Rotor punching and stacking welding defect identification system based on edge calculation

Abstract

The invention discloses a rotor punching and stacking welding defect identification system based on edge calculation, and relates to the technical field of rotor welding identification, and the rotor punching and stacking welding defect identification system comprises a data acquisition module which is used for carrying out real-time acquisition on smoke dust data generated in the welding process, and the smoke dust data comprises smoke concentration, suspended particulate matter concentration and smoke infrared thermal imaging temperature; the model matching module is used for comparing the smoke concentration, the suspended particulate matter concentration and the smoke infrared thermal imaging temperature with corresponding preset reference models respectively to obtain abnormal data with an abnormal trend; and the judgment module is used for performing time sequence analysis on the abnormal data, calculating a standard deviation and a mean deviation rate based on a sliding window, and generating a corresponding early warning growth value. According to the method, a dynamic alarm value generation and attenuation mechanism is constructed, so that false alarm caused by continuous influence of historical data is avoided, and the real-time response accuracy of the integrated value is improved. And through a pre-judgment mechanism, it is ensured that alarm value attenuation is executed only on the premise that data are stable or no early warning growth trend exists.

Description

Technical Field

[0001] The present invention relates to the technical field of rotor welding identification, and specifically to a rotor lap welding defect identification system based on edge computing. Background Art

[0002] In the manufacturing process of high-precision equipment such as industrial motors and generators, the welding quality of rotor laminations directly affects the performance and life of the entire machine. During the welding process, defects such as cold welds, lack of fusion, burn-through, and misaligned welds often occur due to factors such as uneven heat input, material offset, or operating errors. In recent years, with the development of the Industrial Internet of Things and edge computing technologies, an increasing number of intelligent terminal devices have been deployed at manufacturing sites, enabling localized processing and real-time analysis of welding process parameters. In rotor lap welding, the smoke concentration, particulate matter concentration, and infrared thermal imaging characteristics generated during welding have been shown to be highly correlated with welding quality. Therefore, building a defect recognition system based on real-time edge computing capabilities, combined with multi-source smoke data for dynamic modeling and risk assessment, has become an important direction for improving welding quality control. Existing technologies have made some progress in multimodal perception and data fusion, but many challenges remain in rotor punch-over welding scenarios. For example, different types of welding anomalies often evolve and interact with each other, and relying solely on single-point threshold judgment may not cover the entire defect path. At the same time, how to realize efficient data fluctuation analysis, dynamic adjustment of warning values ​​and response mechanism on edge devices, and how to combine post-weld visual inspection results to perform model closed-loop optimization are still key technical issues and breakthrough directions in current research. Therefore, the present invention proposes a rotor punching and overlapping welding defect identification system based on edge computing. Summary of the Invention

[0003] The purpose of the present invention is to provide a rotor lap welding defect identification system based on edge computing to solve the problems mentioned in the above background technology.

