic of "the greater the thickness, the higher the weight" (welding thicker materials requires a larger current, and current stability has a more significant impact on quality). A material thickness baseline of 3mm is set, resulting in the following weighting coefficients: if the real-time thickness D≤3mm, weighting coefficient Kj=1; if the real-time thickness D>3mm, weighting coefficient Kj=D / 3 (e.g., when D=6mm, Kj=6 / 3=2). The current distribution statistical features extracted in step 200 (mean, variance, and range multiplied by Kj) are combined to obtain the current distribution weighted feature, emphasizing the focus on current fluctuations during thick material welding.

[0054] Vibration frequency domain weighted features (positive weight adjustment): The cumulative working time parameter of the equipment (total robot running time recorded by edge nodes, in hours, ranging from 0 to 10000 hours) is used as the adjustment basis, adopting the positive logic of "the longer the duration, the higher the weight" (as the cumulative duration increases, the risk of wear on the robotic arm increases, and vibration features need more attention); the cumulative duration benchmark value is set to 1000 hours, and the weight coefficient is obtained: if the cumulative duration H≤1000 hours, the weight coefficient Kv=1; if the cumulative duration H>1000 hours, the weight coefficient Kv=H / 1000 (e.g., if H=1500 hours, Kv=1500 / 1000). The vibration frequency domain envelope features (amplitude value of each frequency point multiplied by Kv) extracted in step 200 are listed to obtain the vibration frequency domain weighted features, which improves the sensitivity of old equipment to vibration anomalies.

[0055] Step 202: Combine the three weighted features obtained in step 201 to dynamically generate an anomaly detection threshold adapted to the real-time scenario. The process is as follows:

[0056] Retrieve the three weighted feature data of the robot during normal welding over the past 20 days from the simulated memory area, and calculate the historical maximum and minimum values ​​of each feature (e.g., the historical range of the temperature change weighted feature is 0.2~3.5℃ / s, and the historical range of the average value in the current distribution weighted feature is 5~25A, etc.). The current temperature change weighted feature, current distribution weighted feature, and vibration frequency domain weighted feature are mapped to the 0~1 interval according to the formula "(current value - historical minimum value) / (historical maximum value - historical minimum value)" to eliminate the difference in dimensions. Based on the normalized historical normal data, the normal fluctuation range of each feature is calculated: the mean μ and standard deviation σ of the historical data are taken to obtain the normal range of [μ-2σ, μ+2σ] (covering 95% of the normal data). If a feature has been continuously in [μ+1.5σ, μ+2σ] (close to the historical limit) in the last 8 sampling periods, the upper limit of the feature is temporarily widened to μ+2.5σ (to avoid misjudgment due to short-term fluctuations); if a feature has been in [μ-σ, μ+σ] (stable fluctuation) in the last 15 sampling periods, the upper and lower limits of the feature are tightened to [μ-1.5σ, μ+1.5σ] (to improve detection sensitivity).

[0057] A fusion weight is assigned to the three features (based on historical fault data verification, temperature change weighted feature accounts for 30%, current distribution weighted feature accounts for 40%, and vibration frequency domain weighted feature accounts for 30%). The adjusted upper limits of the normal range of each feature are then summed according to their weights to obtain the final anomaly judgment threshold. For example, if the upper limit of the temperature feature is 0.8, the upper limit of the current feature is 0.7, and the upper limit of the vibration feature is 0.9, then the threshold = 0.8 × 30% + 0.7 × 40% + 0.9 × 30% = 0.79.

[0058] By selecting appropriate sensors, designing sampling cycles, and preprocessing (filtering, outlier handling), we ensure that temperature, current, and vibration data accurately reflect the equipment status. We focus on extracting features from three key nodes—the start, midpoint, and end of the welding trajectory—avoiding data redundancy throughout the process and making the features more closely match the characteristics of equipment operation at each stage, thus improving the correlation with fault states. We combine real-time parameters such as welding speed, material thickness, and cumulative equipment time for reverse / positive weight adjustments, enabling features to dynamically adapt to different process conditions and equipment aging states, reducing interference from non-fault factors. Based on historical data and real-time fluctuation trends, we dynamically adjust the threshold range, preserving the baseline of normal conditions while addressing different situations such as short-term fluctuations or stable operation. This significantly reduces the risk of false alarms (relaxed when approaching the upper limit) or missed alarms (tightened when stable) caused by fixed thresholds, providing accurate judgment criteria for abnormal responses. Precise features, adaptive weighting, and dynamic thresholds make the data stored in the virtual unit more valuable for analysis, providing high-quality training data for the progressive fault diagnosis model and helping to improve the model's accuracy in identifying slowly deteriorating equipment states.

[0059] In a preferred embodiment of the present invention, step 3 includes:

[0060] The system compares the current value of the drive motor with the abnormal judgment threshold in real time. When the current value continuously exceeds the abnormal judgment threshold within a continuous sampling period, a power cut-off command is triggered. The system calculates the envelope offset area between the vibration spectrum and the abnormal judgment threshold based on the vibration frequency value of the robotic arm. When the envelope offset area exceeds the preset tolerance threshold, a welding speed adjustment command is generated, and the adjustment range of the welding speed adjustment command increases with the increase of the envelope offset area.

