A power tool operation precision control system based on AI analysis of tightening motion

Abstract

This invention discloses a precise control system for power tool operations based on AI analysis of tightening actions, belonging to the field of precise control technology for power tool operations. It includes a data acquisition layer, a model analysis layer, an execution control layer, and a quality closed-loop layer. The data acquisition layer is used to collect multi-dimensional physical quantity data of the tightening tool in real time during the tightening operation at a sampling frequency of not less than 1kHz. The model analysis layer receives the real-time multi-dimensional physical quantity data transmitted by the data acquisition layer and extracts the local gradient feature set of the torque-angle curve based on the relationship between tightening torque and time-series rotation angle. Simultaneously, a time-series prediction network model is used with the local gradient feature set and environmental parameters as input. In addition to achieving precise control of power tool operations, this invention also constructs a highly robust tightening control system with environmental perception, friction compensation, and preload prediction, and can implement multi-level anomaly interception and precise execution mechanisms.

Description

Technical Field

[0001] This invention relates to the field of precision control technology for power tool operations, and more specifically, to a precision control system for power tool operations based on AI analysis of tightening actions. Background Technology

[0002] In traditional mechanical assembly fields, such as applications with high requirements for bolt tightening processes, like the bolt tightening of automobile engines, the quality of threaded fastener tightening directly determines the product's quality and reliability. Current bolt tightening problems mainly include: missed tightening; incorrect tightening (using the wrong tightening process); and repeated tightening (re-tightening bolts that have already been tightened, reducing their lifespan).

[0003] Most existing bolt torque control schemes are based on fixed working conditions or idealized model training, failing to fully consider the nonlinear fluctuations in the friction coefficient caused by changes in contact surface morphology, lubrication layer consumption, and environmental factors such as temperature, humidity, and salt spray during repeated tightening. For example, in the tightening scenario of new energy vehicle battery packs, temperature changes lead to a decrease in frictional resistance, and traditional fixed parameter schemes cannot sense and compensate for this change in real time, resulting in a particularly prominent deviation between the predicted and actual preload force when the environment changes drastically. At the same time, although the existing third-order error compensation model incorporates modeling of factors such as friction and load changes, it still remains at the offline calibration stage and lacks a mechanism to adjust the model weights online based on real-time environmental parameters. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a precise control system for power tool operations based on AI analysis of tightening actions. In addition to achieving precise control of power tool operations, this invention also constructs a highly robust tightening control system with environmental perception, friction compensation, and preload prediction, and can implement multi-level anomaly interception and precise execution mechanisms.

[0005] To solve the above problems, the present invention adopts the following technical solution: A power tool operation precision control system based on AI analysis of tightening actions includes: a data acquisition layer, a model analysis layer, an execution control layer, and a quality closed-loop layer; The data acquisition layer is used to acquire multi-dimensional physical quantity data of the tightening tool in real time at a sampling frequency of not less than 1kHz during the tightening operation. The model parsing layer is used to receive real-time multidimensional physical quantity data transmitted by the data acquisition layer, and extract the local gradient feature set of the torque angle curve based on the relationship between tightening torque and time-series rotation angle. At the same time, a time-series prediction network model is used with the local gradient feature set and environmental parameters as input to predict the preload value under the current tightening stroke in real time. During the prediction process, the internal weight parameters of the model are dynamically adjusted according to the changes in environmental parameters to correct the torque-preload conversion deviation caused by changes in contact surface morphology, lubrication layer consumption, or nonlinear fluctuations in friction coefficient. The execution control layer is used to receive the preload prediction value output by the model analysis layer, compare the preload prediction value with the pre-stored process target threshold range, generate control commands when the predicted preload deviates from the target threshold range, and adjust the motor output torque curve and / or target speed parameters of the tightening tool in real time. When the predicted preload is within the target threshold range but the torque-angle curve shows abnormal characteristics, generate an alarm signal and lock the tool reverse unlocking program. The quality closed-loop layer is used to establish a unique digital file for each set of tightening connections. The file includes the tightening data of the connection throughout its entire life cycle. After the system completes a preset number of tightening operations, a tightening quality trend analysis report is generated based on all digital files.

[0006] Compared with the prior art, the advantages of this invention are: (1) This invention achieves real-time data representation of lubrication state and environmental variables in the power tool control system through the collaborative work of the contact surface lubrication state indirect measurement module and the multi-source environmental parameter acquisition module. On this basis, the friction coefficient change rate calculation unit in the model analysis layer quantifies the lubrication state change rate and environmental fluctuations into calculable feature inputs through formulas. The environmental compensation coefficient is calibrated through orthogonal experiments to ensure the accuracy of friction coefficient prediction. The environmental adaptive time-series preload prediction module further adopts a four-step prediction architecture of segmented normalization, DCT frequency domain mapping, environmental response weight generation and IDCT reconstruction. The frequency domain coefficient is dynamically scaled through the environmental response weight generation network, so that the frequency domain shift caused by increased friction in the high temperature and high humidity environment is compensated in real time. Finally, the preload prediction accuracy can still be maintained within ±3% within the range of ±15℃ of environmental temperature change.

