Bolt tightening process parameter self-adaptive correction method based on algorithm analysis

Abstract

The present application belongs to the technical field of mechanical assembly, and discloses a bolt tightening process parameter self-adaptive correction method based on algorithm analysis. The high-frequency alternating oscillation section is identified through local change rate sign analysis, the oscillation main period is extracted to construct a self-adaptive filtering window model. Then, the torque signal is weighted and smoothed, and phase lead compensation is implemented to generate a de-oscillation torque trend signal. Further, the equivalent stiffness value is calculated through segmented energy integral, and the current stiffness value is obtained by combining trend extrapolation. At the same time, the statistical reference distribution of historical data is constructed, and the confidence index is generated by combining the oscillation intensity evaluation. Finally, the target torque and speed parameters are dynamically adjusted according to the confidence. The present application realizes high consistency between signal processing and actual physical phenomena, balances the system response speed and stability, improves the stiffness calculation accuracy and tightening quality consistency, and reduces the misjudgment rate and the frequency of manual intervention.

Description

Technical Field

[0001] This invention relates to the field of mechanical assembly technology, and more specifically, to an adaptive correction method for bolt tightening process parameters based on algorithm analysis. Background Technology

[0002] Existing bolt tightening control technologies exhibit complex nonlinear behavior in the microscopic frictional characteristics of the threaded contact surface during actual tightening conditions. This is particularly problematic at lower tightening speeds or when the contact surface material is uneven. The bolt-nut interface frequently switches between adhesive and sliding states, producing a characteristic "squeaking" sound. This physical phenomenon manifests as high-frequency sawtooth fluctuations in the torque signal. Although the torque shows a steady increase from a macroscopic perspective, the amplitude of these microscopic oscillations can be a significant proportion of the instantaneous torque value. This oscillation severely interferes with traditional tightening control systems, especially when adaptive control algorithms require real-time calculation of the torque derivative with respect to rotation angle to assess connection stiffness or monitor changes in the friction coefficient. Differential operations inherently amplify high-frequency noise, transforming the previously acceptable sawtooth fluctuations into sharp jumps after derivative calculation, leading to algorithmic misjudgments of the system state. In practical applications, this derivative distortion can cause the control system to incorrectly assume the bolt has entered the yielding stage or that the connection has broken abnormally, triggering unnecessary warnings or emergency shutdowns, or even prematurely terminating the tightening process or incorrectly reducing the target torque value. More seriously, existing filtering techniques often employ fixed parameter settings, failing to adaptively adjust to the diverse oscillation characteristics exhibited by different material combinations, surface treatment processes, and lubrication conditions. In high-precision assembly fields such as aerospace, automotive transmission systems, and precision instrument manufacturing, this instability in stiffness estimation directly impacts product quality reliability and consistency, even leading to a significant decrease in on-site assembly efficiency. This is because technicians are forced to frequently intervene and manually adjust tightening parameters to address system misjudgments. Furthermore, traditional methods neglect the time asymmetry of viscous and sliding phases when processing oscillating signals. Using symmetrical filters to handle inherently asymmetrical physical phenomena further exacerbates the bias in stiffness calculations, especially at critical turning points in the tightening process, where such bias can lead to catastrophic judgment errors.

[0003] In view of this, the present invention proposes an adaptive correction method for bolt tightening process parameters based on algorithm analysis to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an adaptive correction method for bolt tightening process parameters based on algorithm analysis, comprising: Acquire real-time sampling data of torque and rotation signals during the bolt tightening process; Local rate of change sign analysis is performed on the torque signal to identify segments with high-frequency alternation of rate of change signs, which are marked as viscous-sliding oscillation segments; Based on the zero-crossing time series characteristics of the viscous-slip oscillation segment, the main oscillation period is extracted, and an adaptive filtering window model is constructed. An adaptive filtering window model is used to perform weighted smoothing on the torque signal, and phase lead compensation is performed on the smoothing result to generate a de-oscillating torque trend signal. The piecewise energy integration operation is performed on the de-oscillation torque trend signal and the rotation angle signal to obtain the equivalent stiffness value of each segment interval, and the current tightening stiffness value is obtained by extrapolation through the stiffness change trend. Construct a stiffness statistical reference distribution based on historical tightening data, calculate the standardized deviation of the current tightening stiffness value relative to the stiffness statistical reference distribution, and generate a confidence index. Based on the confidence index, a stiffness acceptance strategy is set, and the target torque correction amount and tightening speed adjustment amount are dynamically calculated in combination with the current tightening stiffness value, and the corrected tightening process parameters are output.