[0061] In this embodiment of the invention, based on the real-time collected drive motor current value and the anomaly judgment threshold generated in step 202, a safety command is triggered through continuous monitoring and logical judgment. The specific process is as follows:

[0062] Edge computing nodes acquire the drive motor current value at a sampling period of 50 milliseconds / time (consistent with the current acquisition period). After each acquisition, the current current value is immediately compared with the anomaly judgment threshold. If the current value is less than or equal to the anomaly judgment threshold, it is marked as "normal state" and the continuous over-limit counter is reset (the counter value is set to zero). If the current value is greater than the anomaly judgment threshold, it is marked as "over-limit state" and the continuous over-limit counter value is incremented by 1 (for example, the first over-limit counter = 1, the second continuous over-limit counter = 2, and so on).

[0063] When the consecutive over-limit counter value reaches 3 times (i.e., 3 consecutive sampling periods, totaling 150 milliseconds), it is determined to be a "continuous current anomaly". At this time, the control logic module of the edge computing node immediately generates a power cut-off command. The command format includes the device ID (e.g., "welding robot #001"), command type ("emergency power cut-off"), trigger reason ("drive motor current exceeds the limit 3 times consecutively, current value XXA, threshold XXA"), and timestamp. The generated power cut-off command is converted into a protocol that the robot control system can recognize (e.g., Profinet) through the communication protocol conversion module and transmitted to the robot's power control unit via industrial Ethernet. After receiving the command, the power control unit immediately cuts off the main power supply circuit of the drive motor and returns a "command executed successfully" feedback signal to the edge node. The edge node records the command trigger time and execution result and stores them synchronously in the virtual memory area.

[0064] Frequency domain data of the robotic arm vibration is extracted from the real-time sensor data stream (from the 0-500Hz spectrum transformed by FFT in step 1, containing the amplitude value at each frequency point); simultaneously, the anomaly judgment threshold generated in step 202 is converted into a "vibration frequency domain threshold envelope"—that is, based on the threshold value, a corresponding upper limit line of amplitude is generated in the 0-500Hz frequency range (for example, the threshold is mapped to "the maximum allowable amplitude A at frequency f"). th(f)”). Calculate the “offset area” between the vibration spectrum and the threshold envelope (only counting the portion of the spectrum exceeding the threshold), specifically including:

[0065] The frequency range of 0-500Hz is divided into 50 intervals of 10Hz each (i.e., 0-10Hz, 10-20Hz, ..., 490-500Hz). The midpoint of each interval (e.g., 5Hz, 15Hz...495Hz) is used as the calculation benchmark. For each midpoint, the actual amplitude A of the vibration spectrum is calculated. a (f) and the amplitude A of the threshold envelope th The difference between (f) and A a (f)≤A th (f), the difference is 0; if A a (f)>A th (f), the difference is A a (f)-A th (f); The width of each frequency interval is 10Hz. Multiply the amplitude difference of the interval by 10Hz to obtain the offset area of ​​a single interval. Accumulate the offset areas of 50 intervals, and the sum is the "enveloping offset area between the vibration spectrum and the anomaly judgment threshold" (unit: Hz•mm / s², amplitude unit is mm / s²).

[0066] A preset tolerance threshold for the envelope offset area is set (based on historical normal welding data, such as 500Hz•mm / s²). If the calculated envelope offset area is less than or equal to the tolerance threshold, it is judged as "slight vibration abnormality" and no adjustment command is generated. If the area is greater than the tolerance threshold, it is judged as "significant vibration abnormality" and a welding speed adjustment command is triggered. The adjustment range is directly proportional to the envelope offset area. The specific rule is: for every 100Hz•mm / s² exceeding the tolerance threshold, the welding speed is reduced by 2% (the reference speed is the current welding speed V0). For example, if the envelope offset area = 700Hz•mm / s² (exceeding the tolerance threshold of 200Hz•mm / s²), then the adjustment range = 200 / 100 × 2% = 4%, that is, the new welding speed = V0 × (1 - 4%). To avoid the welding quality being affected by excessively low speed, the upper limit of the adjustment range is set to 20% (i.e., a maximum reduction of 20%). Even if the offset area far exceeds the threshold, the speed will not be lower than 80% of the original speed. The control logic unit generates welding speed adjustment instructions, including the device ID, target speed value, adjustment range, trigger reason ("vibration envelope offset area XX, exceeding the tolerance threshold XX") and timestamp. These instructions are converted into a robot-recognizable format (such as Modbus TCP) by the communication protocol conversion unit and sent to the robot's motion control unit. At the same time, the instruction information is recorded in the virtual memory area.

[0067] By using a "continuous sampling period exceeding limit" judgment logic (rather than a single exceeding limit), false triggers caused by instantaneous current fluctuations are effectively filtered out, ensuring that power cut-off commands are only triggered during persistent anomalies. This protects equipment safety and reduces unnecessary downtime losses. The degree of vibration anomaly is quantified based on the "envelope offset area," making the welding speed adjustment range proportional to the severity of the anomaly (the larger the area, the larger the adjustment range). This avoids a "one-size-fits-all" approach to adjustment, suppressing excessive vibration while minimizing the impact on welding efficiency and quality. Power cut-off targets fatal current anomalies (rapid loss mitigation), while speed adjustment targets non-fatal vibration anomalies (dynamic optimization). The combination of these two strategies achieves a balance between "ensuring safety in emergencies and maintaining production in minor anomalies," improving the stability of industrial production.

[0068] In a preferred embodiment of the present invention, step 4 involves aggregating the sensor data stream, the anomaly detection threshold, and the welding speed adjustment command into time-series-based virtual units for storage, based on the equipment entity; step 5 includes:

[0069] Step 500: Use the historical vibration frequency value sequence and corresponding welding speed adjustment command record in the time-seriesd simulation unit to construct an initial progressive fault diagnosis model;

[0070] Step 501: Obtain the vibration frequency value sequence and welding speed adjustment command record of the latest acquisition cycle, input the latest data into the initial progressive fault diagnosis model, update the vibration characteristic decision parameters, and generate an updated progressive fault diagnosis model, specifically including:

[0071] Step 5010: Obtain the spectral envelope feature value of the latest vibration frequency value sequence;

[0072] Step 5011: Compare the spectral envelope feature values ​​with the historical spectral benchmark in the initial progressive fault diagnosis model to generate a feature offset data sequence.