[0007] (2) This invention introduces a composite control strategy of incremental PID control with anti-integral saturation and feedforward compensation of friction coefficient change rate in the execution control layer. Compared with traditional single PID control, the feedforward compensation term can respond to the trend change of friction coefficient in advance, while the PID feedback channel handles the instantaneous deviation of preload deviation. The two work together to shorten the time for preload adjustment to the target range. The abnormal curve detection module performs sliding window second derivative calculation on the torque-angle curve. The absolute value of the second derivative sliding mean can keenly capture small curve distortions. When stripping occurs, the torque drops sharply. When thread adhesion occurs, a local plateau appears on the curve. This detection method based on curvature characteristics is better than the traditional threshold method, which buys valuable time for reverse unlocking protection. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of a power tool operation precision control system based on AI analysis of tightening actions according to the present invention; Figure 2 This is a schematic diagram of the data acquisition layer module in a power tool operation precision control system based on AI analysis of tightening actions according to the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0010] Example: Please see Figure 1-2 A power tool operation precision control system based on AI analysis of tightening actions includes: a data acquisition layer, a model analysis layer, an execution control layer, and a quality closed-loop layer; The data acquisition layer is used to acquire multi-dimensional physical quantity data of the tightening tool in real time at a sampling frequency of not less than 1kHz during the tightening operation; The model parsing layer is used to receive real-time multidimensional physical quantity data transmitted from the data acquisition layer. Based on the relationship between tightening torque and timing angle, it extracts the local gradient feature set of the torque angle curve. At the same time, the timing prediction network model uses the local gradient feature set and environmental parameters as input to predict the preload value under the current tightening stroke in real time. During the prediction process, the model's internal weight parameters are dynamically adjusted according to changes in environmental parameters to correct the torque-preload conversion deviation caused by changes in contact surface morphology, lubrication layer consumption, or nonlinear fluctuations in friction coefficient. The execution control layer receives the preload prediction value output by the model parsing layer, compares the preload prediction value with the pre-stored process target threshold range, generates control commands when the predicted preload deviates from the target threshold range, and adjusts the motor output torque curve and / or target speed parameters of the tightening tool in real time. When the predicted preload is within the target threshold range but the torque-angle curve shows abnormal characteristics, an alarm signal is generated and the tool reverse unlocking program is locked. The quality closed-loop layer is used to establish a unique digital profile for each set of tightening connections. The profile includes the tightening data of the connection throughout its entire life cycle. After the system completes a preset number of tightening operations, a tightening quality trend analysis report is generated based on all digital profiles.

[0011] In a specific embodiment of the present invention, the multidimensional physical quantity data includes: tightening torque timing waveform, tightening angle timing waveform, tightening speed timing waveform, tightening axial preload signal, contact surface lubrication state characterization parameters, and environmental parameters, including ambient temperature, ambient humidity, and salt spray concentration. The local gradient feature set includes contact point identification features, yield point features, and friction coefficient change rate features; The data includes multidimensional physical quantity data, predicted preload force values, and final tightening result determination labels.

[0012] Specifically, the data acquisition layer includes a high-speed multi-channel synchronous acquisition module, an indirect measurement module for the lubrication state of the contact surface, a multi-source environmental parameter acquisition module, a preprocessing and feature extraction unit, and a data caching and transmission unit. The high-speed multi-channel synchronous acquisition module is used to synchronously acquire multi-dimensional physical quantity data of the tightening tool during the tightening operation at a sampling frequency of not less than 1kHz. The multi-dimensional physical quantity data includes the tightening torque timing waveform. Tightening angle timing waveform Tightening speed timing waveform And the axial preload signal F(t); The indirect measurement module for contact surface lubrication status is used to calculate quantitative indicators characterizing the lubrication status and the rate of change of the friction coefficient during the tightening start-up phase by comparing the initial engagement torque with the reverse release torque. ,in The initial bonding torque (N·m) is defined as the torque value in the torque angle curve that first exceeds a preset threshold. Torque reading at time The reversing loosening torque is defined as the torque value first dropping to a preset threshold during the reverse rotation process after tightening. The following is the torque reading; this module achieves automatic identification of the contact point through the linkage of a non-contact angle encoder and torque sensor, eliminating the need for manual setting of the starting position; The multi-source environmental parameter acquisition module is used to acquire environmental parameters at the tightening operation site in real time at a sampling frequency of not less than 10Hz. The module adopts a three-in-one integrated environmental sensor and is installed in the non-grip area of ​​the power tool through an air-circuit isolation protection structure to avoid interference from the tool motor heat and the operator's hand temperature on the measurement results. The preprocessing and feature extraction unit is used to perform real-time preprocessing and feature extraction on the raw data output by the high-speed multi-channel synchronous acquisition module, the indirect measurement module of contact surface lubrication state, and the multi-source environmental parameter acquisition module. The data buffer and transmission unit uses a circular buffer structure to cache the most recently used data. The complete multidimensional data frame of each sampling point, with a buffer capacity that covers at least the entire cycle of a single tightening stroke, packages the preprocessed and feature-extracted data according to a custom lightweight binary protocol and sends it to the edge computing node or model parsing layer via wired or short-range wireless communication with a transmission delay of no more than 10ms.

[0013] In a specific embodiment of the present invention, environmental parameters include ambient temperature. Ambient relative humidity and salt spray concentration ; The data frame structure of the custom binary protocol includes a 2-byte frame header identifier, an 8-byte timestamp, a 1-byte data channel number, a 2-byte data length, a variable-length payload, and a 2-byte CRC checksum.

