A method for testing the strength of a bolted joint

Abstract

The application discloses a bolt fastener connection strength test method and relates to the technical field of fastening mechanics analysis. The bolt fastener connection strength test method comprises the following steps: collecting bolt data through an internal sensor in an electric wrench, analyzing and classifying the bolt data, transmitting the bolt data to a relay station, obtaining corresponding response bolt data, then uploading the response bolt data to a cloud platform according to network conditions after the response bolt data is sorted according to priorities, prompting operation through the cloud platform, and feeding back to the relay station. The bolt fastener connection strength test accuracy is improved, and the problem of low bolt fastener connection strength test accuracy caused by discontinuous signals in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fastening mechanics analysis, in particular to a bolt fastener connection strength test method. BACKGROUND

[0002] The intelligent electric wrench integrates sensors for detecting construction torque, angle and other data, and meets the needs of tightening and disassembling bolts of different specifications. At the same time, the intelligent electric wrench automatically records the tightening data (torque, number of turns, time) through the microprocessor and storage chip, becoming an important tool in modern engineering operation process.

[0003] Through the intelligent electric wrench, not only the basic functions that can be realized by conventional wrenches can be realized, but also based on the data recorded during the wrench construction process, analysis can be carried out, and based on the analysis, construction reminders can be carried out, such as reminding the quality and results of tightening and disassembling and making on-site work reminders and judgments, and further statistical analysis can be carried out based on the recorded data, reflecting the quality of the bolts and the aging degree of the bolts, and based on the data classification and statistics of different regions or different wrench individuals, the working process state and engineering quality state of different regions or different wrench individuals can also be reflected. The data in the wrench working process can also be transmitted and displayed.

[0004] For example, the Chinese utility model patent with publication number CN211073368U discloses a fixed-torque electric wrench, which can provide torque monitoring and recording functions and data transmission functions for the electric wrench, and records the tightening process of each group of bolts and nuts on the iron tower in a digital manner. The quality of bolt tightening is monitored in a timely and effective manner, and the construction quality and work efficiency are improved.

[0005] For example, the Chinese invention patent application with publication number CN118990381A discloses a steel sleeve electric wrench, an intelligent construction system and a construction method, which detects the size of the working current through the motor control module, measures the actual tightening torque value of the electric wrench by cooperating with the MCU controller, and realizes numerical display through the display screen; after the electric wrench is connected to the network, the MCU controller can transmit the data to the cloud storage for the user to view.

[0006] However, the above-mentioned prior art at least has the following technical problems:

[0007] When heavy electric wrenches are used for frequent operations, the tedious repetitive actions can cause the process of the operation to be prone to negligence and incomplete work, such as problems of incomplete fastening, incorrect use of bolt types, and incorrect bolt positioning, and the like. An intelligent electric wrench can also integrate various functions, such as camera functions, data acquisition, data calculation and storage functions, and the like, which can be even heavier and more difficult to use. However, the intelligent electric wrench has its irreplaceable data acquisition and analysis functions to improve the engineering quality. The existing intelligent wrench usually displays the data after analysis on its display screen, and can also transmit the data. In this process, most of the data is calculated and analyzed by the hardware of the wrench itself, so that the wrench cannot handle complex data types due to the influence of the size and weight of the wrench, or the hardware with complex data processing capability is configured, but the overall wrench becomes heavy.

[0008] In addition, the intelligent wrench in the prior art cannot handle complex event data, such as bolt connection strength and quality analysis, overall structure strength analysis and risk analysis through data obtained by the wrench of adjacent points or associated mechanical components.

[0009] In addition, the intelligent wrench in the prior art only stores and monitors the detected data in the cloud, and it is difficult to fully utilize the cloud power for big data analysis in real time, especially in a poor network operation environment, the data transmission will be greatly limited and affected, so that the cloud monitoring construction process data, analyzing the operation process quality and the bolt connection strength and quality, the communication is poor, it is difficult to play the function and advantage of the cloud. SUMMARY

[0010] In order to solve the technical problem of low bolt fastener connection strength test accuracy caused by signal discontinuity in the prior art, the embodiments of the present application provide a bolt fastener connection strength test method. The technical scheme is as follows:

[0011] The bolt data reflecting the bolt fastener connection process and state and the bolt connection strength in the action process is collected in real time by the sensor of the electric wrench; the bolt data is analyzed and each action is classified into one of the on-site response type event judged and fed back by the electric wrench in real time, the delayed evaluation type event relying on the background in-depth calculation and analysis, and the semi-response type event discovered by the electric wrench but needing the background to assist in confirming and processing; if one action meets the conditions of multiple types of events at the same time, the on-site response type event judged and fed back by the electric wrench in real time is given priority, and the rest is determined as temporary failure; the data in S1 is transmitted to the relay station; the information transmitted back by the relay station is received and an operation prompt is given, the information is the feedback information processed by the relay station according to the data uploaded by the electric wrench and / or the cloud platform according to the data uploaded by the relay station, and the feedback information is used to reflect the strength risk and operation risk of the bolt fastener individual and / or related structure; the specific acquisition process of the feedback information processed by the relay station according to the data uploaded by the electric wrench is that the relay station pre-processes the data uploaded by the electric wrench, and the pre-processing includes preliminary feedback response to the delayed evaluation type event and the semi-response type event, and the preliminary feedback response means that the relay station processes the events with processing capability and feeds back the feedback information to the electric wrench end; the specific acquisition process of the feedback information processed by the cloud platform according to the data uploaded by the relay station is that the relay station evaluates and sorts the priority of the data to be uploaded; the data to be uploaded is uploaded to the cloud platform in real time according to the network state according to the priority order; the cloud platform receives the data uploaded by the relay station to obtain remote feedback information and transmits it to the relay station; the relay station sends the feedback information to the electric wrench end according to the remote feedback information.

[0012] The embodiment of the application provides a bolt fastener connection strength test method, which is used for a relay end and includes the following steps: receiving data uploaded by an electric wrench; the data is bolt data that is analyzed and classified after the electric wrench collects bolt data reflecting a bolt fastener connection process and state and bolt connection strength in a motion process in real time through a sensor; the classification is one of the following: a field response type event that is judged and fed back by the electric wrench in real time, a delayed evaluation type event that depends on deep calculation and analysis of a background, and a semi-response type event that is found by the electric wrench to be abnormal and needs to be confirmed and processed by the background; if a motion simultaneously meets multiple event conditions, the field response type event that is judged and fed back by the electric wrench in real time is given priority, and the rest is determined to be temporarily invalid; the data in A1 is preprocessed; the preprocessing includes preliminary feedback response of the delayed evaluation type event and the semi-response type event, the preliminary feedback response indicates that the relay station processes events with processing capability and feeds back feedback information to the electric wrench end; priority of data to be uploaded is evaluated and sorted; the data to be uploaded is dynamically uploaded to a cloud platform in real time according to a network state according to the priority order; the relay station sends feedback information to the electric wrench end according to remote feedback information; the feedback information is information that is obtained by the cloud platform from dynamic analysis of the data uploaded by the relay station and transmitted to the relay station.

[0013] The embodiment of the application provides a bolt fastener connection strength test method, which is used for a relay end and includes the following steps: receiving data uploaded by an electric wrench; the data is bolt data that is analyzed and classified after the electric wrench collects bolt data reflecting a bolt fastener connection process and state and bolt connection strength in a motion process in real time through a sensor; the classification is one of the following: a field response type event that is judged and fed back by the electric wrench in real time, a delayed evaluation type event that depends on deep calculation and analysis of a background, and a semi-response type event that is found by the electric wrench to be abnormal and needs to be confirmed and processed by the background; if a motion simultaneously meets multiple event conditions, the field response type event that is judged and fed back by the electric wrench in real time is given priority, and the rest is determined to be temporarily invalid; the data in A1 is preprocessed; the preprocessing includes preliminary feedback response of the delayed evaluation type event and the semi-response type event, the preliminary feedback response indicates that the relay station processes events with processing capability and feeds back feedback information to the electric wrench end; priority of data to be uploaded is evaluated and sorted; the data to be uploaded is dynamically uploaded to a cloud platform in real time according to a network state according to the priority order; the relay station sends feedback information to the electric wrench end according to remote feedback information; the feedback information is information that is obtained by the cloud platform from dynamic analysis of the data uploaded by the relay station and transmitted to the relay station.