[0073] Step 5012: Process the feature offset data sequence, identify sampling periods that continuously exceed the offset threshold, and generate an over-limit period marker sequence;

[0074] Step 5013: Based on the duration of the marked sequence of the out-of-limit period, adjust the vibration feature weight coefficients to generate preliminary corrected weight coefficients;

[0075] Step 5014: Extract the number of interventions from the welding speed adjustment command record, and perform attenuation processing on the preliminary correction weight coefficient based on the intervention frequency to generate updated vibration characteristic decision parameters.

[0076] Step 5015: Write the updated vibration characteristic decision parameters into the initial progressive fault diagnosis model to generate an updated progressive fault diagnosis model.

[0077] In this embodiment of the invention, various types of data related to device operation are aggregated according to device entities and stored as virtual units in time series format. The specific process is as follows:

[0078] Edge computing nodes first distinguish different industrial devices by their unique identifiers (such as the serial number "WR-2023-005" for a welding robot), create an independent storage directory for each device, and then categorize the data that needs to be stored:

[0079] Sensor data stream: includes welding robot end temperature value (one value every 50 milliseconds, with timestamp), drive motor current value (same frequency, with running stage marker), and robotic arm vibration frequency value (same frequency, with frequency domain characteristics).

[0080] Anomaly detection threshold: The dynamic threshold generated in step 202 (updated every 1 second, with generation timestamp and corresponding feature fusion logic description);

[0081] Welding speed adjustment command: The command generated in step 3 (including command ID, trigger time, target speed, adjustment range, trigger reason, etc.).

[0082] Data is aggregated using "time windows": a basic time window is set at 1 minute (which can be adjusted according to the equipment's operating rhythm, such as 5 minutes when the welding cycle is long). The data in each window is arranged in ascending order by timestamp, forming "simulated unit segments". Each segment contains: all sensor data in the window (structured by "timestamp-temperature-current-vibration frequency"); and the anomaly judgment threshold generated in the window (if updated once per second, each 1-minute window contains 60 threshold records).

[0083] Welding speed adjustment commands triggered within this window (if any, sorted by trigger time; otherwise marked "No Command"). The virtual unit for each device is stored in a hierarchical structure of "Device ID - Date - Time Window" (e.g., "WR-2023-005 / 20240901 / 08:00-08:01"), and a dual index is established:

[0084] Time Index: Records the start / end timestamps of each realistic unit segment, supporting queries by time period;

[0085] Data type index: Distinguish between sensor data, thresholds, and instructions within each segment, and support quick extraction by data type (such as retrieving adjustment instructions for a specific time period).

[0086] The storage medium uses local solid-state drives (SSDs) at the edge nodes, while backing up to the cloud-based virtual memory area daily to ensure data security and long-term traceability.

[0087] Step 500, the construction process of the initial progressive fault diagnosis model: retrieve the time-series data of the target equipment over the past 6 months from the simulated unit, which should cover the complete life cycle of "no fault - slight wear - moderate wear" (to be confirmed in conjunction with the equipment maintenance log, such as including one minor bearing wear repair record); extract the frequency domain data of the robotic arm vibration within each time window, focusing on retaining the 50-200Hz frequency band (based on the vibration characteristics of bearing wear, this frequency band is the wear-sensitive area), and splice them into a continuous sequence in chronological order (one data point per minute, including the average amplitude and dominant frequency value of 50-200Hz within the window); extract the instructions aligned with the vibration sequence time, which should include "trigger timestamp, adjustment amplitude, and vibration offset area of ​​the corresponding window", and group them by "vibration window timestamp", and count the number of instructions and average adjustment amplitude within each window (e.g., "2 instructions were triggered within the 08:00-08:01 window, with an average adjustment amplitude of -5%"). Remove data from periods of equipment shutdown, calibration, and maintenance (identified by the "equipment status label" in the virtual unit; for example, data from a window in the "shutdown" state is directly removed); if the amplitude of a certain window changes by more than 30% compared to the previous window (judged as sensor interference), replace it with the average of the three windows before and after; establish a "vibration-command" time correlation table: that is, using the vibration window timestamp as the key, associate the vibration characteristics and adjustment commands within that window to ensure that each vibration data can correspond to the intervention behavior of the same period (if there is no command, mark it as "no intervention").

[0088] Model input features and fault label definitions: Input features (all values ​​normalized to 0-1) include core vibration features: average amplitude (F1) in the 50-200Hz frequency band, dominant frequency value (F2, e.g., 100Hz), and daily amplitude growth rate (F3, i.e., average amplitude of the day / average amplitude of the previous day - 1); intervention correlation features: total number of welding speed adjustment commands in the past 3 days (F4), absolute value of average adjustment amplitude (F5), and average interval between two commands (F6). Based on equipment maintenance logs and vibration features, historical data is labeled into 3 states:

[0089] Condition 1 (no wear): Average amplitude of 50-200Hz < 120% of the historical average during the fault-free period, and no adjustment command;

[0090] Condition 2 (Slight Wear): Average amplitude is 120%-150% of the historical average, and adjustment commands occur 1-2 times per week;

[0091] Condition 3 (moderate wear): The average amplitude is 150%-200% of the historical average, and adjustment commands occur 3-5 times per week.