[0014] Specifically, the preprocessing and feature extraction unit includes an outlier detection and removal submodule, a temporal alignment and resampling submodule, a normalization processing submodule, and a feature extraction submodule; The outlier detection and removal submodule employs an outlier detection algorithm based on the sliding interquartile range to perform real-time detection on the time-series data of each channel. The outlier determination criteria are as follows: ,in This represents the data value of the current sampling point. For the channel data in the sliding window The first quartile within, It is the third quartile. Interquartile range, sliding window length The number of sampling points should be no less than 100; for sampling points identified as outliers, linear interpolation should be used for replacement. , and These are the values ​​of the first normal data point before and after the outlier; The timing alignment and resampling submodule is used to perform timing alignment on multi-channel data, unifying data from different acquisition frequencies to the same time base. Linear interpolation is used for resampling, with the resampling frequency set to be no less than 1kHz. The aligned data sequence is denoted as... , where N is the total number of sampling points in a single tightening stroke; The normalization submodule is used to perform Z-score normalization on the aligned data. The normalization formula is: ,in The original data values, This represents the average value of the data from this channel over a fixed time window. The standard deviation is used to unify the data of each channel into a dimensionless space with a mean of 0 and a variance of 1. The feature extraction submodule is used to extract segmented features of the torque angle curve, time-domain statistical features, and comprehensive evaluation indicators of friction state from the preprocessed data. The segmented feature of the torque-angle curve employs a contact point recognition algorithm based on second-order derivative zero-point detection, segmenting the torque-angle curve. It is divided into a bonding section, a linear tightening section, and a yield point section. The identification conditions are ,in The preset torque gradient threshold, yield point The identification conditions are ,in The average torque gradient of the linear tightening section; Time-domain statistical features are used for torque signals and speed signal Extract the mean Maximum value Root mean square value and peak factor ; The comprehensive evaluation index of friction condition is based on the quantitative index of lubrication condition. Coupled with environmental parameters, calculate the comprehensive evaluation index of friction state. ,in , , These are the reference temperature, reference humidity, and reference salt spray concentration, respectively. , , The preset environmental compensation coefficient is obtained through orthogonal experiment calibration. This index is used to quantify the overall fluctuation of the contact surface friction coefficient due to changes in environmental factors.

[0015] In a specific embodiment of the present invention, the indirect measurement module for the lubrication state of the contact surface is further used to detect significant rise points in the torque signal in real time during the tightening start-up phase using a sliding window variance analysis method. When the torque variance within the window... Exceeding the preset threshold At that time, mark the current angle position as the starting point of contact. , ,in The mean torque within the window, and the window length. Set to 20-50 sampling points; The preprocessing and feature extraction unit is also used to determine the torque rise slope of the linear tightening section. The average upward slope of the first tightening pre-stored data. By comparing the results, the type of tightening operation can be preliminarily determined. If it is, then it is marked as a repeated tightening event, where The preset threshold for determining repeated tightening ranges from 1.3 to 1.8. If it is, then it is marked as a slipped tooth abnormal event, where The preset threshold for detecting stripped teeth is set, with a value ranging from 0.3 to 0.6. All other cases are marked as the first effective tightening event. The data buffer and transmission unit is also used to proactively send the sequence number of the last successfully transmitted data frame to the receiving end when re-establishing a connection after a communication interruption. The receiving end then requests the device to resend the missing data frame based on this sequence number. The range of the missing data frame sequence number is... ,in The sequence number of the last data frame received by the receiving end. The sequence number of the last data frame sent by the device. The retransmission process is initiated at the specified time, and the retransmitted data is taken from the data with the corresponding sequence number in the circular buffer. The high-speed multi-channel synchronous acquisition module is also used to automatically generate a synchronous trigger signal when the tightening tool motor starts by detecting the rate of change of the current signal of the motor driver, and start synchronous sampling of all acquisition channels. The sampling start delay does not exceed 0.1ms. In addition, a Butterworth low-pass filter with a cutoff frequency of 1 / 2 of the sampling frequency is set at the analog front end of each acquisition channel to filter out high-frequency noise signals higher than the Nyquist frequency.

[0016] Specifically, the model parsing layer includes a local gradient feature extraction module for torque angle curves and an environment-adaptive temporal preload prediction module; The torque angle curve local gradient feature extraction module receives real-time multidimensional physical quantity data transmitted from the data acquisition layer and extracts a local gradient feature set from the relationship between tightening torque and time-series rotation angle. This module further outputs linear segment energy features and time-domain statistical features of the tightening process as auxiliary inputs to the time-series prediction network. It reflects the potential energy stored in the elastic deformation zone and is strongly correlated with the preload. The environment-adaptive temporal preload prediction module receives local gradient feature sets and real-time environmental parameters to construct temporal feature vectors. A time-series prediction network based on frequency-domain sparse representation is used to predict the time-series feature vectors and output the predicted value of the axial preload under the current tightening stroke. In the tightening stage, the adaptive segmented normalization submodule is configured to normalize the feature values ​​of the screw-in stage and the end tightening stage respectively, based on the segmented torque-angle curve provided by the data acquisition layer, before frequency domain mapping. and The superscripts *seat* and *yield* represent the statistical extreme values ​​during the screw-in and tightening phases, respectively. This piecewise normalization eliminates the numerical scale differences caused by varying bolt lengths, resulting in higher frequency domain coefficients. The characterization of preload is more generalizable.

[0017] In a specific embodiment of the present invention, the local gradient feature set includes contact point identification features, yield point features, and friction coefficient change rate features; The time-domain statistical characteristics of the tightening process include peak torque. Root mean square of torque Peak factor and speed signal coefficient of variation It is used to quantify the stability of the tightening process.

[0018] Specifically, the local gradient feature extraction module for the torque angle curve includes a contact point identification unit, a yield point capture unit, and a friction coefficient change rate calculation unit; The contact point recognition unit is used in the torque-angle curve In this study, the angular position of the contact point was identified by jointly employing first-order derivative threshold detection and second-order derivative zero-point detection. The judgment condition is , Where T is the tightening torque. To tighten the angle, A preset torque gradient threshold is used to eliminate noise fluctuations during the idling phase; The yield point capture unit is used to identify the yield point angle position after the linear tightening segment of the torque-angle curve by tracking the degree of torque gradient decay. The judgment condition is ,in The average torque gradient of the linear tightening section. The preset yield strength coefficient ranges from 0.1 to 0.3. This unit also records the torque value at the yield point. and angle value This serves as a boundary reference for subsequent preload prediction; The friction coefficient change rate calculation unit is used to characterize the lubrication state parameters of the contact surface. Coupled with environmental parameters, the rate of change of friction coefficient during each tightening stroke is calculated. The calculation formula is: The superscripts (n) and (n-1) represent the current tightening stroke and the previous historical tightening stroke, respectively. For ambient temperature, For ambient relative humidity, Salt spray concentration, For reference environmental benchmark values, , , The preset environmental compensation coefficient was calibrated through orthogonal experiments. Used to quantify the rate of change in frictional state caused by lubrication layer consumption and environmental fluctuations.