[0014] The technical scheme provided by the embodiment of the application has at least the following beneficial effects:

[0015] Through the sensor of the electric wrench, the data in the working process is collected in real time, the hardware of the wrench body can directly process and timely feedback the simple and easy-to-handle data, real-time guidance is provided for the on-site work, preventing the lag caused by big data calculation and the inconvenience caused by the configuration of heavy hardware devices, in addition, the relay station is set to preprocess the problems that the wrench body cannot solve, solving most of the conditions that can be solved on site, in addition, the relay station placed on the construction site can communicate with the wrench in real time without obstruction, and timely process the lightweight computing problems that the wrench body cannot solve, compared with the prior art, the data needs to be transmitted to the cloud for monitoring and subsequent data processing, the wrench involved in the scheme has the advantages of being light and having strong data processing capability, the data in the bolt working process can be processed in time, and relevant strength and quality data analysis results are obtained, to quickly and accurately guide the on-site work, most common abnormalities can be corrected on site in seconds, significantly reducing the number of rework and reinspection, in addition, for abnormal data that is not common, further cloud analysis and feedback can be performed, the cloud accurately analyzes the data in detail, and does not cause data omission and local feedback lag caused by local data pressure.

[0016] Through the relay station, the priority of the data to be uploaded is evaluated, the uploaded data can be accurately sorted by emergency level, and the data with high emergency level is uploaded in priority, so that the emergency can be quickly processed, compared with the prior art, centralized uploading or sequential uploading will cause a large amount of normal data to cause data congestion and abnormal data to be unable to be uploaded in time, which is not conducive to real-time analysis of the cloud and timely prevention and solution of emergency.

[0017] Through the real-time dynamic uploading to the cloud platform according to the network state, the network state can be analyzed and evaluated in real time, the data uploading mode is selected in real time according to the network state, large blocks and multiple channels are selected for concurrent transmission in good network, and small blocks and few channels are selected for concurrent transmission in poor network, and the transmission is automatically adjusted dynamically, which can ensure that the bandwidth is not wasted and can prevent data transmission failure from causing data congestion, in addition, the transmission with priority can also realize that the emergency data is transmitted quickly in priority, even if the network is not good, the data that needs to be analyzed or uploaded in emergency can also be uploaded in time, thereby realizing the timely feedback of the emergency data, improving the quality and safety of the work, the cloud platform analyzes the data and corresponding models, and can analyze the events that cannot be solved or accurately processed by the wrench and the relay station, for example, local correlation data comprehensive analysis, connection strength of a batch or local position, and abnormal risk results of local connection strength obtained by combination analysis of slight abnormal data and normal data, fully utilizing the computing power of the cloud platform, without occupying local computing resources, preventing local lag from causing work to be not smooth, and improving the efficiency of work and the accuracy of bolt quality and strength analysis.

[0018] By adding network compression factor when prioritizing, when encountering network difference, with threshold, it can prevent a large number of packets from rushing in, ensure that the link is not drowned by traffic flood even in poor condition, trigger the flow limiting threshold, reduce the instantaneous rush, and the link can fully utilize the bandwidth as soon as it is good. In addition, the network scaling factor and the way of selecting the upload data in real time according to the network state are analyzed, which realizes the compression or expansion of the priority of all data to be sent according to the availability at the scheduling level, so that the "most important" data always rushes to the front, prevents congestion, and adjusts the slice size and the number of concurrent channels according to the same network availability at the transmission level, maximizes the bandwidth utilization and minimizes the packet loss. The scheduling layer decides which packets should be sent first, and the transmission layer decides how to slice and send the packets that should be sent first. The two complement each other, ensuring that the queue is not chaotic and the transmission is not blocked. Key data wins pre-selection in scheduling and enjoys the safest slicing strategy in transmission, greatly improving its end-to-end success rate. The average delay and maximum throughput of the entire system can be improved at the same time. Low-priority traffic is released when the network is good, while high-priority traffic can still be uploaded quickly when the network is poor. On-site does not require manual repeated fine-tuning. Each time the network fluctuates, the system can automatically complete the double-layer response, improving stability and maintainability.

[0019] By setting the cloud privilege factor, the cloud needs to rush in instantly when it finds high-risk, to ensure immediate response to emergency events and avoid permanent occupation of bandwidth. It is smoothed back to normal over time. Since the data of normal events is usually at the back in the basic sorting process and needs to be uploaded after the high-risk data is uploaded, the corresponding time can be uploaded in priority after the cloud needs to upload the associated normal data or the associated abnormal data for further comprehensive analysis. This setting can quickly analyze and eliminate structural risks in a timely manner. Existing technologies mainly judge single events, but in actual complex projects, even if a single data is normal, it does not mean that the overall or local quality is normal. Therefore, through dynamic priority strategy, some structural risks can be well avoided, and quick response can be made when the network is not good, fully utilizing the cloud power and ensuring the accuracy of bolt data analysis and the safety of the job and the project. In addition, the cloud privilege queue and slicing strategy can significantly reduce the time delay of the first packet reaching the cloud. After the emergency privilege period ends, the local scoring and network adaptation ensure the subsequent retransmission success rate, so that the packet loss rate does not skyrocket due to temporary queuing, realizing the maximum timeliness and high reliability of key data upload. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of a bolt fastener connection strength test method provided by the embodiments of the present application;

[0021] Figure 2 A schematic diagram of the architecture of an end-side electric wrench of a bolt fastener connection strength test method provided by the embodiment of the application is shown in the figure;

[0022] Figure 3 A schematic diagram of bolt fastener data uploading of a bolt fastener connection strength test method provided by the embodiment of the application is shown in the figure;

[0023] Figure 4 A schematic diagram of the architecture of a relay station of a bolt fastener connection strength test method provided by the embodiment of the application is shown in the figure;

[0024] Figure 5 A schematic diagram of the architecture of a cloud platform of a bolt fastener connection strength test method provided by the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0025] The technical solutions in the application will be described below with reference to the accompanying drawings.

[0026] In order to make the technical problems, technical solutions and advantages of the application clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0027] Embodiment one of the application: a bolt fastener connection strength test method is provided for an electric wrench, such as Figure 1 A flowchart of a bolt fastener connection strength test method is shown in the figure, and Figure 3 A schematic diagram of bolt fastener data uploading of a bolt fastener connection strength test method provided by the embodiment of the application is shown in the figure, and the processing flow of the method can include the following steps:

[0028] S1: Real-time acquisition of bolt data reflecting the bolt fastener connection process and state and the bolt connection strength during the action process through the sensor of the electric wrench. The electric wrench includes the conventional component composition and functions of a commonly used electric wrench, including: a power and transmission system, a counterforce and clutch mechanism, a power supply and heat dissipation component, and a shell and protective outer shell, and further including: a sensing and measuring mechanism, a control and communication mechanism, and a human-computer interface and indication mechanism, and specifically including the following components: a torque sensor (strain gauge or torque rotor) to obtain a real-time torque-time curve, peak torque, for judging over-tightening / under-tightening, estimating pre-tightening force, and determining whether the breakaway torque / rotation torque is lower than the design residual force, a rotary encoder (optoelectronic / magneto-electric) to obtain rotation angle and number of turns, and further calculate the number of turns tightened, detect jamming, and identify slipping and sudden jumping events through angular velocity / IMU sensors, a slip / clutch trigger switch provided inside the wrench to prompt early wire slipping or early release of the wrench clutch, a vibration sensor (MEMS) to collect high-frequency RMS and peak value for detecting abnormal friction and fracture sound patterns, an acoustic emission sensor to detect ultrasonic activity for detecting fine cracks and wire slipping progress, a temperature sensor to detect the temperature rise of the wrench head or the bolt for judging the heating caused by long-time loading or jamming, a current / power meter to detect the cross verification of the driving motor current and the torque curve for detecting hard collision, a GNSS / UWB module to record the spatial position of each bolt (for high progress engineering), and a timestamp / RTC precise time to calculate the time effectiveness and aging index. In addition to the basic functions of the above-mentioned sensors, not all of them need to be configured, and users can select them according to actual conditions. The specific sensors used can also be replaced by other sensors with similar functions and external devices. The wrench of the present embodiment mainly obtains relevant data of tightening bolts, loosening bolts, and replacing bolts during the working process, analyzes the characteristic data of the bolts to obtain quality-related data, and analyzes the process-related data to obtain the failure and risk of the bolt fastener caused by non-quality factors.