[0092] The system initiates a "feature-state" co-occurrence analysis algorithm to automatically calculate the distribution range of each feature under different states: For 1000 data points of state one, the system automatically calculates the sample proportion (92%) of F1<0.3, F3<0.02, and F4=0, forming a feature association pattern for state one; For 800 data points of state two, the system automatically calculates the sample proportion (88%) of F1 in the range of 0.3-0.5, F3 in the range of 0.02-0.05, and F4=1-2, forming a feature association pattern for state two; The system automatically extracts the vibration data of state one from the preprocessed data for the previous 3 months, calculates the mean value of the average amplitude of 50-200Hz (e.g., 0.04mm / s²), and stores it as the baseline value (B). Weights are automatically assigned by calculating the coefficient of variation of each feature across different states (i.e., first calculating the standard deviation and mean of all sample values ​​for each feature in states one, two, and three, then using the formula "coefficient of variation = standard deviation / mean" to obtain the coefficient of variation of that feature in each state, and finally synthesizing the overall coefficient of variation of that feature across all states, thus reflecting the feature's ability to distinguish between different states).

[0093] F1 (average amplitude): highest coefficient of variation (0.42), automatically assigned weight 0.35; F3 (daily amplitude growth rate): second highest coefficient of variation (0.31), automatically assigned weight 0.25; F4 (number of commands): coefficient of variation 0.28, automatically assigned weight 0.2; F2 (main frequency value): coefficient of variation 0.15, automatically assigned weight 0.1; F5, F6: coefficient of variation < 0.1, automatically assigned weight 0.05 each; the total weight is automatically verified to 1 by the system to ensure the consistency of the diagnostic logic.

[0094] Step 5010: For vibration data from the "latest acquisition period" (preset to the most recent 30 minutes), automatically execute the following process:

[0095] The system automatically retrieves 30 minutes of vibration frequency domain data from the real-time cache of edge computing nodes, and automatically segments it into 1 minute / frames (synchronized with the time window of the simulated unit). For each frame of data, the system automatically calculates the amplitude values ​​of all frequency points within the 50-200Hz frequency band, and extracts the "maximum amplitude" (vertex of the envelope) and "average amplitude" (overall energy) of that frequency band. The system automatically generates feature values ​​for each window in frequency order: "average amplitude (E1) of 50-200Hz, frequency corresponding to the maximum amplitude (E2)" (e.g., "window 09:00-09:01: E1=0.055mm / s², E2=110Hz"). The system automatically stitches together the features of 30 windows in ascending order of timestamp to form the latest spectral envelope feature sequence (with timestamp).

[0096] Step 5011: Retrieve the historical spectrum benchmark stored in the initial model and automatically complete the following operations:

[0097] Loading reference parameters: B1=0.04mm / s² (average amplitude reference for 50-200Hz), B2=100Hz (frequency reference corresponding to maximum amplitude); For each window of the latest sequence, the offset is automatically calculated: average amplitude offset (D1) = (current E1 - B1) / B1 × 100% (positive value indicates higher than the reference); main frequency offset (D2) = current E2 - B2 (positive value indicates offset to higher frequency); According to the window timestamp order, the system automatically combines D1 and D2 to generate a "feature offset data sequence" (e.g., "09:00-09:01: D1=37.5%, D2=10Hz; 09:01-09:02: D1=40%, D2=12Hz").

[0098] Step 5012: Based on the correlation between feature offset and fault state, automatically perform anomaly identification:

[0099] Load offset judgment threshold: D1=20% (the lowest offset of state two, i.e., 120% of B1), D2=5Hz (the critical value based on wear feature statistics); the system automatically compares D1 and D2 for each window. When D1>20% and D2>5Hz, it is automatically marked as an "abnormal window"; the system automatically traverses the offset sequence. When three or more consecutive windows are "abnormal windows" and the average D1 of these three windows is >25%, it is automatically judged as a "continuous abnormal period"; the start / end window timestamp, duration, and average D1 / D2 of each period are recorded (e.g., "start 09:00-09:01, end 09:02-09:03, lasting 3 minutes, average D1 39%, average D2 11Hz").

[0100] Step 5013: Based on the duration of the exceeding period, dynamically adjust the weight of the vibration characteristics to enhance the influence of abnormal characteristics. The process is as follows:

[0101] Load the initial weights, including F1=0.35 and F2=0.1 (inheriting the automatic allocation result from step 500).

[0102] Automatic adjustment based on preset logic: For 1 minute (1 window), F1 weight = 0.35 + 0.35 × 10% = 0.385, F2 weight = 0.1 + 0.1 × 10% = 0.11; for 2 minutes (2 windows), F1 = 0.35 + 0.35 × 10% × 2 = 0.42, F2 = 0.1 + 0.1 × 10% × 2 = 0.12; for ≥5 minutes, the F1 upper limit is automatically locked at 0.5, and the F2 upper limit is locked at 0. 2. (Avoid excessive influence from features); In case of abnormal interruption, the weights will automatically fall back to the initial value × 1.1 (e.g., F1 = 0.35 × 1.1 = 0.385); The system traverses the latest 30 windows, collects the adjusted F1 and F2 weights for each window, and then calculates the arithmetic mean of the 30 F1 weight values ​​and the arithmetic mean of the 30 F2 weight values ​​respectively. The result is the "preliminary optimized weights" (e.g., F1 = 0.42, F2 = 0.12).