[0019] In a specific embodiment of the present invention, the contact point recognition unit is further used to identify the contact point before the torque increases. In the low torque range, the moment when the nut or bolt begins to make contact with the connected parts is determined by sliding window variance analysis. In the window =variance within 50 First time exceeding the preset threshold At that time, mark the current point as the coarse fit point. Then A precise search for the zero point of the second derivative is then performed within a subsequent 5° range, resulting in a final contact point identification accuracy better than a 1° turn. The yield point capture unit is also used to perform Savitzky-Golay smoothing filtering on the raw torque data, with a window length of 11 points and a polynomial order, to eliminate the interference of sensor noise on gradient calculation, and to obtain the first derivative after filtering. Used for yield determination, yield determination proportionality coefficient Automatically selected based on the stiffness level of the connected components, when the average gradient of the linear segment... This is considered a hard connection. =0.3, when This is considered a soft link. =0.15, this adaptive strategy ensures the accuracy of yield point capture under different connection stiffnesses; The friction coefficient change rate calculation unit is also used to store the friction coefficient change rate of the most recent 50 consecutive tightening cycles. The sequence was processed, and a first-order low-pass filter was used to generate smooth trend values. ,when Exceeding the preset threshold 5 times consecutively When this occurs, a warning signal for the deterioration of the friction system is triggered, indicating that the lubrication layer consumption has entered a rapid change phase, and the target torque needs to be adjusted in advance.

[0020] Specifically, the execution control layer includes a preload deviation judgment module, a dynamic torque and speed adjustment module, an abnormal curve detection module, a reverse unlocking control module, and a program and sleeve interlocking module; The preload deviation determination module is used to receive the predicted axial preload value of the current tightening stroke output by the model analysis layer. Read the target preload threshold range of the current connection pair from the pre-stored process parameter database. Then the deviation was calculated. ,when hour, ,when hour, ,when When within range, It also triggers the abnormal curve detection flag and outputs the deviation amount. And a comparison flag, where flag=1 indicates that it exceeds the upper bound, flag=-1 indicates that it is below the lower bound, and flag=0 indicates that it is within the range; The dynamic torque and speed adjustment module is used to receive the output from the preload deviation determination module. The control quantity is calculated using an incremental PID controller with anti-integral saturation, along with a flag bit. ,in This is the proportional gain coefficient. This is the integral gain coefficient. The differential gain coefficient is used to resist integral saturation using a clamping method. The total control quantity is calculated as follows. Reaching the physical limit of the actuator and When the sign of the integral is consistent with the direction of the amplitude limit, the accumulation of the error by the integral term immediately stops. Simultaneously, this module also receives the rate of change of the friction coefficient output from the model's analytical layer. Add feedforward compensation term ,in The feedforward gain coefficient is the sum of the feedforward compensation term and the PID output to form the total control quantity. ; The abnormal curve detection module is used to detect the real-time acquired torque-angle curve when flag=0 and the preload prediction value is within the target threshold range. Perform second derivative calculation using a sliding window ,in Let i be the angular resolution and i be the sampling point number. In the non-fitting region, calculate the moving average of the absolute value of the second derivative. ,when Exceeding the preset threshold When this occurs, it is determined to be an abnormal torque-angle curve. Abnormality types include stripping, thread adhesion, or abnormal yielding of the connector. The reverse unlock control module receives the abnormal judgment signal from the abnormal curve detection module, generates an alarm signal, and displays it on the tool body and production line terminal. Simultaneously, it locks the tool's forward start and switches the tool status to reverse unlock mode. In reverse unlock mode, it controls the motor to rotate in the reverse direction. The reverse speed curve uses trapezoidal acceleration and deceleration control. The reverse stop condition is when the cumulative reverse angle reaches a preset value. Or the reverse torque is detected to be below the threshold. ; The program and sleeve interlock module is used to read the electronic tag of the current sleeve using an RFID reader / writer installed on the sleeve base before the tightening operation begins. This obtains the sleeve type code and size parameters, and matches the sleeve information with the sleeve specifications specified in the current tightening program issued from top to bottom. If the match fails, an interlock signal is generated, preventing the motor from starting and prompting the operator to replace the sleeve. At the same time, the cumulative number of uses for each sleeve is recorded. ,when Exceeding the preset lifespan threshold A replacement reminder will be triggered at any time.

[0021] In a specific embodiment of the present invention, in the dynamic torque and speed adjustment module, the total control quantity The output is filtered by a first-order low-pass filter. ,in These are the filter coefficients. To control the cycle, the filtered control quantity is used to generate torque and speed adjustment commands to avoid parameter jumps from impacting the tightening process; The anomaly curve detection module is also used for second derivative sequences. A median filter with a length of 5 points is applied to eliminate interference from high-frequency noise from the sensor; when an anomaly is detected, the angle range in which the anomaly occurred is recorded simultaneously. The corresponding torque value serves as the basis for subsequent quality traceability; The reverse unlock control module is also used to monitor the motor current in real time when the tool is in reverse unlock mode. If the current exceeds the overload protection threshold Then temporarily reduce the reverse rotation speed to Continue running the motor at its original speed until the current returns to normal, in order to prevent the motor from overheating and being damaged. The program and sleeve interlocking module also obtains the current spatial attitude information of the tool through the nine-axis inertial measurement unit built into the tightening tool, including roll angle, pitch angle and yaw angle, and compares the real-time attitude with the pre-stored standardized working attitude. When any attitude angle deviates from the standard value by more than the set threshold of roll angle ±15° or pitch angle ±20°, an attitude abnormality alarm signal is generated, and the tightening action is prohibited from starting or the current tightening process is stopped immediately.