[0005] The technical effects and advantages of the adaptive correction method for bolt tightening process parameters based on algorithm analysis in this invention are as follows: This invention, through intelligent identification and processing of oscillation characteristics, can accurately grasp the true tightening trend amidst interference-filled signals. This high adaptability significantly extends the lifespan of connectors, avoiding loosening failure due to insufficient tightening or thread damage caused by overtightening. In terms of production efficiency, this invention greatly reduces the misjudgment rate and the frequency of manual intervention, eliminating the need for operators to frequently pause production lines to adjust parameters as in traditional methods. In high-volume applications, this invention reduces scrap rates and rework rates, creating considerable economic benefits for enterprises. More importantly, the self-learning characteristic of this invention allows it to adapt to different material combinations, surface treatments, and lubrication conditions without the need for manual parameter recalibration. Furthermore, the intelligent early warning mechanism of this invention can identify abnormal operating conditions early, preventing problems before they occur and providing valuable real-time quality monitoring tools for production management, enabling a shift from a passive response to a proactive prevention management model. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the adaptive correction method for bolt tightening process parameters based on algorithm analysis according to the present invention. Detailed Implementation

[0007] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0008] Please see Figure 1 In this embodiment of the invention, the specific implementation process of the adaptive correction method for bolt tightening process parameters based on algorithm analysis includes: Real-time sampling data of torque and rotation angle signals are acquired during the bolt tightening process. The torque signal reflects the change in applied torque force during tightening and is a key indicator of bolt tightening quality; the rotation angle signal records the cumulative change in the bolt's rotation angle. Together, they constitute the basic data for the tightening process. These signals are acquired in real time using high-precision torque sensors and angle encoders, with the sampling frequency typically set between 1000Hz and 2000Hz to ensure the capture of high-frequency oscillation characteristics. The acquired raw data contains all physical characteristics of the tightening process, including torque fluctuations, frictional characteristics of the bolt-nut contact surface, and information on material elastic-plastic deformation. This provides complete physical field information for subsequent accurate analysis and serves as the basic data input for adaptive correction of bolt tightening process parameters.

[0009] Local rate-of-change sign analysis was performed on the torque signal to identify segments with high-frequency alternation of rate-of-change signs, which were marked as viscous-slip oscillation segments. Viscous-slip oscillation is a typical physical phenomenon in bolt tightening, originating from changes in the frictional characteristics of the threaded contact surface. When the contact surface is in a critical friction state, it frequently switches between viscous and slip states, causing characteristic oscillations in the torque signal. By identifying this oscillation characteristic, key physical behaviors during tightening can be accurately captured, providing important basis for subsequent signal processing and parameter correction. Accurate identification of oscillation segments is a fundamental step in ensuring tightening quality and is crucial for preventing quality problems such as stripping, over-tightening, and under-tightening.

[0010] Based on the zero-crossing time-series characteristics of the viscous-slip oscillation region, the dominant oscillation period is extracted, and an adaptive filtering window model is constructed. The dominant oscillation period is a key characteristic of the viscous-slip phenomenon, directly reflecting the dynamic response characteristics of the system. By analyzing the zero-crossing time-series characteristics of the torque change rate within the oscillation region, the frequency characteristics of the oscillation can be accurately extracted, providing a scientific basis for subsequent signal filtering. The adaptive filtering window model is dynamically adjusted based on the extracted oscillation characteristics, ensuring that the filtering effect can effectively suppress oscillation interference while retaining the true trend information of the tightening process. This adaptive filtering strategy based on physical characteristics is significantly superior to traditional fixed-parameter filtering methods, and can automatically optimize the filtering parameters according to the characteristics of different bolts and different tightening stages.