[0004] The present invention can be implemented through the following technical solutions: a rotor overlap welding defect identification system based on edge computing, including a data acquisition module, a model matching module, a judgment module, an alert update module, a statistics module and an early warning module; The data acquisition module is used to collect different types of smoke data in real time during the rotor punching and welding process, including smoke concentration, suspended particulate matter concentration and smoke infrared thermal imaging temperature; The data acquisition module packages and transmits three types of data to the model matching module, including smoke concentration sequence, particulate matter concentration sequence and thermal imaging area sequence; The model matching module compares the smoke concentration data, suspended particulate matter concentration data and smoke infrared thermal imaging temperature obtained by the acquisition module with the standard reference model preset in the system to determine whether an abnormal trend has entered; The judgment module further performs time series fluctuation analysis on the data that has been judged to be abnormal to identify whether there is a trend of drastic changes within a preset time window. Specifically, the following steps are included: Sliding window processing: Set the analysis window for each type of data to the most recent 5 seconds of data, take all data points within the window, and calculate the standard deviation of the sequence within the window, which is used as a fluctuation range indicator; Fluctuation amplitude calculation: Use the standard deviation calculation method to analyze the fluctuation amplitude of the window data. If the amplitude exceeds the corresponding fluctuation threshold (learned from historical data), it is considered that the data currently has volatility risk; Deviation comparison: Compare the mean value in the current window with the normal mean value of the data type. If the difference is greater than the set relative deviation rate (such as 10%), the risk assessment of the current data type will be further aggravated. Specifically, deviation rate = |current window mean - historical mean| / historical mean; Early warning growth value generation: Based on the weighted result of the fluctuation range and the degree of deviation, an early warning growth value is generated. The early warning growth value = fluctuation range × α + deviation rate × β, where α and β are the set weight factors; The warning update module generates corresponding dynamic warning values ​​according to the warning growth values ​​of various types of data, and updates the current values ​​of various dynamic warning values ​​in combination with the time decay mechanism; The statistical module performs weighted statistical integration on each dynamic alert value according to the corresponding weight to obtain an integrated value, which is used to determine the current overall risk level of the system and decide whether to enter the alarm state. Specifically, it includes: Weighted summation: Assign weights to each type of data based on its importance in the overall judgment, and calculate the integrated value after weighting; Threshold judgment: Set the total warning threshold. When the integrated value exceeds the total warning threshold, it is judged as "significant welding abnormality trend"; The early warning module is used to respond to the alarm signal.

[0005] A further technical improvement of the present invention is that the time decay mechanism includes: S1. Before the dynamic warning value is attenuated, a double pre-judgment is performed first: s11. Set a safety threshold for the total warning threshold and determine whether the current integrated value exceeds the safety threshold. If the integrated value has not yet dropped below the safety threshold, delay the execution of the dynamic warning value decay for this type of data. s12. The system reviews the integrated value change rate and fluctuation trend of this type of data within the past set time window. If an upward or unstable trend is identified, no decay is performed; S2. When the system identifies a certain type of data as entering a decayable state, it adjusts the weight of that type of data in calculating the integrated value, so that its impact on the overall warning judgment result is gradually weakened; S3. Assume that the dynamic warning value of the d-th type of data at time t is , then the current integrated value is defined as: Where, Indicates the weight coefficient of the corresponding type of data in the current integrated value calculation, which is used to measure its impact on the overall risk value of the system; S4. When the preset attenuation trigger condition is met, Performs exponential decay adjustment.

[0006] A further technical improvement of the present invention is that if the dynamic warning value of a certain type of data meets the attenuation trigger condition, but the system detects that any other type of data still has an early warning growth value input within the same time window, or its dynamic warning value is in a continuous growth state, then it is determined that there is a co-evolution trend; The system temporarily suspends weight decay for this type of data, retaining its risk influence in the integrated value until the remaining types of data also enter a stable decay state, and then uniformly performs weight reduction processing.

[0007] A further technical improvement of the present invention is that the statistical module is provided with a plurality of stage-by-stage integration value critical thresholds. When the integration value reaches different critical thresholds, the system dynamically increases the collection frequency of various types of smoke data and adopts a higher cumulative weight when updating the dynamic warning value to accelerate the growth rate of the overall integration value of the system, thereby improving the response speed to abnormal conditions.

[0008] A further technical improvement of the present invention is that the system further includes a defect evolution module, which maps the fluctuation trends of smoke concentration, particle concentration and thermal imaging to different stages of the defect formation process by establishing a time-series evolution path diagram of welding defect behavior, so as to identify in advance whether it is in a defect evolution state such as cold welding, incomplete fusion or burn-through, and predict the trend based on the change rate and duration.

[0009] A further technical improvement of the present invention is that the method for trend judgment comprises the following steps: A1. Build a reference model for defect evolution paths: Based on historical welding defect data, the evolution stages corresponding to the defect types are marked, each evolution stage is mapped as a path node, and the defect evolution direction is used as a connecting line to form a complete defect evolution path diagram; A2. Real-time data trend comparison and path matching: Perform real-time sliding window processing on all types of collected data to extract dynamic features within the current time period, including average change trend, fluctuation amplitude, continuous deviation time, and rise / fall rate; Compare the extracted trend features with the established feature intervals of each evolution stage in the evolution path diagram to determine which defect stage node the current node falls into; A3. Dynamically output the defect stage judgment results and feedback the risk level: If the match is successful, the evolution stage to which the current state belongs is marked.