[0022] Specifically, the torque adjustment formula in the dynamic torque and speed adjustment module is as follows: ,in The target torque for the process is expressed in N·m. The torque conversion factor is in units of When the deviation exceeds the upper limit, the target torque is reduced; when it exceeds the lower limit, the target torque is increased. The speed adjustment formula in the dynamic torque and speed adjustment module is as follows: ,in The target rotational speed for the process is expressed in r / min. The unit of the speed conversion factor is... When the preload is too high, reduce the rotation speed to reduce impact energy; when it is too low, increase the rotation speed to improve preload efficiency.

[0023] In a specific embodiment of the present invention, in the dynamic torque and speed adjustment module, when the preload deviation... When the value is zero, the PID controller output automatically returns to zero, but the feedforward compensation term... Still based on the rate of change of friction coefficient Continuous output to maintain stable preload even with slow environmental changes.

[0024] Specifically, the quality closed-loop layer includes a digital archive creation module, a trend analysis report generation module, an anomaly warning push module, and a traceability query module; The digital record creation module generates a unique record identifier for each tightening connection pair and stores the multidimensional physical quantity data of that connection pair in each tightening operation, as well as the preload prediction value output by the model analysis layer. The final tightening result determination label and environmental parameters are archived in the order of tightening completion timestamps to the storage location corresponding to the file identification code, forming a tightening data record covering the entire life cycle of the connection pair. The file identification code is generated by assigning the workstation number where the connection pair is located to the file identification code. Bolt position number and tightening completion timestamp The strings are concatenated, and after cyclic redundancy check, the lower 16 bits are taken and combined with the higher bits of the original string to form a 32-bit hexadecimal code, ensuring that the file identification code of the same connection pair has time distinguishability in different operations; The trend analysis report generation module is used to extract the predicted preload value sequence for each tightening operation from all digital archives after the system has completed a preset number of K tightening operations. Calculate the statistical characteristic value of the sequence, compare the statistical characteristic value with the preset quality control limit, and generate a quality trend analysis report containing trend judgment conclusions, outlier markers and quality improvement suggestions based on the comparison results; The anomaly warning push module is used to automatically generate an early warning task sheet when the trend analysis report generation module determines that there are points exceeding the control limits or trend anomalies. The task sheet includes the anomaly type, the connection sub-file identifier code corresponding to the anomaly point, the time of anomaly occurrence, and the preload deviation. The early warning task sheet is pushed to the handheld terminal of the production line manager and the workshop dashboard system through the Industrial Internet of Things protocol, with a push delay of no more than 30 seconds. The traceability query module receives the connector file identification code or product serial number input by the user, retrieves the corresponding full life cycle data record from the digital file database, displays the preload value curve, judgment label and operator ID of each tightening operation of the connector in a timeline format, and displays a highlighted mark at abnormal operation points. It supports outputting a quality history report of the connector, which includes the number of tightening operations, the date of first tightening, the date of last tightening, the average preload, the standard deviation, and a summary of each abnormal event.

[0025] In a specific embodiment of the present invention, the digital archive creation module is further used to perform lossy compression on the raw torque-angle curve data of each tightening operation. The compression method involves extracting feature points from the torque-angle curve at equal intervals, with an extraction interval of 1° angle step, and applying second-order differential encoding to the extracted sequence. The difference values ​​are stored after Huffman coding with a compression ratio of no less than 5:1; at the same time, the key feature values ​​of the curve are stored as lossless indexes. The trend analysis report generation module is also used to calculate the capability index of the tightening process. ,in and These are the upper and lower specification limits for process design, respectively, and s is the sample standard deviation. When the value is less than 1.33, process optimization suggestions are automatically generated in the trend report, indicating that the target torque needs to be adjusted or the friction coefficient fluctuation needs to be reduced.

[0026] Specifically, the trend analysis report generation module includes a statistical characteristic calculation submodule, a quality control limit calculation submodule, a trend anomaly judgment submodule, and a report generation submodule; The statistical characteristics calculation submodule is used to calculate the sample mean, sample standard deviation, and moving range; The quality control limit calculation submodule is used to calculate the upper limit of the preload value. and control lower limit ,in As a control chart constant, when a single tightening is a single measurement and the range of movement is calculated based on two adjacent points, take [value]. The control limit is used to identify abnormal tightening events that exceed the normal fluctuation range; The trend anomaly detection submodule is used to analyze the sequence. Perform linear regression and calculate the slope. , The unit is kN / cycle, when When it is determined that there is a monotonically increasing or decreasing trend, among which The preset trend threshold ranges from 0.01 to 0.05 kN / time. When C consecutive points... All located in When it is on the same side, it is judged as an abnormal chain length and is also marked as an abnormal trend. The report generation submodule is used to format the above statistical characteristic values, control limits, outlier locations, and trend judgment results into a document containing data tables, control charts, and textual conclusions. The control chart uses the tightening number i as the horizontal axis and the preload value as the vertical axis. Use the vertical axis as the vertical axis and draw the center line. Control upper limit and control lower limit Points exceeding the control limits are marked in red.

[0027] In a specific embodiment of the present invention, the sample mean The unit is kN; Sample standard deviation The unit is kN; Motion range The unit is kN, and its average value is .

[0028] Specifically, the preset number of jobs K in the trend analysis report generation module is dynamically determined according to the following rules: When the production line is in batch production mode, K=50; When the production line operates on a small-batch, multi-variety model, K=20; The range of movement of the predicted preload value after 5 consecutive tightening cycles When all values ​​are less than 0.1kN, trend analysis is triggered in advance, and K is the actual number of times completed, with K being no less than 10 times.