[0011] An adaptive filtering window model is used to perform weighted smoothing on the torque signal, and phase lead compensation is applied to the smoothing result to generate a de-oscillating torque trend signal. Weighted smoothing is an effective means of suppressing high-frequency oscillations. By performing local weighted averaging on the original torque signal through an adaptive filtering window, the influence of oscillating components is significantly reduced. However, the filtering process inevitably introduces phase delay, which affects the accuracy of subsequent stiffness calculations. To address this issue, this method introduces a phase lead compensation mechanism. Through group delay analysis and corresponding time-domain forward shift operations, the phase lag caused by filtering is effectively eliminated, ensuring the time alignment of the processed signal. The de-oscillating torque trend signal retains the true mechanical characteristics of the tightening process while effectively suppressing the oscillating components that interfere with the analysis, laying the foundation for subsequent stiffness calculations.

[0012] Piecewise energy integration is performed on the de-oscillation torque trend signal and the angle signal to obtain the equivalent stiffness value of each segment interval. The current tightening stiffness value is then obtained by extrapolating the stiffness change trend. Piecewise energy integration is a core technology for evaluating the dynamic stiffness characteristics of a bolt tightening system. By calculating the integral value of torque with respect to angle, the system's energy absorption capacity can be accurately reflected. The equivalent stiffness value represents the comprehensive stiffness characteristics of the bolt-nut-connector system at different tightening stages and is a key parameter for predicting the target torque. By performing trend analysis and extrapolation on the equivalent stiffness of multiple consecutive segments, the real-time stiffness value under the current tightening state can be predicted, providing a scientific basis for torque prediction and process parameter correction. This stiffness estimation method based on energy integration is superior to traditional simple slope calculation, effectively suppressing the influence of local fluctuations and providing more stable and reliable stiffness assessment results.

[0013] A stiffness statistical reference distribution based on historical tightening data is constructed. The standardized deviation of the current tightening stiffness value relative to the stiffness statistical reference distribution is calculated, generating a confidence index. The stiffness statistical reference distribution is the foundation of the system's adaptive capability. By collecting and analyzing historical tightening data, the statistical distribution characteristics of stiffness values ​​at specific corner positions are established, providing a reference standard for the current tightening process. The standardized deviation quantifies the degree of difference between the current tightening stiffness and historical reference data, and is an important indicator for evaluating the stability of the tightening process. The confidence index further considers the impact of oscillation intensity on the reliability of stiffness estimation, generating a final confidence score through comprehensive evaluation. This dual evaluation mechanism based on historical data and real-time analysis significantly improves the system's ability to identify and adapt to abnormal operating conditions.

[0014] The system sets a stiffness acceptance strategy based on a confidence level index, dynamically calculates the target torque correction and tightening speed adjustment based on the current tightening stiffness value, and outputs the corrected tightening process parameters. The stiffness acceptance strategy is the core of the system's decision-making mechanism, dynamically adjusting the adoption method of the current stiffness estimate based on the confidence level index. Under high confidence conditions, the system directly uses the current stiffness value for parameter calculation; under low confidence conditions, different response strategies are adopted based on the duration to ensure system stability and reliability. The target torque correction and tightening speed adjustment are the final output control parameters. The former ensures that the expected tightening force is achieved, while the latter suppresses increased oscillation and optimizes the dynamic characteristics of the tightening process by adjusting the tightening speed. This adaptive control strategy based on multi-level decision-making can effectively cope with various complex working conditions and improve the quality stability and efficiency of bolt tightening.

[0015] In this embodiment of the invention, local rate of change sign analysis is performed on the torque signal to identify segments where the rate of change signs alternate frequently, which are marked as viscous-slip oscillation segments, including: Differential operations are performed on adjacent sampling points of the torque signal to obtain the torque change rate sequence. Differential operation is a fundamental step in acquiring the dynamic characteristics of the signal; by calculating the difference in torque values ​​between adjacent sampling points, the instantaneous rate of torque change is obtained. The calculation uses a first-order difference equation: ; in, For the first Torque change rate at each sampling point For the first Torque value at each sampling point This is the torque value from the previous sampling point. The sampling time interval is denoted as . This calculation intuitively reflects the instantaneous rate of change of torque and is an important indicator for identifying the dynamic behavior of the system. In practical applications, to reduce the influence of random noise, center difference or higher-order difference methods can also be used. However, for identifying high-frequency oscillations during the tightening process, a simple first-order forward difference method usually provides sufficient information.