[0010] A further technical improvement of the present invention is that after welding is completed, the system determines whether the welding part is qualified through visual inspection, and on the premise of confirming that it is qualified, traces the integrated value fluctuation sequence during the welding process, dynamically corrects the standard reference model of various types of smoke data according to the integrated value fluctuation range of the sequence, and dynamically adjusts the total warning threshold within the allowable fluctuation range to improve the system's adaptability to normal fluctuations and reduce the false alarm rate.

[0011] A further technical improvement of the present invention is that the method for dynamically adjusting the total warning threshold by the system comprises the following steps: Y1. After the welding process is completed, the system collects and identifies images of the welding seam area; Y2. When the weld seam area is judged to be qualified, the system retrieves the integrated value fluctuation sequence within the welding cycle, recorded as S(t), where t represents the time series; Y3. The integrated value sequence S(t) in the welding cycle currently marked as qualified and the integrated value sequence that has been judged qualified in the past are combined into a sample set to perform qualified fluctuation range modeling, including: The sliding window statistical method is used to extract the average value and maximum fluctuation range of the integrated value in each sample; The allowed fluctuation range of the integrated value to form a qualified sample is: ;in, is the mean, is the deviation tolerance range; If several consecutive qualified integrated value sequences S(t) fall into this interval, the standard reference model is updated; Y4. Based on the fluctuation range modeling results in Y3, the system performs the following operations on the standard reference model and the total warning threshold: Standard reference model: shifts the baseline change model of the currently collected data to the mean center of the qualified samples; Total warning threshold: within the qualified fluctuation range, set the dynamic offset factor ε to increase the total warning threshold Temporarily raised to .

[0012] A further technical improvement of the present invention is that: if the weld seam area is judged as qualified for n consecutive times (n is a positive integer greater than 1) and the fluctuation amplitude of the integrated value sequence S(t) is consistent, the standard reference model and the overall warning threshold are updated; At the same time, the system retains the original standard reference model and the original total warning threshold as a fallback mechanism to prevent missed detections due to model misadaptation.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention avoids false alarms caused by the continuous influence of historical data by building a dynamic warning value generation and decay mechanism, improving the real-time response accuracy of the integrated value. Through the pre-judgment mechanism, it ensures that the warning value decay is only executed when the data is stable or there is no early warning growth trend, effectively enhancing the system's judgment stability of edge states. This invention utilizes multi-source data fusion integration values ​​and introduces a hierarchical response mechanism. It dynamically adjusts the collection frequency and weight parameters of various types of data according to the critical thresholds reached by the integration values, thereby improving the ability to capture early signs of abnormalities and ensuring that the system can timely improve its perception and judgment strength when entering different early warning stages. The present invention further combines visual inspection feedback of welding results to model the fluctuation range of the integrated value, and dynamically corrects the standard reference model and the total warning threshold, so that the system has good on-site adaptability, reduces the possibility of false alarms and missed alarms, and realizes the linkage suppression and synchronous attenuation of risk signals between different data types through the collaborative evolution trend recognition mechanism, which significantly improves the system's comprehensive analysis capability of multi-dimensional abnormal evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0015] Figure 1 is a system block diagram of the present invention; Figure 2 It is a logic flow chart of the present invention. DETAILED DESCRIPTION

[0016] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0017] Example 1: Please refer to Figure 1-2As shown, the present invention provides a rotor lap welding defect identification system based on edge computing. This system is built on an edge computing architecture. The edge computing unit is deployed at the welding site and is physically connected to the data acquisition sensor. It has functions such as local data preprocessing, model matching, volatility judgment, warning value update, and alarm output, thereby realizing on-site intelligent identification and risk response of the welding process. There is no need to rely on remote servers for real-time calculations, which can significantly reduce data transmission delays and improve the timeliness of anomaly detection.