[0029] In a specific embodiment of the present invention, the dynamic K-value strategy solves the adaptability problem of the traditional fixed sample size strategy under different production line modes. In batch mode, it avoids statistical bias caused by too few samples, and in small batch mode, it avoids response lag caused by too many samples. The early triggering mechanism enables rapid early warning when abnormal process signs appear, making trend analysis more targeted and timely, and significantly improving the flexibility and intelligence level of quality closed-loop control.

[0030] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A power tool operation precision control system based on AI analysis of tightening actions, characterized in that, include: Data acquisition layer, model parsing layer, execution control layer, and quality closed-loop layer; The data acquisition layer is used to acquire multi-dimensional physical quantity data of the tightening tool in real time at a sampling frequency of not less than 1kHz during the tightening operation. The model parsing layer is used to receive real-time multidimensional physical quantity data transmitted by the data acquisition layer, and extract the local gradient feature set of the torque angle curve based on the relationship between tightening torque and time-series rotation angle. At the same time, a time-series prediction network model is used with the local gradient feature set and environmental parameters as input to predict the preload value under the current tightening stroke in real time. During the prediction process, the internal weight parameters of the model are dynamically adjusted according to the changes in environmental parameters to correct the torque-preload conversion deviation caused by changes in contact surface morphology, lubrication layer consumption, or nonlinear fluctuations in friction coefficient. The execution control layer is used to receive the preload prediction value output by the model analysis layer, compare the preload prediction value with the pre-stored process target threshold range, generate control commands when the predicted preload deviates from the target threshold range, and adjust the motor output torque curve and / or target speed parameters of the tightening tool in real time. When the predicted preload is within the target threshold range but the torque-angle curve shows abnormal characteristics, generate an alarm signal and lock the tool reverse unlocking program. The quality closed-loop layer is used to establish a unique digital file for each set of tightening connections. The file includes the tightening data of the connection throughout its entire life cycle. After the system completes a preset number of tightening operations, a tightening quality trend analysis report is generated based on all digital files.

2. The power tool operation precision control system based on AI analysis of tightening actions according to claim 1, characterized in that, The data acquisition layer includes a high-speed multi-channel synchronous acquisition module, a contact surface lubrication state indirect measurement module, a multi-source environmental parameter acquisition module, a preprocessing and feature extraction unit, and a data caching and transmission unit. The high-speed multi-channel synchronous acquisition module is used to synchronously acquire multi-dimensional physical quantity data of the tightening tool at a sampling frequency of not less than 1kHz during the tightening operation. The multi-dimensional physical quantity data includes the tightening torque timing waveform. Tightening angle timing waveform Tightening speed timing waveform And the axial preload signal F(t); The indirect measurement module for the lubrication state of the contact surface is used to calculate quantitative indicators characterizing the lubrication state and the rate of change of the friction coefficient by comparing the initial engagement torque with the reverse release torque during the tightening start-up phase. ,in The initial bonding torque (N·m) is defined as the torque value in the torque angle curve that first exceeds a preset threshold. Torque reading at time The reversing loosening torque is defined as the torque value first dropping to a preset threshold during the reverse rotation process after tightening. The following torque readings are obtained by using a non-contact angle encoder linked with a torque sensor to automatically identify the contact point, eliminating the need for manual setting of the starting position; The multi-source environmental parameter acquisition module is used to acquire environmental parameters of the tightening operation site in real time at a sampling frequency of not less than 10Hz. The module adopts a three-in-one integrated environmental sensor and is installed in the non-grip area of ​​the power tool through an air-circuit isolation protection structure to avoid interference from the tool motor heat and the operator's hand temperature on the measurement results. The preprocessing and feature extraction unit is used to perform real-time preprocessing and feature extraction on the raw data output by the high-speed multi-channel synchronous acquisition module, the contact surface lubrication state indirect measurement module, and the multi-source environmental parameter acquisition module. The data caching and transmission unit employs a circular buffer structure to cache the most recently accessed data. The complete multidimensional data frame of each sampling point, with a buffer capacity that covers at least the entire cycle of a single tightening stroke, packages the preprocessed and feature-extracted data according to a custom lightweight binary protocol and sends it to the edge computing node or model parsing layer via wired or short-range wireless communication with a transmission delay of no more than 10ms.

3. The power tool operation precision control system based on AI analysis of tightening actions according to claim 2, characterized in that, The preprocessing and feature extraction unit includes an outlier detection and removal submodule, a temporal alignment and resampling submodule, a normalization processing submodule, and a feature extraction submodule. The outlier detection and removal submodule employs an outlier detection algorithm based on sliding interquartile range (IIR) to perform real-time detection on the time-series data of each channel. The outlier determination criteria are as follows: ,in This represents the data value of the current sampling point. For the channel data in the sliding window The first quartile within, It is the third quartile. Interquartile range, sliding window length The number of sampling points should be no less than 100; for sampling points identified as outliers, linear interpolation should be used for replacement. , and These are the values ​​of the first normal data point before and after the outlier; The timing alignment and resampling submodule is used to perform timing alignment on multi-channel data, unifying data from different acquisition frequencies to the same time base, and resampling using linear interpolation. The resampling frequency is set to be no less than 1kHz. The aligned data sequence is denoted as follows. , where N is the total number of sampling points in a single tightening stroke; The normalization submodule is used to perform Z-score standardization on the aligned data. The standardization formula is: ,in The original data values, This represents the average value of the data from this channel over a fixed time window. The standard deviation is used to unify the data of each channel into a dimensionless space with a mean of 0 and a variance of 1. The feature extraction submodule is used to extract segmented features of the torque angle curve, time-domain statistical features, and comprehensive evaluation indicators of friction state from the preprocessed data. The segmented features of the torque-angle curve employ a fitting point recognition algorithm based on second-order derivative zero-point detection to segment the torque-angle curve. It is divided into a bonding section, a linear tightening section, and a yield point section. The identification conditions are ,in The preset torque gradient threshold, yield point The identification conditions are ,in The average torque gradient of the linear tightening section; The time-domain statistical features are used for torque signals. and speed signal Extract the mean Maximum value Root mean square value and peak factor ; The comprehensive evaluation index of friction state is based on the quantitative index of lubrication state. Coupled with environmental parameters, calculate the comprehensive evaluation index of friction state. ,in , , These are the reference temperature, reference humidity, and reference salt spray concentration, respectively. , , The preset environmental compensation coefficient is obtained through orthogonal experiment calibration. This index is used to quantify the overall fluctuation of the contact surface friction coefficient due to changes in environmental factors.