[0016] The positive and negative signs of each element in the torque rate of change sequence are extracted to generate a symbol sequence. Symbol extraction is an effective means of simplifying signal analysis by highlighting the alternating characteristics of the signal by retaining only the direction information of the rate of change. The extraction process converts each element in the torque rate of change sequence into its symbol representation: ; The generated symbol sequence contains only three values: +1, 0, and -1, intuitively reflecting the states of torque increase, remain unchanged, or decrease. This symbolization significantly simplifies the complexity of subsequent analysis while retaining the key information needed for oscillation identification. In practical implementations, considering the impact of measurement noise, a small threshold is typically introduced. ,when The time is considered zero to avoid frequent sign changes caused by minute noise.

[0017] A fixed-length detection window is set along the symbol sequence, and the number of distinct flips between adjacent symbols within each window is counted. Sliding window detection is a core technology for identifying local oscillation characteristics. It analyzes the symbol change characteristics within the window by sliding a fixed-length window across the signal. Each window covers a continuous symbol sequence, and the window length is typically set to 2-4 times the expected oscillation period to ensure complete capture of oscillation characteristics. At each sliding window position, the number of changes between adjacent symbols is counted, i.e., when... Each time is recorded as one flip. This analysis method based on sign change is not sensitive to signal amplitude and can effectively identify frequent alternations in the direction of torque change, which is a typical characteristic of viscous-slip oscillation.

[0018] When the ratio of the number of flips to the length of the sliding window exceeds a preset flip density threshold, the segment covered by that sliding window is marked as a viscous-sliding oscillation segment. Flip density is a direct indicator of the intensity of oscillation; it is determined by calculating the symbol flip frequency within a unit window length to assess the signal's oscillation characteristics. Flip density is defined as: ; in, For the flipped density, The number of times the window is flipped. The sliding window length is defined as follows: When the sign-flipping density exceeds a preset threshold (typically set to 0.3-0.5, indicating an average sign-flipping occurrence every 2-3 sampling points), the segment is considered to exhibit significant viscous-slip oscillation characteristics. This statistically based judgment method is simple, efficient, and effectively distinguishes between normal fluctuations and abnormal oscillations, providing accurate oscillation segment location for subsequent signal processing.

[0019] Adjacent viscous-slip oscillation segments are merged, and the start and end sampling indices of each merged segment are recorded. Segment merging is a necessary step in processing continuous oscillations. By merging oscillation segments that are close in time, a more complete description of the oscillation characteristics is obtained. The merging process checks all marked oscillation segments. When the interval between two segments is less than a preset minimum interval threshold (usually half of the expected oscillation period), they are considered different parts of the same oscillation process and merged. Finally, the start and end sampling indices of each merged oscillation segment are recorded to form a complete description of the oscillation segment. This segment merging strategy avoids repeated identification and fragmented processing of the same oscillation process, improves the coherence and accuracy of subsequent analysis, and provides a more complete data foundation for the overall assessment of oscillation characteristics.

[0020] In this embodiment of the invention, based on the zero-crossing time-series characteristics of the viscous-slip oscillation segment, the main oscillation period is extracted, and an adaptive filtering window model is constructed, including: Within the viscous-slip oscillation region, the zero-crossing points of the torque rate of change sequence (from positive to negative or vice versa) are detected, forming a zero-crossing index sequence. Zero-crossing detection is a classic method in frequency analysis, identifying the key time points of the signal's periodic changes by pinpointing when the signal crosses zero. Within the oscillation region, the zero-crossing points of the torque rate of change sequence correspond to local extreme points of torque, accurately reflecting the time-domain characteristics of the oscillation. The detection process traverses all sampling points within the oscillation region. When two adjacent points in the torque rate of change sequence have opposite signs, linear interpolation is used to determine the precise zero-crossing position, and the corresponding sampling index is recorded, forming a complete zero-crossing index sequence. This zero-crossing-based analysis method is computationally simple and efficient, has low sensitivity to noise and amplitude fluctuations, and can stably extract the time-series characteristics of the oscillation, providing a reliable basis for period estimation.