[0018] Specifically, it includes data acquisition module, model matching module, judgment module, alert update module, statistics module and early warning module; The data acquisition module is used to collect different types of smoke data from multiple sources in real time during the rotor punching and welding process, including: Smoke density data: Using a laser scattering smoke sensor, the concentration value is collected every 100 milliseconds and output as single-channel time series data. Ten sets of sampling values ​​are aggregated per second to form a smoke concentration sequence within 1 second. , where each C is a sampling value; Suspended particulate matter concentration data: A PM particle size counting sensor is used to collect data every 200 milliseconds, recording the mass concentration of particles in different particle size ranges. Five sets of PM2.5 and PM10 data pairs are obtained per second to form a particle concentration sequence: ; Smoke infrared thermal imaging temperature: A mid-infrared thermal imager is used to capture the temperature distribution image of the welding area, updating 5 frames per second to obtain a sequence of thermal imaging area areas. Each pixel in the temperature distribution image corresponds to a real-time temperature value. In this embodiment, the area of ​​the area with a temperature above 450° is extracted from the temperature distribution image. The data acquisition module packages and transmits three types of data collected every second to the model matching module, including smoke concentration sequence, particulate matter concentration sequence and thermal imaging area sequence; The model matching module compares the smoke concentration data, suspended particulate matter concentration data, and smoke infrared thermal imaging temperature obtained by the acquisition module with the standard reference model preset in the system to determine whether to enter the abnormal trend analysis stage; Concentration model matching: Calculate the smoke concentration average curve over the past 5 seconds and compare it with the standard concentration trend established for historical normal welding samples. Use the Pearson correlation coefficient for similarity matching. If the similarity exceeds the preset concentration threshold, it is determined that the concentration trend is abnormal. Specifically, similarity = vector dot product / (product of two vector moduli). If the similarity is greater than 0.85, it is considered abnormal. Particle model matching: Count the growth rates of PM2.5 and PM10 in the current time period. If the increase in the concentration of either PM2.5 or PM10 particle size within 1 second, that is, the difference between the current second and the previous second exceeds the preset particle concentration threshold, it is determined that the particle trend is abnormal; Thermal image model matching: Calculate the high temperature area in each frame of thermal image The area change of the high temperature area includes: Image frame input: Convert each frame of thermal image into a two-dimensional temperature matrix T(x, y): each pixel point (x, y) corresponds to a temperature value in °C; Setting temperature thresholds , as a benchmark value for identifying “high temperature areas”; Extract the pixel set of high temperature area A: A={(x,y)|T(x,y)≥ }, that is, only retain those pixels whose temperature is greater than or equal to the threshold, and obtain their number N; Calculate the area of ​​the high temperature area by the formula , obtain high temperature area , where S is the physical area represented by a single pixel, which is calculated based on the focal length of the thermal imager, the size of the imaging surface, and the fixed distance from the welding area to the thermal imager through geometric calibration; If the high temperature area Continuously expands in three consecutive frames of images, or exceeds the preset extreme temperature threshold If there is a hot spot, it is judged that the heating state is abnormal; Right now (t)> (t-1)> (t-2), or T(x, y) ≥ , it is considered as "abnormal heating state" during welding; The judgment module further performs time series fluctuation analysis on the data that has been judged to be abnormal to identify whether there is a trend of drastic changes within the preset time window. Specifically, it includes: Sliding window processing: Set the analysis window for each type of data to the most recent 5 seconds of data, take all data points within the window, and calculate the standard deviation of the sequence within the window, which is used as a fluctuation range indicator; Fluctuation amplitude calculation: Use the standard deviation calculation method to analyze the fluctuation amplitude of the window data. If the amplitude exceeds the corresponding fluctuation threshold (learned from historical data), it is considered that the data currently has volatility risk; Deviation comparison: Compare the mean value in the current window with the normal mean value of the data type. If the difference is greater than the set relative deviation rate (such as 10%), the risk assessment of the current data type will be further aggravated. Specifically, deviation rate = |current window mean - historical mean| / historical mean; Early warning growth value generation: Based on the weighted result of the fluctuation range and the degree of deviation, an early warning growth value is generated. The early warning growth value = fluctuation range × α + deviation rate × β, where α and β are the set weight factors; The alert update module generates corresponding dynamic alert values ​​according to the early warning growth values ​​of various data, and updates the current values ​​of various dynamic alert values ​​in combination with the time decay mechanism. Specifically: Single growth calculation: accumulate the warning growth value entered each time; Decay mechanism: If there is no new growth input for a certain type of data within a set time interval, its current dynamic warning value will decrease exponentially to prevent old anomalies from continuously affecting the judgment results; Time decay mechanisms include: S1. Before the dynamic warning value is attenuated, a double pre-judgment is performed first: s11. Set a safety threshold for the total warning threshold and determine whether the current integrated value exceeds the safety threshold. If the integrated value has not yet dropped below the safety threshold, delay the execution of the dynamic warning value decay for this type of data. s12. The system reviews the integrated value change rate and fluctuation trend of this type of data within the past set time window. If an upward or unstable trend is identified, no decay is performed; S2. When the system identifies a certain type of data as entering a decayable state, it adjusts the weight of that type of data in calculating the integrated value, so that its impact on the overall warning judgment result is gradually weakened; S3. Assume that the dynamic warning value of the d-th type of data at time t is , then the current integrated value is defined as: Where, Indicates the weight coefficient of the corresponding type of data in the current integrated value calculation, which