[0103] Step 5014: Based on the intervention frequency of the welding speed adjustment command, automatically complete the weight calibration: The system automatically extracts the latest 30 minutes of adjustment commands from the simulation unit, counts the total number (e.g., 6 times), and automatically calculates the intervention frequency = 6 times / 30 minutes = 0.2 times / minute; loads the baseline intervention frequency = 0.1 times / minute (the average value automatically counted by the system based on historical normal production data), and automatically calculates it according to the formula "attenuation coefficient = 1 / (1 + actual frequency / baseline frequency)" (when 0.2 times / minute, the attenuation coefficient = 1 / (1 + 0.2 / 0.1) = 1 / 3); automatically multiplies the preliminary optimized weights by the attenuation coefficient (e.g., F1 = 0.42 × 1 / 3 = 0.14, F2 = 0.12 × 1 / 3 = 0.04), and normalizes all feature weights (ensuring the sum is 1).

[0104] Step 5015: Automatically perform parameter replacement and verification to complete model update: Automatically replace the corresponding parameters of the initial model with the update weights from step 5014 (e.g., F1=0.14, F2=0.04), retaining other parameters (e.g., F3=0.25); The system automatically extracts the vibration characteristics of the "non-abnormal window" within the latest 30 minutes and automatically calculates them according to the formula "new baseline = (old baseline × 9 + latest mean of non-abnormal features) / 10" (e.g., old B1=0.04mm / s², latest mean of non-abnormal E1=0.042mm / s², new B1=0.0402mm / s²); Automatically retrieve historical data from the past hour (including samples with known fault states, the state is set by the system). (Automatic log annotation) Input the data into the updated model to obtain the diagnostic results; Statistically measure the percentage of samples whose model diagnostic results match the actual state (e.g., 55 out of 60 windows match within 1 hour, accuracy = 55 / 60); If the accuracy improves by ≥5% compared to the initial model (e.g., from 85% to 92%), the update is confirmed to be effective; If the accuracy decreases (e.g., from 85% to 82%), the baseline intervention frequency is automatically reduced (e.g., from 0.1 times / minute to 0.08 times / minute), and steps 5014-5015 are re-executed; Generate the "Updated Progressive Fault Diagnosis Model", and synchronously store the model parameters, update timestamps, and verification results in the "Model Log" folder of the virtual unit.

[0105] By aggregating data by equipment entity and storing it in a time-series format, the problems of scattered and weakly correlated data are solved. This creates a complete time chain of "sensor data - threshold - command," providing high-quality, highly correlated historical data for model building and improving data query and traceability efficiency. The model is built based on six months of historical data from the equipment itself, combined with sensitive frequency ranges and command intervention characteristics, enabling the initial model to accurately capture the progressive failure patterns of the equipment (such as the wear vibration patterns of specific bearings). Through comparison of the latest vibration characteristics with historical benchmarks, weight adjustment of exceeding limits, and attenuation processing of intervention frequency, the model's decision parameters can adapt to changes in equipment status in real time (such as increased wear, frequent interventions, etc.), maintaining a high recognition rate for progressive failures over the long term. The model update process incorporates intervention information from welding speed adjustment commands, allowing diagnostic results to reflect the "impact of intervention measures on equipment status" (such as whether vibration is alleviated after adjustment), improving the practicality of the diagnosis and providing a more realistic basis for predictive maintenance.

[0106] In a preferred embodiment of the present invention, step 6 includes:

[0107] Step 600: Use the updated progressive fault diagnosis model to process the historical vibration frequency value sequence, extract the vibration spectrum envelope feature values ​​of continuous timestamps, and generate a feature evolution sequence sorted by acquisition time.

[0108] Step 601: Locate the trigger time node of the welding speed adjustment command in the feature evolution sequence and generate a feature sequence with time stamps;

[0109] Step 602: Analyze the spatiotemporal correlation between the feature value shift in the time-stamped feature sequence and the adjustment command to generate predictive diagnostic results for the wear state of the robotic arm bearing, specifically including:

[0110] Step 6020: Based on the time-stamped feature sequence, calculate the gradient of feature value changes between adjacent time nodes to generate a feature gradient sequence;

[0111] Step 6021: Perform time window integration on the feature gradient sequence to generate the cumulative offset;

[0112] Step 6022: Locate the corresponding segment in the cumulative offset according to the trigger time node of the welding speed adjustment command, and extract the average offset of the corresponding segment.

[0113] Step 6023: Calculate the bearing wear compensation coefficient by combining the average offset value with the adjustment range of the welding speed adjustment command;

[0114] Step 6024: Based on the cumulative offset and bearing wear compensation coefficient, generate a predictive diagnostic result of the wear state of the robotic arm bearing;

[0115] Step 603: Output a robotic arm vibration diagnosis report containing time-stamped feature sequences, correlation analysis results, and wear prediction results.

[0116] In this embodiment of the invention, historical data is processed by an updated progressive fault diagnosis model to extract the variation law of vibration characteristics over time. The specific process is as follows:

[0117] The time-series data of the target device over the past 7 days was retrieved from the simulated unit (covering a sufficient time span to reflect gradual changes). The focus was on extracting the vibration frequency value sequence of the robotic arm (one time window per minute, totaling 7 × 24 × 60 = 10080 windows). Based on the optimized sensitive frequency band (50-200Hz) in the updated progressive fault diagnosis model, two types of spectral envelope features were extracted from the vibration frequency domain data of each time window: the average amplitude of the 50-200Hz band (denoted as Favg, reflecting the overall vibration energy); and the frequency value with the largest amplitude within this band (denoted as Fpeak, reflecting the dominant vibration frequency). The extracted Favg and Fpeak were arranged in ascending order by timestamp (from the first window 7 days ago to the latest window), forming a structured sequence of "timestamp-Favg-Fpeak," i.e., the "feature evolution sequence."