[0029] As Figure 2As shown, the architecture schematic diagram of the end-side electric wrench of the bolt fastener connection strength test method provided by the embodiment of the application, the sensor data is collected through the sensor in the end-side electric wrench, and is classified according to the classification events, wherein the classification events include: on-site response, semi-response and delay evaluation, is uploaded to the relay station, waits for the pre-processing feedback of the relay station, and sends the on-site feedback to the end-side electric wrench, wherein the on-site feedback includes green light, yellow light, buzzer and text instruction, the end-side electric wrench adjusts the operation according to the feedback, analyzes the bolt data, and classifies each action as one of the on-site response type event judged and fed back by the electric wrench in real time, the delay evaluation type event relying on the background deep calculation and analysis, and the semi-response type event discovered by the electric wrench but needing the background assistance to confirm and process, and the specific classification also needs to be determined according to the specific software and hardware configuration of the wrench, the relay station and the cloud platform, for example, in the process of power equipment construction, such as the installation or maintenance and replacement of parts of the power tower, power transformation equipment and other equipment: Figure 4 As shown, the architecture schematic diagram of the relay station of the bolt fastener connection strength test method, first, the relay station receives the data uploaded by the wrench, pre-processes the semi-response and delay evaluation events, after the on-site feedback, the relay station detects the network state, then performs priority evaluation, including: risk, criticality, timeliness and aging, and receives the urgent / association instructions of the cloud to jointly perform network bottom protection and right insertion, second, the uploading queue is generated by sorting, the relay station selects the slice size and the number of concurrent to perform adaptive breakpoint continuation and retry, and finally feeds back to the relay station.

[0030] Example A: fastening process, the data of the sensor is: the measured peak value is 395 N·m (specification: M20x2.5, 10.9 level, high-strength bolt; design pre-tightening torque 400±40 N·m), the number of turns is greater than the standard, there is no slip sign, and the IMU angular velocity is smooth, so that all indicators fall within the allowed window, and the green light “qualified” is directly turned on, at this time the wrench buzzes once, the data is locally backed up and uploaded; the cloud defaults to qualified for reference.

[0031] Example B: loosening process, the starting torque is 150 N·m (theoretical residual 160 N·m), the torque-angle curve smoothly decreases and has no high-frequency vibration, the starting torque is consistent with the residual force, the disassembly is normal, the data is locally backed up and uploaded.

[0032] The above examples A and B are both on-site response type events, and the wrench gives an explicit prompt, which usually does not need to be analyzed further.

[0033] Example C: Tightening process, measured peak 720 N.m (M24, 10.9 grade, target 650 ± 65 N.m), torque slope normal, angle satisfied, micro-rebound too large 18 N.m, analysis result: over-torque + large rebound = possible bolt hole paint crushing or gasket yielding, wrench then bright yellow light; label "half-response", local backup and upload relay station preliminary recalculation or cloud secondary determination, recalculation includes specification deviation recalculation, rebound-stiffness diagnosis and aggregation / batch correlation analysis, specification deviation recalculation is to call the design database to additionally consider the temperature of the day, lubrication coefficient, and to re-correct the calibration matrix, re-judge, rebound-stiffness diagnosis is: difference FFT analysis with the same tower position, same batch M24 historical curve, if rebound 18 N.m exceeds statistical 95% percentile, so it is possible that the gasket / paint layer yields, confirm whether the large rebound is normal paint compaction or abnormal softening / yielding, aggregation / batch correlation is: check whether there are multiple over-torques within 10 minutes in the same tower section through spatial-temporal clustering (DBSCAN), check whether the material batch code matches, whether the bolts of a certain batch are generally too hard or the hole position plating is too thick, determine whether this is an isolated event or a batch process deviation, and give a "single-point re-inspection" or "whole batch sampling" suggestion.

[0034] Example D: Loosening process, breakaway torque 450 N.m, exceeds batch mean x 2, torque-angle curve drops multiple times, temperature rise ΔT + 9 °C, judge as possible rust or seizure, wrench cannot distinguish whether to replace the hole position, on-site prompt "need background support", upload the whole wave curve by the relay station or the cloud platform for further analysis.

[0035] Examples C and D above are both half-response events, the wrench finds abnormalities but cannot make a conclusion, and needs background assistance.

[0036] Example E: Tightening process, peak 620 N.m (within the specification window), 4 Hz micro-vibration pattern appears on the curve, as a single data cannot be determined, it needs to be compared with the batch vibration template in the cloud to identify early fatigue, the wrench then prompts "delayed evaluation" label; backup and upload, the cloud performs batch FFT clustering for further analysis.

[0037] Example F: Loosening process, breakaway torque 60 N.m, (according to the residual stress ≥ 140 N.m) is obviously too low, the residual clamping force may be insufficient, no slip is seen, the breakaway torque is too low, the nut may be self-loosening; need to aggregate the same batch in the tower section, delayed evaluation, further analysis by the spatial-temporal aggregation algorithm in the cloud.

[0038] Examples E and F above are both delayed evaluation events, the wrench cannot see the problem locally, but needs batch / historical comparison.

[0039] From the above examples, it can be seen that the wrench can immediately judge the qualified / abnormal for the on-site response type event. The fixed threshold or single rule in the wrench firmware is required. The information comes from the original sensing signal of the current action (torque, angle, slip sign, etc.). The algorithm is simple and can be concluded in milliseconds and prompt the operator through the buzzer / indicator light. The wrench can determine part of the semi-response type event and needs the backstage assistance for confirmation. This type of event usually has boundary abnormalities or mixed multiple signs. The wrench can detect the "suspected abnormal" trigger condition, such as: peak slightly exceeds / undershoots + large rebound, significant high rotation torque + multiple slips, but cannot determine whether it is material yield, hole position problem, or lubrication / temperature caused. It needs to be combined with the design database, the friction coefficient of the same batch, and the historical curve of the same batch to make a final conclusion. The local storage and computing power of the wrench are insufficient to complete it. It will turn on the yellow light and upload the complete waveform, waiting for the relay station or the cloud to give the "re-tightening / replacement" instruction. The wrench cannot directly identify the delayed evaluation type event. Single data may appear normal and needs to be compared with a batch or history. For example: micro-vibration marks 4Hz, slight and consistent low rotation torque; slight deviation at multiple points in the same tower section. This kind of trend / aggregated abnormality can only be revealed in a space-time, multi-sample statistical or machine learning model. The wrench usually cannot store enough comparison samples and has no batch calculation ability. Therefore, it is silent uploaded for batch processing in the cloud. If a single action meets multiple event conditions, the real-time judgment and feedback of the on-site response type event by the electric wrench is given priority. The rest is temporarily disabled. A single tightening action may trigger multiple rules. For example: the torque peak is 15% lower than the lower limit. The wrench analyzes it as under-tightening (which should belong to "on-site response type"). At the same time, it detects that the rotation torque is significantly lower than the batch average, which is suspected of self-loosening (which can be classified as "semi-response" or "delayed evaluation"). At this time, the safety and assembly quality on site are ensured. The "on-site response type" event is the event that the wrench can determine in milliseconds and must immediately remind the operator. If other types of logic are also prompted at the same time, it will cause multiple or even contradictory prompts, which will interfere with the construction. At this time, if the action meets the "on-site response" condition, a single result corresponding to the "on-site response" is output. The rest of the classification labels (semi-response, delayed evaluation) are no longer prompted to the operator, but are written into the meta field of the data package and uploaded together with the record. Temporary failure is not discarded. The relay station or the cloud can still view these backup labels in the background. The wrench working site has only one clear instruction to avoid misoperation. All clues are preserved for future review. This setting carries only one main label for a record. The algorithm and log are more concise.

[0040] S2: Transfer the data in S1 to the relay station. The data here includes not only the original sensing curve data but also the derived features and labels, which facilitate the relay station to continue scoring, rearranging, or locally reviewing the data.

[0041] S3: receiving the information returned by the relay station and prompting the operation, the information is the feedback information processed by the relay station according to the data uploaded by the electric wrench and / or the feedback information processed by the cloud platform according to the data uploaded by the relay station, the feedback information is used to reflect the strength risk and operation risk of the bolt fastener and / or related structure.

[0042] The specific acquisition process of the feedback information processed by the relay station according to the data uploaded by the electric wrench is as follows: the relay station pre-processes the cloud platform data uploaded by the electric wrench, and the pre-processing includes preliminary feedback response to delay evaluation events and semi-response events, the preliminary feedback response means that the relay station processes events with processing capability and feeds back the feedback information to the electric wrench end.