is used to measure its impact on the overall risk value of the system; S4. When the preset attenuation trigger condition is met, Perform an exponential decay adjustment, and the decrement calculation is as follows: Where, The weight attenuation rate coefficient preset by the system, indicating the degree of impact reduction per unit time; is the weight of this type of data at the previous moment; The time interval since the last update. The specific weight update events of each type of data are recorded as the first timestamp, and the current judgment time is the second timestamp. The difference between the second timestamp and the first timestamp. The system periodically runs this judgment mechanism to ensure that the attenuation operation has continuity and quantifiable basis in time advancement, and avoid abnormal jumps in weights due to sampling delays; By weighting The gradual adjustment of the dynamic alert value gradually reduces its impact on the calculation of the integrated value without destroying the historical continuity of the dynamic alert value. This method effectively reduces the risk of false alarms caused by short periods of no input, enhances the ability to mitigate changes in data trends, and improves the system's stability in judging boundary conditions and the interpretability of subsequent risks. If the dynamic warning value of a certain type of data meets the attenuation trigger condition, but the system detects that any other type of data still has an early warning growth value input within the same time window, or its dynamic warning value is in a continuous growth state, it is determined that there is a co-evolution trend; The system temporarily suspends weight decay for this type of data, retaining its risk influence in the integrated value until the rest of the data also enters a stable decay state, and then uniformly performs weight reduction processing; The statistical module performs weighted statistical integration on each dynamic alert value according to the corresponding weight to obtain the integrated value, which is used to determine the current overall risk level of the system and decide whether to enter the alarm state. Specifically, it includes: Weighted summation: Assign weights to each type of data based on its importance in the overall judgment, and calculate the integrated value after weighting; Threshold judgment: Set the total warning threshold. When the integrated value exceeds the total warning threshold, it is judged as "significant welding abnormality trend"; The statistics module is equipped with multiple critical thresholds for integration values ​​at different stages. When the integration value reaches different critical thresholds, the system dynamically increases the frequency of collecting various types of smoke and dust data and adopts a higher cumulative weight when updating the dynamic warning value to accelerate the growth rate of the system's overall integration value, thereby improving the response speed to abnormal conditions. Specifically, they include: Z1. Preprocessing and weighted integration calculation of dynamic warning values ​​for various types of data: The statistics module periodically obtains the dynamic warning values ​​of various data (smoke concentration, suspended particulate matter concentration, infrared thermal imaging temperature) at the current moment. The dynamic warning values ​​are derived from the cumulative update results of the warning growth value of the abnormal fluctuation trend of each type of data; The system performs a weighted superposition of the current dynamic alert values ​​of various data types based on preset weight factors (e.g., the weight of smoke concentration data is 0.3, suspended particulate matter data is 0.2, and infrared thermal imaging data is 0.5), and obtains the current integrated value to reflect the actual impact of each type of data in anomaly detection. Z2. Perform a primary comparison with the overall warning threshold: The system has a set of graded critical thresholds, such as the first-level critical threshold is , the secondary critical threshold is , the third critical threshold is ; In this embodiment, the critical threshold of each level is obtained by: By collecting the integrated value data series of historical normal welding cycles and defective welding cycles, the normal value interval and abnormal transition area are identified using clustering algorithm; Based on the integrated value growth slope of the risk acceleration segment in the anomaly evolution path and combined with manual assessment results, critical thresholds at three inflection point levels are formulated; If the current integration value does not exceed the first critical threshold , the system maintains normal detection mode and continues to operate at the default sampling frequency without adjusting the response level; Z3, when the integration value reaches the first critical threshold hour: The system automatically increases the frequency of data collection for various types of smoke and dust for the first time, including: The smoke concentration sampling period is reduced from 100ms to 80ms; The suspended particulate matter sampling period was adjusted from 200ms to 150ms; The infrared thermal imaging frame rate is increased from 5 frames per second to 7 frames per second; At the same time, the system will use a higher cumulative weight for the early warning growth value of various types of data when updating the dynamic warning value, for example: The original weight of smoke concentration type data was 1.0, which was adjusted to 1.2; The weight of infrared thermal imaging data type has been increased from 1.2 to 1.5; Z4, when the integration value reaches the secondary critical threshold hour: The system has once again increased the frequency of collecting various types of data: The smoke concentration sampling period is reduced to 50ms; Thermal imaging frame rate increased to 10 frames per second; The suspended particulate matter sampling period is adjusted to 100ms; When updating the dynamic alert values ​​of various data types, the cumulative weights of all early warning growth values ​​have been further increased. For example, the infrared thermal imaging category has been increased to 2.0, and the smoke category has been increased to 1.5. Z5, the integration value reaches the third critical threshold hour: The system enters the highest responsiveness mode: The acquisition period of all sensors reaches the minimum threshold (e.g. 30ms); The thermal image enters the continuous frame high temperature recognition state; Dynamic alert values ​​corresponding to all types of data are updated with the preset maximum cumulative weight to improve response speed; The early warning module is used to respond to alarm signals and control the execution of output behaviors, such as prompting alarm information, triggering device linkage, or uploading system records; including; Local alarm output: activate the sound and light alarm, and display the current alarm type data and value on the operation panel interface; Logging function: Automatically record alarm number, time, warning growth value composition and collected data snapshots for later traceability and model optimization.