[0118] Step 601: Associate the welding speed adjustment command with the characteristic evolution sequence to identify the nodes that influence the intervention behavior on the vibration characteristics. The specific process is as follows:

[0119] Extract the trigger timestamps (accurate to the minute) and adjustment ranges (e.g., -5%, -8%) of all commands in the past 7 days from the "Welding Speed ​​Adjustment Command" record of the virtual unit, and form a "Command Timestamp - Adjustment Range" list (e.g., "2024-09-01 08:30--5%; 2024-09-01 14:15--8%").

[0120] Traverse the feature evolution sequence to find the time window that perfectly matches the command timestamp (or has an error ≤ 1 minute), and add a command marker after the feature value of the window. For example, if the command for 2024-09-01 08:30 corresponds to "08:30-0.05mm / s²-105Hz" in the feature sequence, then mark it as "08:30-0.05mm / s²-105Hz-command (-5%)". Retain the unmarked window features to finally form a "time-stamped feature sequence", which contains continuous vibration feature changes and clearly marks the trigger position and intensity of the intervention command.

[0121] Step 6020: The rate of characteristic change at adjacent time points is calculated to reflect the rate of degradation of vibration characteristics. The specific process is as follows:

[0122] For time-stamped feature sequences, two consecutive time windows are selected in chronological order (with an interval of 1 minute between adjacent windows), for example, window t1 (08:30) and window t2 (08:31). The average amplitude (Favg) for the 50-200Hz frequency band is calculated as: Gradient Gaavg = (Favg of window t2 - Favg of window t1) / time interval (1 minute). A positive value indicates an increase in amplitude (deterioration), and a negative value indicates a decrease (improvement).

[0123] For the frequency value (Fpeak) with the largest amplitude in the 50-200Hz frequency band: gradient Gpeak = (Fpeak of window t2 - Fpeak of window t1) / 1 minute. A positive value indicates that the main frequency has shifted to a higher frequency (a typical feature of accelerated wear). The Gavg and Gpeak of all adjacent windows are spliced ​​in time sequence to form a "characteristic gradient sequence", for example: "08:30-08:31: Gavg=0.002mm / s² / minute, Gpeak=1Hz / minute; 08:31-08:32: Gavg=0.001mm / s² / minute, Gpeak=0.5Hz / minute".

[0124] Step 6021: Quantize the total feature shift over a period of time by integrating the gradient sequence over a time window. The specific process is as follows:

[0125] A sliding time window (1 hour in size, i.e., 60 adjacent windows) is used, sliding in 10-minute increments (ensuring coverage of all time periods with overlap to avoid information omission). The feature gradient sequences within each time window are accumulated: the cumulative Favg offset Savg = the sum of all Gavg values ​​within the window (reflecting the total increase in average amplitude within 1 hour); the cumulative Fpeak offset Speak = the sum of all Gpeak values ​​within the window (reflecting the total shift in dominant frequency within 1 hour). The cumulative offset sequences are generated by sorting by the window start time, forming a "time window - Savg - Speak" sequence, for example: "08:00-09:00: Savg=0.1mm / s², Speak=50Hz; 08:10-09:10: Savg=0.11mm / s², Speak=52Hz".

[0126] Step 6022: Analyze the cumulative offset during the period when the positioning adjustment command is triggered, and examine the vibration changes before and after the intervention. The specific process is as follows:

[0127] For each welding speed adjustment command (e.g., trigger time 08:30), its corresponding segment is defined as "30 minutes before the command to 30 minutes after the command" (a total of 1 hour, covering the complete impact cycle before and after the intervention), i.e., 08:00-09:00. The window corresponding to this time period (08:00-09:00) is found from the cumulative offset sequence, and its Savg and Speak are extracted. If the time period spans multiple windows (e.g., if the command is triggered at 08:55, the segment is 08:25-09:25, covering the 08:20-09:20 and 08:30-09:30 windows), the average of the Savg and Speak values ​​for these windows is taken as the segment offset. The arithmetic mean of the extracted segment offsets (e.g., Savg = 0.1 mm / s², 0.12 mm / s²) is calculated and used as the "mean offset" for the segment corresponding to that command.

[0128] Step 6023, Calculation of bearing wear compensation coefficient: This coefficient is used to quantify the mitigating effect of welding speed adjustment commands on bearing wear. The specific process is as follows:

[0129] The larger the adjustment range (e.g., the greater the speed reduction), the stronger the vibration suppression effect, and the larger the compensation coefficient (representing the proportion of wear reduction due to intervention). Based on the adjustment range of the command (taking the absolute value, such as -5% which is the range of 5%), a positive correlation is established between the range and the compensation coefficient: Compensation coefficient K = adjustment range × 0.02 (based on the historical data pattern that "for every 1% increase in adjustment range, the wear rate decreases by about 2%)); combined with the average offset value extracted in step 6022 (e.g., Savg average value = 0.11 mm / s²), K is corrected: if the average offset value > the historical normal offset value for the same period (e.g., 0.08 mm / s²), it indicates that the intervention effect is limited, and K is multiplied by 0.8; if ≤ 0.08 mm / s², it indicates that the intervention is effective, and K remains unchanged. For example: if the adjustment range is 5% and the average offset value is 0.11 mm / s² (> 0.08), then K = 5% × 0.02 × 0.8 = 0.0008 (i.e., 0.08%).