[0021] The index difference between adjacent zero-crossing points in the zero-crossing index sequence is calculated and denoted as the half-cycle sample number sequence. Half-cycle analysis is a crucial step in oscillation period estimation; by calculating the time interval between adjacent zero-crossing points, the half-cycle characteristics of the oscillation are directly obtained. The calculation process involves taking the difference between adjacent elements in the zero-crossing index sequence to obtain continuous half-cycle sample numbers: ; in, For the first Number of sampling points per half-cycle For the first A sampling index that crosses zero. This serves as the sampling index for the previous zero-crossing point. The half-cycle sample sequence directly reflects the changing characteristics of the oscillation period, containing frequency modulation information and nonlinear features of the oscillation. Compared to directly analyzing the whole period, half-cycle analysis can capture more detailed oscillation asymmetry and provide richer oscillation characteristic information.

[0022] Median statistics are performed on the half-cycle sampled sequence, and twice the median is converted into a time quantity, denoted as the main oscillation period. Median statistics are a robust method for obtaining stable period estimates. By selecting the median value of the half-cycle sequence, the influence of extreme values ​​and outliers is effectively avoided. The main reason for choosing the median instead of the mean is that sudden periodic changes or measurement anomalies may occur during the oscillation process. These outliers will significantly affect the mean, but have a smaller impact on the median. Twice the median corresponds to the complete oscillation period. Multiplying it by the sampling time interval converts it into the actual time quantity. ; in, The main cycle is oscillation. The median of the half-cycle sample number sequence. The sampling time interval is denoted as . This median-based period estimation method has strong anti-interference capabilities and can provide stable and reliable oscillation period estimates in complex tightening environments.

[0023] The filter window width is obtained by multiplying the oscillation period by a preset window scaling factor. Determining the window width is a crucial step in constructing an adaptive filter. Dynamic calculations based on the oscillation period ensure that the filter window matches the oscillation characteristics. The formula for calculating the window width is: ; in, This represents the width of the filter window (number of sampling points). This is the window scaling factor. The main cycle is oscillation. The sampling time interval is defined as . The window factor is typically set between 2 and 4, with the specific value depending on the stability of the oscillation and the desired filtering effect. A larger factor provides a smoother filtering result but may lead to over-smoothing; a smaller factor retains more detail but may result in insufficient filtering. This window design based on the oscillation period ensures that the filter's cutoff frequency matches the oscillation frequency, effectively suppressing oscillation components while preserving low-frequency trend information.

[0024] Based on the filter window width, a weighted coefficient vector of corresponding length is generated to construct an adaptive filter window model. Weighted coefficient generation is the final step in filter window construction; by designing a suitable weight distribution, the frequency response characteristics of the filter are optimized. The basic weighted coefficients can be modeled using a Gaussian function. ; in, For the first The weight of each sampling point Index for the center point of the window, The standard deviation parameter of the Gaussian function is typically set to 1 / 4 to 1 / 6 of the window width. The Gaussian window exhibits excellent time-frequency localization characteristics, providing a smooth response in both the time and frequency domains and reducing spectral leakage during filtering. Finally, the weight vector is normalized to ensure that the sum of all weights is 1, maintaining the average energy of the signal. This adaptive window design based on the Gaussian model considers both the oscillating frequency characteristics and provides a smooth filtering response, laying the foundation for subsequent signal processing.