4. The power tool operation precision control system based on AI analysis of tightening actions according to claim 1, characterized in that, The model parsing layer includes a torque angle curve local gradient feature extraction module and an environment-adaptive temporal preload prediction module; The torque angle curve local gradient feature extraction module is used to receive real-time multidimensional physical quantity data transmitted from the data acquisition layer, and extract a local gradient feature set from the relationship between tightening torque and time-series rotation angle. The torque angle curve local gradient feature extraction module further outputs linear segment energy features and time-domain statistical features of the tightening process as auxiliary inputs to the time-series prediction network. It reflects the potential energy stored in the elastic deformation zone and is strongly correlated with the preload. The environment-adaptive temporal preload prediction module is used to receive the local gradient feature set and real-time environmental parameters to construct a temporal feature vector. A time-series prediction network based on frequency-domain sparse representation is used to predict the time-series feature vector and output the predicted value of axial preload under the current tightening stroke. In the tightening stage, the adaptive segmented normalization submodule is configured to normalize the feature values ​​of the screw-in stage and the end tightening stage respectively, based on the segmented torque-angle curve provided by the data acquisition layer, before frequency domain mapping. and The superscripts *seat* and *yield* represent the statistical extreme values ​​during the screw-in and tightening phases, respectively. This piecewise normalization eliminates the numerical scale differences caused by varying bolt lengths, resulting in higher frequency domain coefficients. The characterization of preload is more generalizable.

5. The power tool operation precision control system based on AI analysis of tightening actions according to claim 4, characterized in that, The local gradient feature extraction module for the torque angle curve includes a contact point identification unit, a yield point capture unit, and a friction coefficient change rate calculation unit. The contact point recognition unit is used in the torque-angle curve In this study, the angular position of the contact point was identified by jointly employing first-order derivative threshold detection and second-order derivative zero-point detection. The judgment condition is , Where T is the tightening torque. To tighten the angle, A preset torque gradient threshold is used to eliminate noise fluctuations during the idling phase; The yield point capturing unit is used to identify the yield point angle position after the linear tightening segment of the torque-angle curve by tracking the degree of torque gradient decay. The judgment condition is ,in The average torque gradient of the linear tightening section. The preset yield strength coefficient ranges from 0.1 to 0.

3. This unit also records the torque value at the yield point. and angle value This serves as a boundary reference for subsequent preload prediction; The friction coefficient change rate calculation unit is used to characterize the lubrication state parameters of the contact surface. Coupled with environmental parameters, the rate of change of friction coefficient during each tightening stroke is calculated. The calculation formula is: The superscripts (n) and (n-1) represent the current tightening stroke and the previous historical tightening stroke, respectively. For ambient temperature, For ambient relative humidity, Salt spray concentration, For reference environmental benchmark values, , , The preset environmental compensation coefficient was calibrated through orthogonal experiments. Used to quantify the rate of change in frictional state caused by lubrication layer consumption and environmental fluctuations.

6. The power tool operation precision control system based on AI analysis of tightening actions according to claim 1, characterized in that, The execution control layer includes a preload deviation determination module, a dynamic torque and speed adjustment module, an abnormal curve detection module, a reverse unlocking control module, and a program and sleeve interlocking module. The preload deviation determination module is used to receive the predicted axial preload value of the current tightening stroke output by the model analysis layer. Read the target preload threshold range of the current connection pair from the pre-stored process parameter database. Then the deviation was calculated. ,when hour, ,when hour, ,when When within range, It also triggers the abnormal curve detection flag and outputs the deviation amount. And a comparison flag, where flag=1 indicates that it exceeds the upper bound, flag=-1 indicates that it is below the lower bound, and flag=0 indicates that it is within the range; The dynamic torque and speed adjustment module is used to receive the output from the preload deviation determination module. The control quantity is calculated using an incremental PID controller with anti-integral saturation, along with a flag bit. ,in This is the proportional gain coefficient. This is the integral gain coefficient. The differential gain coefficient is used to resist integral saturation by employing a clamping method to calculate the total control quantity. Reaching the physical limit of the actuator and When the sign of the integral is consistent with the direction of the amplitude limit, the accumulation of the error by the integral term immediately stops. Simultaneously, this module also receives the rate of change of the friction coefficient output from the model's analytical layer. Add feedforward compensation term ,in The feedforward gain coefficient is the sum of the feedforward compensation term and the PID output to form the total control quantity. ; The abnormal curve detection module is used to detect the real-time collected torque-angle curve when flag=0 and the preload prediction value is within the target threshold range. Perform second derivative calculation using a sliding window ,in Let i be the angular resolution and i be the sampling point number. In the non-fitting region, calculate the moving average of the absolute value of the second derivative. ,when Exceeding the preset threshold When this occurs, it is determined to be an abnormal torque-angle curve. Abnormality types include stripping, thread adhesion, or abnormal yielding of the connector. The reverse unlock control module receives the abnormal judgment signal from the abnormal curve detection module, generates an alarm signal, and displays it on the tool body and production line terminal. Simultaneously, it locks the tool's forward start and switches the tool status to reverse unlock mode. In reverse unlock mode, it controls the motor to rotate in the reverse direction, using a trapezoidal acceleration / deceleration control for the reverse speed curve. The reverse stop condition is when the cumulative reverse angle reaches a preset value. Or the reverse torque is detected to be below the threshold. ; The program and sleeve interlock module are used to read the electronic tag of the current sleeve by an RFID reader / writer installed on the sleeve base before the tightening operation begins, obtain the sleeve type code and size parameters, and match the sleeve information with the sleeve specifications specified in the current tightening procedure issued from top to bottom. If the match fails, an interlock signal is generated to prevent the motor from starting and prompt the operator to replace the sleeve. At the same time, the cumulative number of uses of each sleeve is recorded. ,when Exceeding the preset lifespan threshold A replacement reminder will be triggered at any time.