[0130] Step 6024: Based on the cumulative offset and compensation coefficient, determine the current wear status and predict future trends. The specific process is as follows:

[0131] Take the sum of the cumulative offset Savg over the past 24 hours (e.g., 2.5 mm / s²), subtract the sum of the compensation coefficients of all adjustment commands during the same period (e.g., 0.3), and obtain the actual wear correlation value (2.2 mm / s²). Based on historical data in the updated progressive fault diagnosis model, establish a correspondence between "actual wear correlation value - wear level": 0-1.5 mm / s² is State 1 (no significant wear); 1.5-3 mm / s² is State 2 (slight wear); 3 mm / s² is State 3 (moderate wear). For example, the actual value of 2.2 mm / s² is State 2. Calculate the actual wear correlation value growth rate over the past 7 days (e.g., an increase of 0.3 mm / s² per day), and combine it with the current value (2.2) to predict the time to reach level 2 (3 mm / s²): (3 - 2.2) / 0.3 ≈ 2.7 days, that is, it may enter the moderate wear stage in about 3 days. Generate a predictive diagnostic result containing "current wear level (state 2), predicted level for the next 3 days (state 3), and recommended maintenance time (within 3 days)".

[0132] Step 603: Integrate the above analysis results into an intuitive report, the specific content and format of which are as follows:

[0133] The report displays the Favg and Fpeak values ​​for key time points over the past 7 days (e.g., 0:00 each day, when the instruction is triggered) in tabular form, with instruction markers (e.g., "08:30-Instruction-5%"). It also charts the changes in Favg within one hour after the instruction is triggered (e.g., Gavg drops from 0.002 to 0.001 after the instruction, indicating effective intervention). The report calculates the average compensation coefficient under different adjustment amplitudes (e.g., K average 0.0007 at 5% amplitude, K average 0.0012 at 8% amplitude), illustrating the relationship between intervention intensity and effect. It presents the current level, future trend curves (marking the time point for reaching level 2), and recommended maintenance measures (e.g., "Stop and replace joint bearings within 3 days") in a text + chart format. The report is generated in PDF format, named according to "Equipment Identifier-Date" (e.g., "Welding Robot-A3 Workshop-012-20240908.pdf"), synchronously stored in the "Diagnostic Report" folder of the virtual unit, and pushed to the equipment management platform.

[0134] The feature evolution sequence visually displays the continuous changes of vibration characteristics (amplitude, dominant frequency) over time, making equipment degradation trends (such as a slow increase in amplitude and a shift in dominant frequency towards higher frequencies) readily apparent, facilitating the rapid identification of early signs of progressive failures. The time-stamped feature sequence clearly marks the trigger points of adjustment commands, allowing direct observation of the command's impact on vibration characteristics (such as a slowdown in amplitude growth after the command), providing a basis for evaluating intervention effectiveness. Through gradient calculation, cumulative offset analysis, and compensation coefficient correction, vibration characteristics are quantitatively correlated with wear status, accurately determining the current wear level and predicting future degradation time, providing a precise time window for predictive maintenance and reducing unplanned downtime. The report integrates feature sequences, correlation analysis, and prediction results, offering comprehensive and intuitive content, improving equipment management efficiency.

[0135] like Figure 2 As shown, embodiments of the present invention also provide a machine learning system, comprising:

[0136] The acquisition module is used to acquire real-time sensor data streams of industrial equipment at the edge computing node. The data streams include the end temperature value of the welding robot, the current value of the drive motor, and the vibration frequency value of the robotic arm.

[0137] The threshold generation module is used to extract the feature values ​​of the starting point, midpoint and ending point of the welding trajectory based on the real-time sensor data stream through the edge intelligent processing unit, and dynamically generate anomaly judgment thresholds based on the feature values.

[0138] The control module is used to trigger a power cut-off command when the current value of the drive motor exceeds the threshold, or to generate a welding speed adjustment command when the vibration frequency value of the robotic arm deviates from the normal range, based on an anomaly detection threshold.

[0139] The storage module is used to aggregate sensor data streams, anomaly detection thresholds, and welding speed adjustment commands into time-series-based virtual units for storage, based on the device entity.

[0140] The update module is used to construct a progressive fault diagnosis model and update the decision parameters of the progressive fault diagnosis model based on the historical vibration frequency value sequence and corresponding control command records in the simulated unit, so as to obtain the updated progressive fault diagnosis model.

[0141] The report output module is used to output a report on the evolution sequence of robot arm vibration characteristics based on the updated progressive fault diagnosis model.

[0142] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0143] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0144] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0145] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A machine learning method, characterized in that, The method includes: Step 1: Collect real-time sensor data streams from industrial equipment at the edge computing node. The data streams include the end temperature value of the welding robot, the current value of the drive motor, and the vibration frequency value of the robotic arm. Step 2: Based on the real-time sensor data stream, the feature values ​​of the starting point, midpoint, and ending point in the welding trajectory are extracted by the edge intelligent processing unit, and an anomaly judgment threshold is dynamically generated based on the feature values. Step 3: Based on the abnormal judgment threshold, when the current value of the drive motor exceeds the threshold, a power cut-off command is triggered, or when the vibration frequency value of the robotic arm deviates from the normal range, a welding speed adjustment command is generated. Step 4: Aggregate the sensor data stream, anomaly detection threshold, and welding speed adjustment command into time-series simulated units according to the equipment entity and store them. Step 5: Based on the historical vibration frequency value sequence and corresponding control command records in the simulated unit, construct an initial progressive fault diagnosis model and update the decision parameters of the initial progressive fault diagnosis model to obtain an updated progressive fault diagnosis model. Step 6: Based on the updated progressive fault diagnosis model, output a report on the evolution sequence of the robot arm's vibration characteristics.