[0025] In this embodiment of the invention, an adaptive filtering window model is used to perform weighted smoothing on the torque signal, and phase lead compensation is applied to the smoothing result to generate a de-oscillating torque trend signal, including: The weighted coefficient vector of the adaptive filtering window model is slid point-by-point along the torque signal. The dot product of the sampled value within the window and the weighted coefficient at each position is calculated to obtain the preliminary smoothed signal. Window sliding is the core operation for weighted smoothing; by moving the filtering window to cover the entire signal, global filtering is achieved. At each sampling point, the window center is aligned with the current point, and the weighted sum of all sampling points within the window and their corresponding weighted coefficients is calculated. ; in, For the first Smoothed torque value at each sampling point The original torque signal, These are the corresponding weighting coefficients. The value represents the window length. For points near the signal boundary, symmetrical or constant extension methods are used to handle cases where the window exceeds the signal range. This weighted average-based filtering method effectively suppresses high-frequency oscillation components while preserving the low-frequency trend information of the signal, making it an effective means of removing viscous-sliding oscillation interference.

[0026] The group delay is calculated by determining the time delay introduced by the filter center relative to the current sampling point based on the filter window width. Group delay calculation is a fundamental step in phase compensation, determining the time delay introduced by the filtering process through theoretical analysis. For symmetrical weighted windows, the group delay is primarily determined by the window's center position and can be approximated as half the window width. This theoretical calculation assumes perfectly symmetrical weighting coefficients and a uniform distribution of signal frequency components. In practical applications, the group delay can be determined more accurately by analyzing the filter's phase-frequency characteristics or by verifying with test signals. Accurate group delay estimation is crucial for achieving precise phase compensation and directly affects the time alignment of subsequent analyses.

[0027] The initial smoothed signal is shifted forward by the number of points corresponding to the group delay in the sampling index direction to obtain the phase-aligned signal. Time-domain forward shifting is a direct method to achieve phase compensation, eliminating the phase delay introduced by filtering by shifting the signal time series. The forward shift operation advances the value of each sampling point by a specified delay: ; in, For the first Signal values ​​after phase alignment of each sampling point For the initial smoothing of the signal, from then on The value of each point. This time-domain forward shift method is intuitive and effective, and can accurately compensate for the fixed group delay introduced by the linear phase filter, ensuring that the phase characteristics before and after signal processing remain consistent, and providing a time-aligned signal for subsequent stiffness calculations.

[0028] For the missing region caused by the forward shift at the end of the phase-aligned signal, linear extrapolation is performed using the end slope to fill it, generating a de-oscillation torque trend signal. Extrapolation filling is a necessary step in handling boundary problems, filling the data gaps caused by the forward shift in the time domain using a reasonable prediction method. The filling process first calculates the local slope at the end of the phase-aligned signal, usually obtained through linear regression of the last few valid points; then, based on this slope, linear extrapolation is performed to fill the missing region at the end after the forward shift. ; in, For extrapolation region number Fill value for each point, This represents the last valid value of the phase alignment signal. To calculate the terminal slope, The total length of the signal is given. This slope-based linear extrapolation method is simple and effective, maintaining the continuity of the signal trend, avoiding abrupt changes at the boundary, and generating a complete deoscillation torque trend signal, providing continuous and smooth input data for subsequent stiffness analysis.

[0029] In this embodiment of the invention, piecewise energy integration is performed on the de-oscillation torque trend signal and the rotation angle signal to obtain the equivalent stiffness value of each segment interval, and the current tightening stiffness value is obtained by extrapolation through the stiffness change trend, including: The rotation angle signal is divided into multiple continuous segmented intervals according to a preset angular step size. Angle segmentation is a preliminary step in energy integration, ensuring physical consistency of the analysis intervals through equal-angle division. The division process begins at the tightening start point and gradually divides the rotation angle signal along a preset angular step size (typically 5° to 10°), forming a series of continuous angular intervals. Each interval corresponds to a fixed angular segment of bolt rotation, and the interval boundary is determined by the sampling point reaching a specific angle. This uniform division based on angle ensures the physical comparability of each segment, facilitating subsequent analysis of the stiffness evolution during bolt tightening. In practical applications, the selection of the angular step size needs to balance the requirements of analytical accuracy and computational efficiency. Too small a step size increases computational load and may introduce interference from local fluctuations, while too large a step size may lose important details of change.