7. The power tool operation precision control system based on AI analysis of tightening actions according to claim 6, characterized in that, The torque adjustment formula in the dynamic torque and speed adjustment module is as follows: ,in The target torque for the process is expressed in N·m. The torque conversion factor is in units of When the deviation exceeds the upper limit, the target torque is reduced; when it exceeds the lower limit, the target torque is increased. The speed adjustment formula in the dynamic torque and speed adjustment module is as follows: ,in The target rotational speed for the process is expressed in r / min. The unit of the speed conversion factor is... When the preload is too high, reduce the rotation speed to reduce impact energy; when it is too low, increase the rotation speed to improve preload efficiency.

8. The power tool operation precision control system based on AI analysis of tightening actions according to claim 1, characterized in that, The quality closed-loop layer includes a digital archive creation module, a trend analysis report generation module, an anomaly early warning push module, and a traceability query module; The digital record creation module is used to generate a unique record identifier for each set of tightening connections, and to store the multi-dimensional physical quantity data of the connection in each tightening operation, as well as the preload prediction value output by the model analysis layer. The final tightening result determination label and environmental parameters are archived in the order of tightening completion timestamps to the storage location corresponding to the file identification code, forming a tightening data record covering the entire life cycle of the connection pair. The file identification code is generated by assigning the workstation number of the connection pair to the file identification code. Bolt position number and tightening completion timestamp The strings are concatenated, and after cyclic redundancy check, the lower 16 bits are taken and combined with the higher bits of the original string to form a 32-bit hexadecimal code, ensuring that the file identification code of the same connection pair has time distinguishability in different operations; The trend analysis report generation module is used to extract the predicted preload value sequence of each tightening operation from all digital archives after the system has completed a preset number K tightening operations. Calculate the statistical characteristic value of the sequence, compare the statistical characteristic value with the preset quality control limit, and generate a quality trend analysis report containing trend judgment conclusions, outlier markers and quality improvement suggestions based on the comparison results; The abnormality warning push module is used to automatically generate a warning task sheet when the trend analysis report generation module determines that there is a point exceeding the control limit or an abnormal trend. The task sheet includes the abnormality type, the connection sub-file identifier code corresponding to the abnormal point, the time of abnormality occurrence and the preload deviation. The warning task sheet is pushed to the handheld terminal of the production line manager and the workshop dashboard system through the industrial Internet of Things protocol, with a push delay of no more than 30 seconds. The traceability query module is used to receive the connector file identification code or product serial number input by the user, retrieve the corresponding full life cycle data record from the digital file database, display the preload value curve, judgment label and operator ID of each tightening operation of the connector in a timeline manner, and display a highlighted mark at abnormal operation points. It supports outputting a quality history report of the connector, which includes the number of tightening operations, the date of first tightening, the date of last tightening, the average preload, the standard deviation, and a summary of each abnormal event.

9. A power tool operation precision control system based on AI analysis of tightening actions according to claim 8, characterized in that, The trend analysis report generation module includes a statistical feature calculation submodule, a quality control limit calculation submodule, a trend anomaly determination submodule, and a report generation submodule. The statistical characteristic calculation submodule is used to calculate the sample mean, sample standard deviation, and moving range; The quality control limit calculation submodule is used to calculate the upper limit of the preload value. and control lower limit ,in As a control chart constant, when a single tightening is a single measurement and the range of movement is calculated based on two adjacent points, take [value]. The control limit is used to identify abnormal tightening events that exceed the normal fluctuation range; The trend anomaly determination submodule is used to analyze the sequence. Perform linear regression and calculate the slope. , The unit is kN / cycle, when When it is determined that there is a monotonically increasing or decreasing trend, among which The preset trend threshold ranges from 0.01 to 0.05 kN / time. When C consecutive points... All located in When it is on the same side, it is judged as an abnormal chain length and is also marked as an abnormal trend. The report generation submodule is used to format the above statistical characteristic values, control limits, outlier locations, and trend determination results into a document containing data tables, control chart images, and textual conclusions. The control chart image uses the tightening sequence number i as the horizontal axis and the preload value as the vertical axis. Use the vertical axis as the vertical axis and draw the center line. Control upper limit and control lower limit Points exceeding the control limits are marked in red.

10. A power tool operation precision control system based on AI analysis of tightening actions according to claim 8, characterized in that, The preset number of tasks K in the trend analysis report generation module is dynamically determined according to the following rules: When the production line is in batch production mode, K=50; When the production line operates on a small-batch, multi-variety model, K=20; The range of movement of the predicted preload value after 5 consecutive tightening cycles When all values ​​are less than 0.1kN, trend analysis is triggered in advance, and K is the actual number of times completed, with K being no less than 10 times.

Images