[0043] The need for pre-processing in the embodiment is flexibly selected according to the configuration, operation capability and data type possessed by the intermediate station, and the main purpose of pre-processing is to feedback as soon as possible, so as to avoid delay of engineering progress and delay of dangerous early warning caused by too long feedback time of the cloud platform or network condition limited by the construction site, for example:

[0044] Lightweight recalculation of semi-response events: after the data (for example C above: M24 bolt in slight over-torque (720N·m) + rebound 18N·m) arrives at the relay station, the system will immediately do a lightweight recalculation to determine whether it needs to be corrected immediately on site, the relay station reads the upper and lower torque limits T n , T x of the bolt specification from the local simplified model (or lookup table), loads the friction coefficient K y and temperature correction coefficient α p sampled on site on the same day, and uses the formula ; in one bolt fastening action, the real-time torque is constantly measured by the torque sensor inside the electric wrench at a fixed sampling frequency, and the highest instantaneous torque value recorded is called T p , the torque peak value is converted into equivalent axial force T', the error of lubrication and temperature on the pre-tightening force is eliminated, and T' is compared with the allowed window [T n , T xCompare. If T′ falls within the window, it is judged as qualified. If it still falls outside the window, it is judged as continuously abnormal. The relay station returns the result to the wrench through the ACK channel. When it is qualified, the green light is on and beeps once, allowing the operator to continue with the next bolt. When it is continuously abnormal, the yellow light is on and beeps twice, and附带文字指令如“逆转10°后再按650N·m复拧”,提示现场立即修正。再例如:对半响应事件,变电站隔离开关底座,M16×2.0,8.8级高强螺栓,设计预紧矩:210N·m±10%(窗口189–231N·m),实时检测到203N·m(落在设计窗口),拧紧中连续检测到2次短暂打滑,每次持续约80ms,扳手固件据此自动把事件标为半响应,并亮黄灯,提示“疑似打滑”,IMU检测到的高频振动RMS比同批平均值高1.8σ,支持“螺纹或垫片微滑”这一判断,但是,轻量规则仅能处理数值超限(欠扭 / 过扭 / 大回弹),打滑趋势分析需对Slip位置、持续时间、波形细节与历史模板比对,超出嵌入式5ms规则范畴,同一规格螺栓在其他塔位偶有短滑而仍合格,中继站无法判断这两次80ms滑移是否导致螺纹微损伤,需要云端调用波形DTW模型+批次聚集,综合多颗螺栓滑移模式,才可给出“可接受磨合”或“潜在螺孔损伤”结论。再例如,在给220kV变电站GIS平台拧紧一颗M20×2.5、10.9级地脚螺栓时,扳手测得峰值扭矩392N·m(设计窗口360–440N·m)与总旋转角5.1°均完全达标,但在3.8–4.2kHz高频段捕获到0.28g的微振RMS,比同批施工样本均值高出约2.2个标准差,扳手因此将该记录打上(延迟评估类)静默上传,中继站只附上“未预处理”标签后排队上云,因为判断这一振动偏高是否意味着螺栓早期疲劳或平台整体松弛,必须依赖云端对同塔位数十条FFT波形做批量统计与空间聚类,中继站本地无法给出结论或现场指令。

[0045] It should be noted that there are some inaccuracies in the original Chinese text, such as "附带文字指令如‘逆转10°后再按650N·m复拧’" which seems incomplete in expression. The above translation is based on the existing text as accurately as possible.The specific acquisition process of the feedback information processed by the cloud platform according to the data uploaded by the relay station is as follows: the relay station performs priority evaluation and sorting on the data to be uploaded. The data to be uploaded includes, in principle, all original data and calculation process data of the wrench and the relay station. The cloud platform will review and store all the data in idle time, and analyze relevant statistical data, such as analyzing the quality of a batch of bolts or the aging state of the bolts of a certain device. However, during real-time construction, the uploaded data needs to have a priority, and different events correspond to different importance and impact levels. The main data to be uploaded in priority is the unprocessed semi-response type and the delay evaluation type, the on-site response type event, which has been properly handled, and the related data of which no other problems are found only needs to be uploaded later. There are also semi-response type events, and the part that has been handled on site also needs to be uploaded later in the sorting. When calculating the priority, the risk intensity, denoted as the original risk value r, such as under-torque, over-torque, continuous slipping, abnormally high-frequency vibration, abnormal temperature rise, etc. is mainly considered. This kind of data is the most direct measure of safety. The specific acquisition process is as follows: the wrench samples the original signals such as torque, angle, slipping flag, IMU vibration, temperature rise, etc. at a fixed sampling frequency during work. The firmware extracts features {x i} from each frame of signal, including peak torque difference, rising slope, rebound amplitude, slip count, vibration RMS, etc. i is the type of extracted features, and the pre-burned weight is calculated as r, where, is the "dangerous degree weight" assigned to each abnormal feature x i . These weights are usually determined by first having a seed value, then offline calibration, and finally online fine-tuning. The purpose of the seed value is to first make the model run, and then to fine-tune it accurately. The seed value can usually be scored by engineers, who give each feature an importance level of 1-5 according to the FMEA or AHP approach, and then normalize it to 0-1 to obtain it. Alternatively, the relative value of each feature change 1σ to the theoretical influence on the axial force or failure probability is taken as the weight, and then offline hyperparameter search is performed to make the weight approach the optimal value. Prepare labeled data, such as original waveform + wrench feature corresponding label: OK / NOK / repair / failure 4 class labels, then define the target: maximize the recall rate of high-risk samples Top-k, or minimize the overall F1 loss, then search, strategy: use grid or Bayesian / or HyperOpt traversal a1…a i ∈[0,1],or directly train logistic regression or GBDT, and then use cross-validation to pick aᵢ * , where aᵢ * represents the i-th component in the final selected optimal dangerous degree weight, which is the optimal influence factor of the overall risk evaluation. Then normalize and issue;

[0046] The specific calculation formula of the original risk value r is as follows:

[0047] ;

[0048] In specific use, the optimal vector can be linearly normalized so that =1, m represents the total amount of extracted feature types, and the weight is usually required to be re-verified for a certain period of time, and finally online fine-tuning is performed to keep sensitive to scene drift. First, coarse weights are given by experts or physical models, then historical labels are used for offline hyperparameter search calibration, and finally online monitoring is performed through the cloud to make slight self-adaptation. This method can quickly land and continuously approach the optimal value according to environmental, material and operator differences. The denominator is always 1 according to the design, but in actual engineering implementation, retaining the denominator is beneficial to supporting feature channel gating, data quality self-adaptation, and can also prevent risk value scale drift caused by slight deviation of the weight sum from 1 during floating point operation and online fine-tuning, thereby ensuring the stability and scalability of the system.

[0049] k is the structural criticality, denoted as engineering criticality k, for example, the importance of the node where the bolt is located to the overall structure, the anchor bolt corresponding to the secondary material connection, the power equipment flange corresponding to the general cable support, which is usually marked by a design database or BIM. Specifically, each bolt has a node level bound in a work order or BIM model (tower leg anchor bolt 1.0, secondary material connection 0.3…), the wrench scans the code or reads the bolt unique identifier through NFC, the bolt unique identifier is the "identity card" assigned by the system to each monitored bolt or each hole position, and the corresponding weight k∈[0,1] is directly queried in the local mirror table. In addition, it is not necessary to "scan the code or NFC" for each bolt in the field to obtain k; the criticality value can be delivered once before work or automatically matched by position, for example, in the tower assembly and support row scenarios, the wrench can batch load the "tower section parameter table". Before construction, the relay station writes the CSV (bolt serial number→k) of the tower section into the wrench once, the wrench works in sequence, automatically increments the index, and the worker is already working according to the drawing sequence. Once downloaded, it can be used offline without the need to scan each code. For example, in the scenario of outdoor iron towers and power equipment pedestals, the position acquisition criticality mapping (UWB / GNSS coarse positioning) can be used. The wrench obtains the coordinate grid in the preset interval of the tower leg NE / SE / SW / NW, and the relay station queries the local LUT: NE tower leg=k=1.0, secondary material area=0.3. At this time, the UWB base station needs to be in place at the tower foundation, and the positioning accuracy only needs to be a few tens of centimeters. In addition, there are other ways to determine the structural criticality, such as hierarchical and sequential work, and the above methods can be combined to determine the structural criticality. Time urgency t is mainly reflected in the remaining time-to-live ratio, which is the remaining time to the end of the construction process or the acceptance of the process. The closer to the deadline, the higher the score. An exponential steep increase model can be used to obtain the process scheduling table, which records the start time and the end time of the node. The relay station calculates t in real time according to the following formula, given the wrench timestamp:

[0050] ;

[0051] The queuing aging parameter a mainly considers the waiting time length, and the real-time cumulative time of data staying in the relay station queue, and forms an aging score after normalization; it is ensured that low-risk packets can still be transmitted after long-term queuing, and are not "starved to death"; specifically, the entry time T i is recorded in the cache queue, and a is calculated according to the following formula each time the scheduling is performed, to normalize the aging value, so that the longer the waiting time, the closer a is to 1:

[0052] ;

[0053] is the maximum allowed queuing time, which is much larger than the preset maximum queuing time set in advance by the preset personnel; represents the current time; wherein, the basic upload priority result :

[0054] ;

[0055] wherein, p is a risk amplification index, the risk amplification index is set in advance by the preset personnel, and p>1, the higher r is, the higher the risk impact index is, and the lower r is, the lower the risk impact index is; wherein, the risk impact index is represented by p times of r, and 1-e -λ×t is the timeliness impact index; the risk weight w r , the criticality weight w k , the timeliness weight w t and the queuing aging weight w a are respectively the weights corresponding to the risk impact index, the structural criticality, the timeliness impact index and the queuing aging parameter, and the range of each weight is 0 to 1, and the sum of the four weights is 1; these weights and other similar weight setting modes of the present application are similar, and can be initially set according to engineering experience and then debugged; the timing of debugging can be, the historical data of the scheduling system can be combined for tuning in the later period, or the weights can be searched through existing mature multi-objective optimization (such as maximum recall + minimum delay).