2. The machine learning method according to claim 1, characterized in that, Step 2 includes: Based on real-time sensor data stream, the temperature change rate characteristics of the starting point of the welding trajectory, the current distribution statistics of the midpoint of the welding path, and the vibration frequency domain envelope characteristics of the welding termination point are extracted. Temperature change rate features are combined with real-time welding speed parameters for reverse weight adjustment to generate temperature change weighted features; current distribution statistical features are combined with real-time welding material thickness parameters for positive weight adjustment to generate current distribution weighted features; vibration frequency domain envelope features are combined with equipment cumulative working time parameters for positive weight adjustment to generate vibration frequency domain weighted features. Anomaly detection thresholds are dynamically generated by integrating weighted features of temperature change, current distribution, and vibration frequency domain.

3. The machine learning method according to claim 2, characterized in that, Step 3 includes: The system compares the current value of the drive motor with the abnormal judgment threshold in real time. When the current value continuously exceeds the abnormal judgment threshold within a continuous sampling period, a power cut-off command is triggered. The system calculates the envelope offset area between the vibration spectrum and the abnormal judgment threshold based on the vibration frequency value of the robotic arm. When the envelope offset area exceeds the preset tolerance threshold, a welding speed adjustment command is generated, and the adjustment range of the welding speed adjustment command increases with the increase of the envelope offset area.

4. The machine learning method according to claim 3, characterized in that, Step 5 includes: An initial progressive fault diagnosis model is constructed using the historical vibration frequency value sequence and corresponding welding speed adjustment command record in the time-seriesd simulated unit. The latest vibration frequency value sequence and welding speed adjustment command record of the latest acquisition cycle are obtained. The latest data is input into the initial progressive fault diagnosis model, the vibration characteristic decision parameters are updated, and an updated progressive fault diagnosis model is generated.

5. The machine learning method according to claim 4, characterized in that, Obtain the latest vibration frequency value sequence and welding speed adjustment command record from the latest acquisition cycle. Input the latest data into the initial progressive fault diagnosis model, update the vibration characteristic decision parameters, and generate an updated progressive fault diagnosis model, including: Obtain the spectral envelope feature values ​​of the latest vibration frequency value sequence; The spectral envelope feature values ​​are compared with the historical spectral benchmark in the initial progressive fault diagnosis model to generate a feature offset data sequence. Process the feature offset data sequence, identify sampling periods that continuously exceed the offset threshold, and generate an over-limit period marker sequence; Based on the duration of the out-of-limit period marker sequence, the vibration characteristic weight coefficients are adjusted to generate preliminary corrected weight coefficients; The number of interventions in the welding speed adjustment command record is extracted, and the initial correction weight coefficient is attenuated based on the intervention frequency to generate updated vibration characteristic decision parameters. The updated vibration characteristic decision parameters are written into the initial progressive fault diagnosis model to generate an updated progressive fault diagnosis model.

6. The machine learning method according to claim 5, characterized in that, Step 6 includes: The historical vibration frequency value sequence is processed using an updated progressive fault diagnosis model, and the vibration spectrum envelope feature values ​​of consecutive timestamps are extracted to generate a feature evolution sequence sorted by acquisition time. Locate the trigger time node of the welding speed adjustment command in the feature evolution sequence and generate a time-stamped feature sequence; Analyze the spatiotemporal correlation between the degree of feature value shift in the time-stamped feature sequence and the adjustment command to generate predictive diagnostic results of the wear state of the robotic arm bearing; The output includes a robotic arm vibration diagnosis report containing time-stamped feature sequences, correlation analysis results, and wear prediction results.

7. The machine learning method according to claim 6, characterized in that, Analyzing the spatiotemporal correlation between the eigenvalue shift in the time-stamped feature sequence and the adjustment commands generates predictive diagnostic results for the wear state of the robotic arm bearings, including: Based on the time-stamped feature sequence, the gradient of feature value changes between adjacent time nodes is calculated to generate a feature gradient sequence. Perform time window integration on the feature gradient sequence to generate cumulative offset; Based on the trigger time node of the welding speed adjustment command, locate the corresponding segment in the cumulative offset and extract the average offset of the corresponding segment; The bearing wear compensation coefficient is calculated by combining the average offset value with the adjustment range of the welding speed adjustment command. Based on the cumulative offset and bearing wear compensation coefficient, predictive diagnostic results for the wear condition of the robotic arm bearing are generated.

8. A machine learning system that implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire real-time sensor data streams of industrial equipment at the edge computing node. The data streams include the end temperature value of the welding robot, the current value of the drive motor, and the vibration frequency value of the robotic arm. The threshold generation module is used to extract the feature values ​​of the starting point, midpoint and ending point of the welding trajectory based on the real-time sensor data stream through the edge intelligent processing unit, and dynamically generate anomaly judgment thresholds based on the feature values. The control module is used to trigger a power cut-off command when the current value of the drive motor exceeds the threshold, or to generate a welding speed adjustment command when the vibration frequency value of the robotic arm deviates from the normal range, based on an anomaly detection threshold. The storage module is used to aggregate sensor data streams, anomaly detection thresholds, and welding speed adjustment commands into time-series-based virtual units for storage, based on the device entity. The update module is used to construct an initial progressive fault diagnosis model and update the decision parameters of the initial progressive fault diagnosis model based on the historical vibration frequency value sequence and corresponding control command records in the simulated unit, so as to obtain an updated progressive fault diagnosis model. The report output module is used to output a report on the evolution sequence of robot arm vibration characteristics based on the updated progressive fault diagnosis model.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images