[0030] Numerical integration is performed on the de-oscillation torque trend signal within each segmented interval to obtain the cumulative torque energy value for that interval. Energy integration is the core calculation for evaluating system stiffness; the energy absorption within the interval is obtained by integrating the accumulated torque over the angle. The calculation employs a numerical integration method, accumulating the torque signal within the interval according to the angle: ; in, For the first The cumulative torque energy value in each interval For the interval of the first The de-oscillation torque trend value at each sampling point This represents the angle increment corresponding to that point. The cumulative energy value intuitively reflects the energy absorption characteristics during bolt tightening and is the physical basis for evaluating system stiffness. Compared to the simple torque-angle slope calculation, the energy integration method considers complete information within the interval, providing more stable and reliable stiffness assessment results and reducing the impact of local fluctuations.

[0031] The equivalent stiffness value for each segment is obtained by dividing the accumulated torque energy of each segment by the corresponding angle step size. Equivalent stiffness calculation is the final step in energy analysis, converting the accumulated energy into a stiffness metric through standardization. The calculation uses the energy-angle ratio method. ; in, For the first The equivalent stiffness values ​​for each interval. This represents the cumulative torque energy value within this range. The preset angle step size is used. The equivalent stiffness value is expressed in the form of energy / angle, which intuitively reflects the energy input required for unit angle deformation and is a comprehensive characterization of the system's stiffness characteristics. This energy-based stiffness definition is superior to the traditional simple slope calculation, as it considers complete information within the interval and can provide more stable stiffness assessment results, providing a reliable basis for real-time monitoring and control of the bolt tightening process.

[0032] Linear regression is performed on the equivalent stiffness values ​​of several recent consecutive segmented intervals to obtain the slope of stiffness change. Linear regression is an effective method for analyzing stiffness trends; by fitting the stiffness values ​​of the most recent intervals, the variation law of system stiffness can be obtained. The regression process selects the most recent N consecutive segmented intervals (usually N=3-5), with the center angle of the interval as the independent variable and the corresponding equivalent stiffness value as the dependent variable. The least squares method is used to fit the straight line. ; in, For angle Stiffness prediction at the location, The slope of the stiffness change. The fitting intercept. The slope of the stiffness change. This intuitively reflects the trend of system stiffness changing with rotation angle and is a key parameter for predicting future stiffness. A positive slope indicates that the bolt is entering the fastening phase, and the stiffness increases with rotation angle; a negative slope may indicate that the bolt is approaching or has exceeded its yield point, and the stiffness begins to decrease. This trend-based stiffness analysis can predict system behavior in advance, providing a basis for decision-making in active control strategies.

[0033] The current tightening stiffness value is obtained by multiplying the slope of the stiffness change by the extrapolation step size and then adding it to the equivalent stiffness value of the latest segmented interval. Extrapolation calculation is the final step in obtaining real-time stiffness, using trend prediction to obtain a stiffness estimate for the current moment. The calculation formula is: ; in, This is the current tightening stiffness value. This represents the equivalent stiffness value for the newly completed interval. The slope of the stiffness change. The extrapolation step size is typically set to half an angle. This trend-based extrapolation stiffness estimation method overcomes the lag of piecewise analysis, providing near real-time stiffness predictions and timely decision-making basis for dynamic adjustment of tightening parameters. By comprehensively considering historical data and changing trends, the extrapolation method maintains computational stability while rapidly responding to changes in system characteristics, enabling precise monitoring and control of the tightening process.

[0034] In this embodiment of the invention, a stiffness statistical reference distribution based on historical tightening data is constructed, the standardized deviation of the current tightening stiffness value relative to the stiffness statistical reference distribution is calculated, and a confidence index is generated, including: A sample set of stiffness values ​​at the same rotation angle is extracted from historical tightening records. The mean and standard deviation of this sample set are calculated to construct a stiffness statistical reference distribution. Constructing the reference distribution is a fundamental step in adaptive evaluation, establishing a standardized evaluation benchmark through historical data statistics. The construction process first selects stiffness data points close to the current rotation angle from historical tightening records to form a sample set; then, the statistical characteristics of this set are calculated, including the mean. and standard deviation Construct a normal distribution model As a reference distribution, it reflects the normal range of stiffness variation at a specific corner position, providing a statistical benchmark for the current tightening process. In practical applications, the reference distribution can be hierarchically managed according to bolt specifications, material properties, and tightening conditions, improving the relevance and accuracy of the evaluation. This adaptive evaluation method based on historical data can adapt to different product and process conditions, significantly improving the system's versatility and robustness.