[0056] Since part of the data has been processed in the preprocessing stage, this part of the data is usually uploaded later, so it is necessary to increase the discount item χ(d) of the processed data on the basis of the basic upload priority result, and the final upload priority result is:

[0057] ;

[0058] Wherein, the value of K is affected by many factors, for example, if the processed package still has residual risk, it is hoped that they will arrive at the cloud faster, K should be large, if the on-site confirmation is reliable enough, the upload can be delayed, K should be small, at this time, FMEA / SIL can be used to determine the "tolerable residual risk" for the construction stage, the lower the residual risk, the lower the value of K, K can be regarded as a linear mapping of "risk residual tolerance" in the range of 0-1, in addition, the more crowded the network, the stronger the inhibition of processed packages to prevent bandwidth contention, K will be small, in addition, it is also related to the aging window and operation and maintenance strategy, the specific initial value can be more experienced and set artificially, 0.45-0.55 can be used for the same industry and link to run through, and the completion rate and retransmission rate are used as feedback to fine-tune with simple PID or Bandit, then K will automatically converge to the minimum value that meets the SLA. According to the priority order, the data that needs to be uploaded is uploaded to the cloud platform in real time and dynamically according to the network state, the relay station does not send the data in the cache to the cloud in a fixed interval and fixed way, but feeds the packages one by one to the network according to the priority, and adjusts the sending strategy at any time according to the network fluctuations or queue changes, the specific strategy includes selecting the slice size and concurrent channel mode, the relay station measures the effective throughput, round-trip delay and packet loss rate every few seconds to tens of seconds, and obtains the current network availability n in real time, the highest upload priority result record in the queue is taken first, that is, the data with high risk, high key, time urgency or being promoted by the cloud is dequeued first, the on-site processing or low-risk is delayed, and the data is adaptively divided and sent according to the network condition, for example, the first package is taken, the original package is cut into (2-8KB) fragments according to the current network strategy, and the metadata such as total_chunks and chunk_id is recorded, then concurrent or breakpoint scheduling is performed, the single slice size and concurrent channel number are dynamically adjusted according to the availability n, when the ACK is successful, the next slice is sent, when the timeout or packet loss occurs, the retry or small block length or fewer channels are performed, after all the slices are ACKed, cloud_flag=1 is set, if the network is disconnected, the unfired slices are re-entered into the queue to wait, this operation selects the package first and then cuts the slices: ensures that the business priority is always based on "complete package" as the granularity, the slicing strategy is aligned with the network: the current n is good network, then cut large blocks and concurrent multiple channels, poor network, then cut small blocks and fewer concurrent channels, the subsequent ACK and retransmission are based on the fragments: convenient for breakpoint continuation and packet loss recovery.The slice algorithm described above can realize automatic identification of network segment by the relay station, give the block length and concurrency parameters, queue according to business first, then slice and send, ACK (Acknowledgment, acknowledgment) / timeout callback adjustment, through this state machine + self-adjusting mechanism, without manual intervention, the block can be reduced to 2-4KB in poor network, and increased to 32KB in good network, and the first piece of high-risk packet is always the fastest outbound. Regarding the slicing strategy, engineers can refer to the throughput curve and packet loss curve of similar sites in the past, or directly use the recommended default value as the initial template. The relay station automatically detects online and adjusts adaptively. The initial template can be set offline in advance, and the network availability is mapped to the network classification gear. The corresponding segmentation template is read and adjusted in combination with the slice confirmation and timeout in the last sending window. The network classification gear is usually 3-4 network gears, but other number of gears can also be set. The initial template refers to a set of general transmission parameter table pre-configured for fast online in the absence of historical link data of the site. The template gives the recommended slice size, concurrent channel number and timeout threshold upper limit according to the fixed network availability interval. Once it is issued, the relay station can start the upload process accordingly, and automatically fine-tune through online detection and AIMD mechanism in subsequent operation. For example: initialization: the system presets three network states (GOOD≥0.75, FAIR 0.35-0.75, POOR<0.35), which correspond to block length 32kB / 8kB / 4kB and concurrency 8 / 4 / 1 respectively. Real-time detection: the relay station sends a detection block every 8s to measure the throughput B_now, calculates the availability n=B_now / B_peak, and updates the current network gear according to the hysteresis logic, and adjusts adaptively: in the FAIR and POOR gears, if the same piece of 3xRTT (Round-Trip Time, Round-Trip Time) is not ACKed or the packet loss is >15%, the block length is halved (lower limit 2kB), and the number of concurrent pipes is halved (lower limit 1). When 10 consecutive pieces of ACK are successful, increase the concurrency by 1 (upper limit 8) according to AIMD. Cloud hot adjustment: when the 30min retention rate is >20% or the retransmission rate is >10%, the cloud can adjust the chunk_size list by one gear through OTA; when the indicators recover, automatically return to the previous gear.

[0059] The cloud platform receives the data uploaded by the relay station for dynamic analysis to obtain remote feedback information and transmits the remote feedback information to the relay station; the relay station sends the feedback information to the electric wrench end according to the remote feedback information, and the relay station pushes the bolt data to the cloud end portal; the platform immediately checks and writes into the high-speed message queue, first ensures that the data is not lost and can be concurrent, the cloud platform continuously listens and receives the data packets uploaded from each relay station, each data packet contains the tightening characteristics, risk analysis value, event identification, timestamp and state label of the corresponding bolt, and the stream computing task reads the queue data and derives the brief characteristics (such as slip count, rebound amplitude, etc.) at the same time. Specifically, the cloud service will parse the received raw data, disassemble each feature index, and store it in a distributed database or data lake, preparing for subsequent analysis. The feature index specifically refers to the numerical characteristics reflecting safety hazards or working condition changes extracted from the signals (such as torque, angle, vibration, temperature, etc.) collected by the wrench during bolt tightening, such as peak torque difference (reflecting tightening strength), rebound amplitude (reflecting whether it is loose), rising slope (reflecting loading speed), slip count (reflecting slip phenomenon), vibration RMS (reflecting vibration intensity), and temperature rise rate (reflecting thermal abnormalities), etc. Then call the online model to give each record a risk score and risk type. Specifically, use multiple models (such as anomaly detection, rule logic, clustering analysis, trend prediction, etc.) to make preliminary judgments on the data, and identify events or key data with potential risks. These models are automatically invoked according to data dimensions, risk levels, and scene history labels. The system selects the most suitable model for judgment according to the event type or field structure. For example, detect whether there are multiple slips, torque abnormalities, and node rebound surges. If the judgment result shows that the current event has safety risks, quality hazards, or key values, the system will generate feedback decisions. The generated feedback content (including feedback target, feedback level, instruction type, etc.) is packaged into a remote control message and sent to the corresponding relay station. For example, the feedback content "need to supplement the adjacent bolt data" or "force the event to return all original waveforms" will mark the privileged event, associated data list, update upload priority, etc. After receiving the feedback, the relay station analyzes the cloud-end issued privileged instructions, re-evaluates the local data packets or updates the sorting rules, and adds the specified data to the priority upload queue to trigger the upload operation.The wrench uploads real-time torque-angle curve, vibration RMS, etc. original features, and uses the IsolationForest model. The IsolationForest model can be trained offline in the cloud in advance. The training data includes normal curve, fault curve, typical normal: abnormal according to a certain proportion, for example, 20:1, which meets the IF "abnormal is rare" hypothesis, etc. dataset, input: peak torque, rebound amplitude, rising slope, slip count, vibration RMS, torque 10%→90% rise time, final tightening angular velocity and total angle, firmware side first calculation and normalization (z-score), pooling: sliding window size 1 (pure real-time), threshold: abnormal score >0.55 judge Suspect, the determination result is: the curve: T_peak=570N·m (target 650±65), abnormal score 0.61, then judge under-tightening Suspect, the cloud generates JSON feedback content, the relay station acts, after receiving, the wrench OLED pops up a yellow light + short vibration, prompting "under-tightening, reverse 30° and tighten again", this operation only takes a very short time for single stream detection, ensuring that the operator can receive the correction instruction before the next bolt. For example, life prediction and trend prediction can also be performed. The anchor bolt of the seaside wind turbine foundation has been in service for 1 year. The wrench uploads "breakout torque + environmental humidity + salt spray index" during weekly inspection. The LSTM+Weibull mixed model is used. LSTM predicts the future 60-day breakout torque decay curve (input sequence length: past 52 weeks). The predicted value is mapped to the residual axial force, which is fed into the Weibull failure model to estimate the loosening probability curve. The model estimates that the probability of this bolt (or group of bolts) dropping below the safety threshold in the next 30 days is 22%, with a confidence interval of 17%-30%. The feedback content is a weekly email or an operation and maintenance dashboard. If the anchor bolt loosening probability is greater than 20% in 30 days, please re-tighten the entire bolt during construction. The relay station does not need to act immediately. The operation and maintenance manager adds the tower leg to the subsequent list, and the wrench task list automatically appears. Therefore, the cloud platform can access and reliably store the obtained data in high concurrency. The cloud platform can perform real-time anomaly detection in seconds, determine and feedback to the site immediately when a single bolt is under-tightened / over-tightened or the sensor drifts. It can perform batch and spatial aggregation analysis in minutes to find systemic quality problems in "same batch" and "same tower leg", intercept rework in advance, and perform trend prediction and life estimation in hours to days to predict the pre-tightening force decay curve and generate preventive maintenance plans. The model and parameters are offline retrained, and new models, new scheduling templates, and wrench calibration coefficients are updated in seconds. The platform provides unified visualization and reporting, operation and maintenance dashboards, construction progress boards, and quality traceability reports in one-stop presentation. It compares multiple sites and optimizes globally, analyzes batch defects, supplier quality, and regional climate impact across sites, and traces permissions and operations at all levels of the life cycle. It integrates with BIM / ERP / CMMS, writes real-time tightening results back to BIM, and automatically generates work orders and spare parts plans.