[0019] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by those skilled in the art based on actual conditions. Example 2: Compared to Example 1, the system in Example 2 further includes a defect evolution module. This module establishes a time-series evolution path diagram of welding defect behavior and maps the fluctuation trends of smoke concentration, particle concentration, and thermal imaging to different stages of the defect formation process. This module can identify in advance whether the defect is in a cold weld, lack of fusion, or burn-through state, and predict the trend based on the change rate and duration.

[0020] The method of trend judgment includes the following steps: A1. Build a reference model for defect evolution paths: Based on a large amount of historical welding defect data, the evolution stages corresponding to the defect types are manually marked, such as: Phase A: A short-term slight increase in smoke concentration → corresponds to the signs of "cold solder joint"; Phase B: PM2.5 continues to rise, and the temperature zone at the edge of the thermal image expands → a trend of “unfused”; Stage C: High-temperature patches gather and the temperature rises rapidly in the infrared image → a sign of "burn-through"; Map the above stages into path nodes and connect them with the defect evolution direction to form a complete defect evolution path diagram; A2. Real-time data trend comparison and path matching: Perform real-time sliding window processing on all types of collected data to extract dynamic features within the current time period, including average change trend, fluctuation amplitude (standard deviation), continuous deviation time, and rise / fall rate; Compare the extracted trend features with the established feature intervals of each stage in the evolution path diagram, and use the dynamic threshold matching method to determine which defect stage node the current data falls into; A3. Dynamically output the defect stage judgment results and feedback the risk level: If the match is successful, the evolution stage to which the current state belongs is marked, and the corresponding risk level is calculated based on the severity of the stage (such as "Warning Level I" or "Warning Level II"); Example 3: Compared to Example 2, the system in Example 3 determines whether the welded part is qualified through visual inspection after welding is completed. If qualified, it traces the integrated value fluctuation sequence during the welding process, dynamically corrects the standard reference model of various types of smoke data based on the integrated value fluctuation range of the sequence, and dynamically adjusts the overall warning threshold within the allowable fluctuation range to improve the system's adaptability to normal fluctuations and reduce the false alarm rate.