[0060] In addition to the slice-concurrent self-adaptation, the queue sorting layer also performs a whole compression priority on all services, and the high and low order between packets is maintained, but the absolute score is reduced simultaneously due to the network difference, triggering the flow limiting threshold, and reducing the instantaneous influx. (Assuming that the total capacity of all data packets is 600 MB waiting to be uploaded, and the data packets are cut into 2 kB, the number of pieces is approximately 300,000. Even if the single piece is light, the continuous arrangement into the sending window will still cause the ACK queue, retransmission buffer, and CPU interrupt to increase), at this time, in the poor network, the whole compression is set to multiply the scores of most packets by 0.3, only a small batch of data packets with scores still higher than the threshold can pass the gate, and the rest of the data packets are directly left in the buffer and do not participate in the current round of uploading. When the network is poor, the scores of all packets are multiplied by a factor less than 1, and the threshold is dynamically raised. Only a small batch of the most critical data can pass the gate to avoid "peak" type slice and retry, and the pressure on the network layer and system resources is controlled within a bearable range. As the network improves, the threshold is dynamically lowered, and more packets will be uploaded in batches. When the network is good, the factor is approximately equal to 1, and there is almost no compression. In addition, the fluctuation of network availability will affect the network compression factor, and in turn affect the upload priority value of each event. If the whole queue is recalculated frequently, unnecessary calculation pressure and frequent changes in upload order will be caused. Therefore, a rhythmic updating strategy can be adopted, which does not update all priorities every time the network fluctuates, but refreshes the whole queue once every certain period. Between two refreshes, the current priority value remains unchanged. If the network deteriorates severely, an emergency compression recalculation is triggered, or the events are divided into multiple levels according to the priority, and the data in the high priority buffer area is dynamically scheduled frequently, and the data in the medium and low priority buffer areas is scheduled less frequently. Because only the high priority events participate in the upload window competition, the network compression factor will change in real time, and the time urgency and queue aging value will also change over time. Therefore, when calculating the priority value, the whole queue is not recalculated frequently, and the discount item may change with the state. Therefore, the discount item is usually not real-time rearranged, but only the priority is refreshed at the update period.

[0061] It needs to be explained that the threshold here does not mean that low priority data packets are completely discarded, but is used to control which packets can enter the sending window and occupy the link bandwidth. The rest of the data packets are left in the local buffer and do not participate in the current round of uploading. When the network is poor, the scores of all packets are multiplied by a factor less than 1, and the threshold is dynamically raised. Only a small batch of the most critical data can pass the gate to avoid "peak" type slice and retry, and the pressure on the network layer and system resources is controlled within a bearable range. As the network improves, the threshold is dynamically lowered, and more packets will be uploaded in batches. When the network is good, the factor is approximately equal to 1, and there is almost no compression. In addition, the fluctuation of network availability will affect the network compression factor, and in turn affect the upload priority value of each event. If the whole queue is recalculated frequently, unnecessary calculation pressure and frequent changes in upload order will be caused. Therefore, a rhythmic updating strategy can be adopted, which does not update all priorities every time the network fluctuates, but refreshes the whole queue once every certain period. Between two refreshes, the current priority value remains unchanged. If the network deteriorates severely, an emergency compression recalculation is triggered, or the events are divided into multiple levels according to the priority, and the data in the high priority buffer area is dynamically scheduled frequently, and the data in the medium and low priority buffer areas is scheduled less frequently. Because only the high priority events participate in the upload window competition, the network compression factor will change in real time, and the time urgency and queue aging value will also change over time. Therefore, when calculating the priority value, the whole queue is not recalculated frequently, and the discount item may change with the state. Therefore, the discount item is usually not real-time rearranged, but only the priority is refreshed at the update period.

[0062] ;

[0063] wherein, n is the real-time network availability, For the current effective throughput, For the peak throughput, it can be selected to detect once in a certain time, for example, detect every 5-10s, and the peak can be selected as the 24h peak throughput; , , , are all segment points, ; , , are the compression coefficients of the left end points (start) of each segment, and the value range is ; , , are the compression coefficients of the right end points (end) of each segment, and the value range is . , , are all greater than 0, and are exponential factors for controlling the steepness of the interpolation curve of each segment. The specific size can be set according to the actual steepness requirement, and in addition, it can also be fine-tuned in actual use. Therefore, the 0th segment is , at this time it is a poor network, and the network compression factor rises from min to max , and the power controlling the smoothness is ; Similarly, the first segment is , at this time it is a transition network, and rises from to ; the second segment is , at this time it is a good network, and rises from to , and usually . The network compression factor g(n) and the slice use the same network availability n, but g(n) is calculated only once, changing "who can enter the window", and the slice may be self-adjusted every piece ACK / timeout, changing "how to send in the window". The two work together, the slice+small block controls the single piece size, and g(n) overall compression controls the number of active packets. The two work together to prevent queue burst under extreme link, and the first piece of the real high-risk packet always arrives at the cloud first. Therefore, the priority of the relay station to the cloud platform at this time is to sort the upload according to the size of , and in the upload process, the size of the upload block is further adjusted by the slice algorithm, wherein .

[0064] Embodiment three: on the basis of other unchanged in embodiment two, in specific scenarios, the network grading of three segments cannot be applied to all scenarios, and the selection of segmentation also depends on the network situation on site. Usually, three segments and four segments can meet most scenarios, and of course, it can also be divided into multiple segments. The following gives the general formula for dividing into multiple segments:

[0065] ;

[0066] Where K is the total number of segments; m is the current segment number, which can range from 0 to k-1, therefore the nth segment number is... The range of real-time network availability for a segment is [n m ,n m+1 ), n m For the first left endpoint of segment, n m+1 For the first The right endpoint of the segment; (n) is the indicator function, nϵ[n m ,n m+1 At that time, it was 1, and the rest were 0; This is the starting compression value for the current segment. The end compression value of the current segment; It is a real number and greater than 0, used to control the steepness of the current segment of the curve. The larger the value, the slower the growth in the early stage and the steeper the rise in the later stage. For the normalized interpolation coordinates within the current segment, compared to the explicit three or four segments, the function with an indefinite number of segments in this embodiment divides the entire input interval into an arbitrary number of sub-segments and defines a compression interval and interpolation curve shape for each segment. This general formula can accurately fit various complex resource scheduling or performance compression requirements. It supports expanding the number of segments as needed and refining the response strategy for each segment. It is suitable for multi-task, multi-level, and multi-state systems, and is particularly suitable for scenarios requiring adaptive control, automatic optimization, or policy learning.

[0067] Example 4: Based on Example 1, 2, or 3, the cloud platform receives data uploaded by the relay station, performs dynamic analysis to obtain remote feedback information, and transmits it to the relay station. After the cloud platform provides feedback, it determines that relevant and related data need to be uploaded. This data may be of high or low priority, but the platform determines it to be an urgent event. Therefore, uploading data according to the priority calculation method of Example 1, 2, or 3 may result in the inability to extract key urgent data in a timely manner. Therefore, this example corrects and provides feedback on the upload priority of Example 1, 2, or 3, forming a new priority ranking. When the cloud detects a high-risk event, it needs to "instantly jump the queue" to ensure immediate response to the emergency event, while avoiding permanent bandwidth occupation and smoothly returning to normal over time. Therefore, a cloud-based priority factor z(τ) needs to be added.

[0068] In particular, when the cloud platform dynamically analyzes the uploaded data, it finds that a certain key event needs to supplement the uploading of its "associated data" (such as other bolt data of the same tower section, historical trend waveforms, etc.) to complete further determination. The cloud platform encapsulates the identification of the required associated data (such as event ID, record label, and data type list) into a feedback message and issues it to the corresponding relay station through the remote control channel. After receiving the feedback, the relay station parses the associated data list and locally retrieves all corresponding records to be uploaded or cached. The associated records are uniformly marked with an "associated privilege" mark, and the privilege trigger time is recorded. For each marked associated record, the cloud privilege factor is calculated according to the privilege model. The relay station reorders the queue of all privileged associated records and other data to be transmitted based on the local final priority from high to low, and preferentially slices and schedules the uploading of the associated data. The cloud privilege factor can be exponentially privileged or decayed, that is, where μ>0,τ≥0: the number of seconds elapsed after the relay station receives the cloud "urgent" (urgent request) command, μ is the decay coefficient that determines the speed of privilege from 1 to 0, which can be configured by the operation and maintenance or automatic tuning platform (commonly used 0.1-0.3), μ and τ are reciprocal units, so the product unit is eliminated, making it easy to calculate. Through cloud privilege, instantaneous queuing can be achieved, for example, when τ=0, z=1, ensuring that the cloud-priority data packet is the first to be processed. It can also be smoothly regressed, for example: as the increases, the privilege effect index decays, and it does not permanently occupy the bandwidth. At this time, the priority result is:

[0069] or ;

[0070] Need to explain, "urgent" command usually means that it needs to be processed preferentially at this moment; with the passage of time, its "urgency" should naturally decay, otherwise it becomes permanent queuing, achieving real-time scheduling. For example: when τ=0, z(τ)=1, at this time, it is 100% dependent on cloud privilege, that is, the cloud-priority data is directly uploaded with the highest priority. When τ tends to infinity (privilege expires), z(τ) tends to 0, ϖ n The control is smoothly transferred from "cloud decides" to "local scoring", ensuring that cloud emergency commands can take effect immediately, and the bandwidth is not permanently occupied. The privilege effect can be "automatically recovered" over time.

[0071] Example Five: Based on Example One or Two or Three or Four, a network bottom line is designed. When the network is very poor, the real-time network availability n=0. At this time, the worst link still needs to be left for the key data packet. When the network is good, n=1 (ideal link), the bottom line can be reduced to ϖ n bandwidth for key data packets. When the network is good, n=1 (ideal link), the bottom line can be reduced to ϖ x, give more space to the base priority, ϖ n The minimum bandwidth ratio reserved for critical traffic, which represents the minimum bandwidth ratio reserved for critical traffic by the system in the worst case of the network (real-time network availability n = 0), ensures that in the case of almost zero bandwidth, a "lifeline" is still left for high priority or emergency events to avoid being completely starved, x is the maximum bandwidth ratio reserved for critical traffic by the system, and the maximum bandwidth ratio reserved for critical traffic by the system in the ideal case of the network (n = 1), still leaves a certain share for critical traffic (avoids being occupied by concurrent or large traffic) while allowing the base priority to also obtain sufficient bandwidth for transmission, and ϖ n , ϖ x Can be set by the operation and maintenance on the platform, so the network bottom line ϖ x (n) is specifically:

[0072] , ;

[0073] Among them, the network bottom line ϖ x (n) is a dynamic bandwidth bottom line function, which represents the bandwidth ratio reserved for high priority traffic by the platform at the current time, and can be obtained by sending fixed-size probe packets every fixed time, for example, 5-10s. Since these packets may be lost and retransmitted in the network, only the application layer data bytes that are successfully received and ACK confirmed by the cloud are counted. The cumulative valid traffic bytes in this period are divided by the measurement time to obtain the real-time effective throughput, and then the ratio of the highest real-time effective throughput observed in the past period is obtained to obtain the real-time network availability n. Then the final priority P f is obtained. The priority of transmission is sorted according to the size of the final priority, and the specific formula is as follows: when network awareness is not enabled, and no emergency queue is needed, use: no cloud privilege, no network compression, only network bottom line combined with base priority P ; or when emergency events need to be queued, but no network compression is done locally, use: with cloud privilege, no network compression, smooth transition to network bottom line combined with base priority P b . ; or when no emergency queue is needed, but network awareness compression is needed, no cloud privilege, network compression, first multiply the network compression factor, then do network bottom line, ; or when both cloud privilege and network compression are included, the complete process is: first multiply the network compression factor, and use no cloud privilege combined with network bottom line P q . The priority obtained through the above formula can achieve the purpose of cloud privilege, so that the associated data can be quickly uploaded, problems can be discovered and fed back in time, and the security is improved. It needs to be explained that for network bottom line, for the case of containing both cloud privilege and network compression factor, when the network availability n=0, the network is extremely poor, and since g(0)=0, P q =0, which is the worst network, but still needs to leave a bottom line bandwidth to transmit key packets, so at this time it is 100% pure bottom line , with the increase of n, the decision-making right is smoothly given from the "bottom line bandwidth" to the basic priority result.

[0074] Further, the embodiment of the application provides a bolt fastener connection strength test method, which is used for a relay end and includes the following steps: receiving data uploaded by an electric wrench; the data is bolt data reflecting a bolt fastener connection process and state and bolt connection strength in a motion process collected by the electric wrench in real time and analyzed and classified; the classification is one of a field response event judged and fed back by the electric wrench in real time, a delayed evaluation event depending on background in-depth calculation and analysis, and a semi-response event discovered by the electric wrench but needing background assistance to confirm and process; if a plurality of event conditions are met at one time, the field response event judged and fed back by the electric wrench in real time is given priority, and the rest is determined as temporary failure; pre-processing is performed on the data in A1; the pre-processing includes preliminary feedback response of the delayed evaluation event and the semi-response event, and the preliminary feedback response indicates that the relay station processes events with processing capability and feeds back feedback information to the electric wrench end; priority evaluation and sorting are performed on data to be uploaded; data to be uploaded is uploaded to a cloud platform in real time and dynamically according to a network state according to the priority order; the relay station sends feedback information to the electric wrench end according to remote feedback information; the feedback information is information that the cloud platform receives data uploaded by the relay station, dynamically analyzes the remote feedback information, and transmits the remote feedback information to the relay station.

[0075] Further, the embodiment of the present application provides a bolt fastener connection strength test method, which is used for a cloud platform and includes the following steps: receiving data uploaded by a relay station; the data uploaded by the relay station is bolt data reflecting a bolt fastener connection process and state and bolt connection strength in a motion process collected by an electric wrench through a sensor in real time, which is analyzed and classified, and then is evaluated and sorted in priority by the relay station, and the data to be uploaded is uploaded in real time and dynamically according to network state according to the priority order; the classification is specifically one of a field response event judged and fed back by the electric wrench in real time, a delay evaluation event depending on background in-depth calculation and analysis, and a semi-response event discovered by the electric wrench but needing background assistance to confirm and process; if a motion simultaneously meets multiple event conditions, the field response event judged and fed back by the electric wrench in real time is given priority, and the rest is determined as temporarily invalid; the cloud platform receives the data uploaded by the relay station to obtain remote feedback information and transmits the remote feedback information to the relay station; the remote feedback information is sent to the electric wrench end by the relay station and is operated and prompted by the electric wrench end.

[0076] It should be added that, as shown in Figure 5 the architecture diagram of the cloud platform of the bolt fastener connection strength test method provided by the embodiment of the present application, the cloud platform first receives and stores data from the relay station, and performs flow and batch dynamic analysis, including: anomaly detection, clustering, trend prediction, generates urgent or associated data requirements, then issues urgent / urgent_associated commands to the relay station, receives the upload state and feedback of the relay station, and again performs flow and batch dynamic analysis according to the upload state and feedback of the relay station.