[0021] The method for dynamically adjusting the total warning threshold of the system includes the following steps: Y1. After the welding process is completed, the system uses an industrial camera to collect and identify images of the weld seam area. In this embodiment, image algorithms (such as edge detection and defect recognition) are used to determine whether the weld is qualified. Specifically, any mature technology in the existing technology can be used for monitoring; Y2. When the weld seam area is judged to be qualified, the system retrieves the integrated value fluctuation sequence within the welding cycle, specifically including: Calling time period: from the start of welding to the triggering of visual inspection; Acquisition content: Dynamic warning value sequences of all types of data (smoke concentration, particulate matter concentration, infrared thermal image); Calculation method: Calculate the integrated value curve by weighting the dynamic alert value at each moment, recorded as S(t), where t represents the time series; Y3. The integrated value sequence S(t) in the welding cycle currently marked as qualified and the integrated value sequence that has been judged qualified in the past are combined into a sample set to perform qualified fluctuation range modeling, including: The sliding window statistical method is used to extract the average value and maximum fluctuation range of the integrated value in each sample; The allowed fluctuation range of the integrated value to form a qualified sample is: ;in, is the mean, is the deviation tolerance range; If several consecutive qualified integrated value sequences S(t) fall into this interval, the standard reference model is updated; Y4. Based on the fluctuation range modeling results in Y3, the system performs the following operations on the standard reference model and the total warning threshold: Standard reference model: offsets the baseline change model of the currently collected data to the mean center of qualified samples to improve the field adaptability of the model; Total warning threshold: within the qualified fluctuation range, set the dynamic offset factor ε to increase the total warning threshold Temporarily raised to , improve the system's fault tolerance under fluctuation conditions and avoid false alarms; If the weld seam area is judged as qualified for n consecutive times (n is a positive integer greater than 1) and the fluctuation amplitude of the integrated value sequence S(t) is consistent, the standard reference model and the overall warning threshold are updated; At the same time, the system retains the original standard reference model and the original total warning threshold as a fallback mechanism to prevent missed detections due to model misadaptation.

[0022] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0023] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. The rotor punching and welding defect identification system based on edge computing is characterized by: include: The data acquisition module collects smoke data generated during the welding process in real time. The smoke data includes smoke concentration, suspended particulate matter concentration and smoke infrared thermal imaging temperature. The model matching module compares the smoke concentration, suspended particulate matter concentration, and smoke infrared thermal imaging temperature with the corresponding preset reference models to obtain abnormal data with abnormal trends; The judgment module performs time series analysis on abnormal data, calculates the standard deviation and mean deviation rate based on the sliding window, and generates the corresponding warning growth value; The alert update module generates corresponding dynamic alert values ​​according to the early warning growth values ​​of various data, and updates various dynamic alert values ​​in combination with the time decay mechanism; The statistical module performs weighted statistical integration on each dynamic warning value according to the corresponding weight to obtain the integrated value and judge whether the integrated value reaches the total warning threshold; The early warning module sends a response alarm signal when the integrated value exceeds the total warning threshold.

2. The rotor punching and welding defect identification system based on edge computing according to claim 1 is characterized in that: Time decay mechanisms include: S1. Before the dynamic warning value is attenuated, a double pre-judgment is performed first: s11. Set a safety threshold for the total warning threshold and determine whether the current integrated value exceeds the safety threshold. If the integrated value has not yet dropped below the safety threshold, delay the execution of the dynamic warning value decay for this type of data. s12. The system reviews the integrated value change rate and fluctuation trend of this type of data within the past set time window. If an upward or unstable trend is identified, no decay is performed; S2. When the system identifies a certain type of data as entering a decayable state, the system adjusts the weight of this type of data when calculating the integrated value; S3. Assume that the dynamic warning value of the d-th type of data at time t is , then the current integrated value is defined as: Where, Indicates the weight coefficient of the corresponding type of data in the current integration value calculation; S4. When the preset attenuation trigger condition is met, Performs exponential decay adjustment.