Claims

1. A bolt fastener connection strength test method for an electric power wrench, characterized by, Comprise the following steps: S1: Real-time acquisition of bolt data reflecting the bolt fastener connection process and state and bolt connection strength in the action process through the sensor of the electric wrench; According to the bolt data, each action is classified into one of the following categories: a site response event judged and fed back by the electric wrench in real time, a delayed evaluation event relying on background in-depth calculation and analysis, and a semi-response event discovered by the electric wrench but needing background assistance for confirmation and processing; If a single action meets the conditions of multiple events, the site response event judged and fed back by the electric wrench in real time is given priority, and the rest is determined as temporary failure; S2: Transmit the bolt data of the bolt fastener connection process and state and bolt connection strength in S1 to the relay station; S3: Receive the information returned by the relay station and give operation prompts, the information being feedback information processed by the relay station according to the data uploaded by the electric wrench and / or the cloud platform according to the data uploaded by the relay station, the feedback information being used to reflect the strength risk and operation risk of the bolt fastener individual and / or related structure; The specific acquisition process of the feedback information processed by the relay station according to the data uploaded by the electric wrench is as follows: The relay station pre-processes the data uploaded by the electric wrench, which includes preliminary feedback response to the delayed evaluation event and the semi-response event, indicating that the relay station processes events with processing capability and feeds back the feedback information to the electric wrench end; The specific acquisition process of the feedback information processed by the cloud platform according to the data uploaded by the relay station is as follows: The relay station evaluates and sorts the priority of the data to be uploaded; According to the priority order, the data to be uploaded is uploaded to the cloud platform in real time and dynamically according to the network state; The cloud platform receives the data uploaded by the relay station, dynamically analyzes the remote feedback information, and transmits it to the relay station; The relay station sends the feedback information to the electric wrench end according to the remote feedback information.

2. A method of testing the strength of a bolted joint according to claim 1, characterized in that The specific process of the relay station evaluating and sorting the priority of the data to be uploaded is as follows: For each event, according to the bolt data sampled by the wrench at a fixed sampling frequency, the original risk value reflecting the safety hazard strength of this fastening process is obtained after feature extraction; Through the node and type query of the bolts of this event, the engineering criticality for quantifying the influence degree of the bolts on the overall structure stability is obtained; The time urgency is obtained by normalizing the operation time of this event relative to the start and end time of the process, which indicates that the closer to the end, the more urgent; Record the enqueue time of the event in the relay station upload queue, and the real-time cumulative time of the data staying in the relay station queue, after normalization and processing, form the queuing aging parameter, which indicates the priority demand that should be raised as the waiting time increases; Assign weights to the original risk value, engineering criticality, time urgency and queuing aging parameter, and discount the processed events to obtain the basic priority evaluation result of the event; Sort all events to be uploaded in descending order of their basic priority evaluation results.

3. The method of testing the strength of a bolted joint according to claim 1, wherein, The specific process of dynamically uploading the data to be uploaded according to network state in real time according to the priority order is as follows: S31: The relay station sends a probe packet at a set frequency, measures the effective throughput, round-trip delay and packet loss rate of the current link, and normalizes them into a network availability index; S32: The data packet corresponding to the event to be uploaded is maintained in a queue sorted from high to low according to the priority evaluation result; S33: The upload thread takes out the data packet corresponding to the event with the highest priority evaluation result from the head of the queue and prepares to send; S34: According to the latest network availability mapping to the network classification gear, the corresponding segmentation template is read and adjusted in combination with the slice confirmation and timeout in the latest sending window to determine the slice size and concurrent channel number of this sending; S35: The data packet ready to be sent is cut into several segments according to the determined slice size, sent simultaneously or sequentially according to the concurrent channel, and each segment is numbered and given breakpoint resume metadata; S36: Whenever an acknowledgement of a certain segment is received, the next segment is immediately sent, and if a segment is not acknowledged within a set timeout threshold round-trip delay, the segment size is reduced and sent again until successful; S37: When all segments are successfully acknowledged, the event is marked as uploaded, and the next highest priority event corresponding data packet is taken to start uploading.

4. The method of testing the strength of a bolted joint according to claim 1, wherein The specific process of receiving the data uploaded by the relay station by the cloud platform to obtain remote feedback information and transmit it to the relay station is as follows: The cloud platform continuously monitors and receives data packets uploaded from each relay station, and the data packets include: twist feature corresponding to the event, risk analysis value, event identification, timestamp and state label; The cloud service analyzes the received data packets, extracts each feature index, and stores them in the database; The cloud analysis module calls the applicable model to judge the data and generates remote feedback information, the applicable model refers to the model selected by the cloud platform automatically or by default according to the data type, business scenario and analysis target, which is used to complete a specific data processing task, and the model includes: one or more of statistical model, machine learning model and rule engine; The feedback information is packaged into a remote control message and sent to the corresponding relay station, and the remote feedback information includes: feedback target, feedback level and instruction type.

5. The method of testing the strength of a bolted joint according to claim 2, wherein, The relay station further includes: When performing the priority evaluation and sorting, the network compression factor is combined with the basic priority evaluation result to obtain an optimized priority evaluation result, and the priority evaluation results are sorted from large to small according to the optimized priority evaluation result; The network compression factor is a proportional function for adjusting the priority result value of the event under dynamic network conditions, ensuring that the relay station uploads critical data when the bandwidth is tight and limits irrelevant data when the bandwidth is abundant; The specific process of combining the network compression factor with the basic priority evaluation result is as follows: The relay station periodically obtains the network availability; And based on network availability, the network compression factor is calculated or queried from a preset network segmentation strategy lookup table; The base priority evaluation result of each event is corrected using the network compression factor to obtain an optimized priority evaluation result.

6. A method of testing the strength of a bolted joint according to claim 5, wherein, The specific process of combining the base priority evaluation result with the network compression factor further includes: The upload gate threshold is set according to the current transmission capacity, which refers to the maximum capacity of the relay station that can stably and safely upload data to the cloud platform per unit time under the current network state; Only the event with the optimized priority evaluation result greater than the threshold can enter the next round of the sending window, and the remaining events remain in the upload queue; The relay station continuously monitors the network availability and queue changes, re-evaluates the network compression factor and the base priority evaluation result of each event before each round of scheduling, and realizes dynamic adjustment and rearrangement of the upload priority.

7. A method of testing the strength of a bolted joint according to any one of claims 5-6, characterized in that, Further including: When the cloud platform analyzes and feeds back that the associated data needs to be further uploaded for cloud analysis, the relay station corrects the priority evaluation result of the associated data in the base priority evaluation result or the optimized priority evaluation result analysis, to ensure that the associated data of the emergency analysis event can be uploaded in time.

8. A method of testing the strength of a bolted joint according to any one of claims 5-6, characterized in that, When the relay station analyzes the base priority evaluation result or the optimized priority evaluation result, the network bottom line is corrected, all events are reordered from high to low according to the corrected priority evaluation result, a new upload queue is generated, and the minimum guaranteed bandwidth is ensured for critical data upload under any network condition.

9. A bolt fastener connection strength test method for a relay station, characterized by, The steps include: A1: receiving data uploaded by the electric wrench; The data is data analyzed and classified by the electric wrench based on bolt data reflecting the bolt fastener connection process and state during the action process and the bolt connection strength collected in real time by the sensor; The classification is one of the on-site response events judged and fed back by the electric wrench in real time, the delay evaluation events relying on background in-depth calculation and analysis, and the semi-response events found by the electric wrench but needing background assistance to confirm and handle; If a single action meets multiple event conditions, the on-site response event judged and fed back by the electric wrench in real time is given priority, and the rest is temporarily invalid; A2: preprocessing the data in A1; The preprocessing includes preliminary feedback response to the delay evaluation events and semi-response events, and the preliminary feedback response means that the relay station processes events with processing capacity and feeds back the feedback information to the electric wrench end; A3: priority evaluation and sorting of data to be uploaded; A4: uploading the data to be uploaded to the cloud platform in real time according to the network state according to the priority order; A5: the relay station sends feedback information to the electric wrench end according to the remote feedback information; The feedback information is information that the cloud platform receives data uploaded by the relay station, dynamically analyzes the remote feedback information, and transmits the information to the relay station.

10. A method for testing the strength of bolt fastener connections, used in a cloud platform, characterized in that, The steps include: B1: receiving data uploaded by the relay station; The data uploaded by the relay station is: the bolt data reflecting the bolt fastener connection process and state and the bolt connection strength in the action process collected by the electric wrench through the sensor in real time, which is analyzed and classified, and the priority is evaluated and sorted by the relay station, and the data to be uploaded is uploaded in real time and dynamically according to the network state according to the priority order; The classification is specifically one of the on-site response event judged and fed back by the electric wrench in real time, the delay evaluation event relying on the background in-depth calculation and analysis, and the semi-response event found by the electric wrench but needing the background to assist in confirming and processing; If a single action meets multiple event conditions at the same time, the on-site response event judged and fed back by the electric wrench in real time is given priority, and the rest is determined as temporary failure; B2: The cloud platform receives the data uploaded by the relay station, dynamically analyzes the remote feedback information, and transmits it to the relay station; The remote feedback information is sent to the electric wrench end by the relay station and is operated by the electric wrench end.

Images