3. The rotor punching and welding defect identification system based on edge computing according to claim 2 is characterized in that: If the dynamic warning value of a certain type of data meets the attenuation trigger condition, but the system detects that any other type of data still has an early warning growth value input within the same time window, or its dynamic warning value is in a continuous growth state, it is determined that there is a co-evolution trend; The system temporarily suspends weight decay for this type of data, retaining its risk influence in the integrated value until the remaining types of data also enter a stable decay state, and then uniformly performs weight reduction processing.

4. The rotor punching and welding defect identification system based on edge computing according to claim 1 is characterized in that: The statistical module is provided with multiple stage-by-stage integration value critical thresholds. When the integration value reaches different critical thresholds, the system dynamically increases the collection frequency of various types of smoke data and adopts a higher cumulative weight when the dynamic warning value is updated.

5. The rotor punching and welding defect identification system based on edge computing according to claim 4 is characterized in that: The system includes a defect evolution module, which establishes a time-series evolution path diagram of welding defect behavior, maps the fluctuation trends of smoke concentration, particle concentration and thermal imaging to different stages of the defect formation process, and predicts trends based on the rate of change and duration.

6. The rotor punching and welding defect identification system based on edge computing according to claim 5 is characterized in that: The method of trend judgment includes the following steps: A1. Build a reference model for defect evolution paths: Based on historical welding defect data, the evolution stages corresponding to the defect types are marked, each evolution stage is mapped as a path node, and the defect evolution direction is used as a connecting line to form a complete defect evolution path diagram; A2. Real-time data trend comparison and path matching: Perform real-time sliding window processing on all types of collected data to extract dynamic features within the current time period, including average change trend, fluctuation amplitude, continuous deviation time, and rise / fall rate; Compare the extracted trend features with the established feature intervals of each evolution stage in the evolution path diagram to determine which defect stage node the current node falls into; A3. Dynamically output the defect stage judgment results and feedback the risk level: If the match is successful, the evolution stage to which the current state belongs is marked.

7. The rotor punching and welding defect identification system based on edge computing according to claim 1 is characterized in that: After welding is completed, the system uses visual inspection to determine whether the welded part is qualified. On the premise of confirming that it is qualified, it traces the integrated value fluctuation sequence of the welding process, dynamically corrects the standard reference model of various types of smoke data according to the integrated value fluctuation range of the sequence, and dynamically adjusts the total warning threshold within the allowable fluctuation range.

8. The rotor punching and welding defect identification system based on edge computing according to claim 7 is characterized in that: The method for dynamically adjusting the total warning threshold of the system includes the following steps: Y1. After the welding process is completed, the system collects and identifies images of the welding seam area; Y2. When the weld seam area is judged to be qualified, the system retrieves the integrated value fluctuation sequence within the welding cycle, recorded as S(t), where t represents the time series; Y3. The integrated value sequence S(t) in the welding cycle currently marked as qualified and the integrated value sequence that has been judged qualified in the past are combined into a sample set to perform qualified fluctuation range modeling, including: The sliding window statistical method is used to extract the average value and maximum fluctuation range of the integrated value in each sample; The allowed fluctuation range of the integrated value to form a qualified sample is: ;in, is the mean, is the deviation tolerance range; If several consecutive qualified integrated value sequences S(t) fall into this interval, the standard reference model is updated; Y4. Based on the fluctuation range modeling results in Y3, the system performs the following operations on the standard reference model and the total warning threshold: Standard reference model: shifts the baseline change model of the currently collected data to the mean center of the qualified samples; Total warning threshold: within the qualified fluctuation range, set the dynamic offset factor ε to increase the total warning threshold Temporarily raised to .

Images