[0035] Calculate the absolute value of the difference between the current tightening stiffness value and the mean of the stiffness statistical reference distribution, and divide it by the standard deviation of the stiffness statistical reference distribution to obtain the standardized deviation value. Standardized deviation calculation is a core step in quantifying the degree of anomalies, achieving objective assessment through statistical standardization. The calculation uses the standard Z-score formula. Standardized deviation value. The Z-score intuitively reflects the degree of deviation of the current stiffness from the normal range and is a dimensionless indicator for quantifying anomalies. Compared with simple absolute or relative differences, the Z-score takes into account the fluctuation characteristics of historical data, providing more objective evaluation results and is suitable for unified evaluation of different tightening stages and different types of bolts.

[0036] The standardized deviation values ​​are mapped to a pre-defined confidence decay curve to obtain the initial confidence level. Confidence mapping is the step of transforming anomaly into confidence, and it uses a non-linear function to achieve an intuitive expression of the score. The mapping adopts an exponential decay model: ; in, As the initial confidence level, This is the standardized deviation value. The attenuation coefficient is typically set to 0.1-0.2. This exponential function-based mapping model provides a high confidence level close to 1 when the deviation is small, but the confidence level decreases rapidly as the deviation increases, consistent with the general laws of statistical evaluation. The confidence level ranges from 0 to 1; the closer to 1, the more reliable the current stiffness value, while the closer to 0, the more likely there are abnormal or unstable factors. This continuous scoring mechanism is more flexible than simple binary judgments and can support subsequent multi-level decision-making strategies.

[0037] The oscillation intensity coefficient is calculated based on the peak-to-peak value of the torque change rate within the current viscous-slip oscillation range. Oscillation intensity assessment is a crucial step in supplementing confidence levels, adjusting the assessment results by quantifying the severity of the oscillations. The calculation process first extracts the maximum and minimum values ​​of the torque change rate within the currently active oscillation range to calculate the peak-to-peak value; then, the peak-to-peak value is normalized to the current mean torque value to obtain the relative oscillation intensity; finally, it is compared with a preset benchmark oscillation intensity to derive the oscillation intensity coefficient. Oscillation intensity directly affects the reliability of stiffness estimation; strong oscillations increase the uncertainty of signal processing results, reducing the accuracy of stiffness prediction. This supplementary assessment based on physical characteristics overcomes the limitations of purely statistical methods, providing a more comprehensive reliability evaluation.

[0038] The initial confidence level is multiplied by a reduction factor negatively correlated with the oscillation intensity coefficient to generate a confidence index. Confidence adjustment is the final step in the comprehensive evaluation, achieving multi-factor assessment by considering the impact of oscillations. The adjustment uses a multiplicative model: ; in, As the final confidence level indicator, As the initial confidence level, This is an adjustment factor (usually set to 0.2-0.5). This represents the oscillation intensity coefficient. This adjustment mechanism appropriately reduces the confidence level when the oscillation intensity increases, reflecting the negative impact of oscillations on the reliability of stiffness estimation. The final confidence index comprehensively considers both statistical anomalies and physical stability dimensions, providing a comprehensive and reliable assessment basis for subsequent decision-making. This multi-factor fusion assessment method significantly improves the system's adaptability to complex operating conditions and can more accurately identify real anomalies and temporary fluctuations.

[0039] In this embodiment of the invention, a stiffness acceptance strategy is set according to a confidence index, and the target torque correction amount and tightening speed adjustment amount are dynamically calculated in combination with the current tightening stiffness value. The corrected tightening process parameters are then output, including: Set confidence thresholds and low-confidence duration thresholds. Threshold settings are a fundamental configuration for decision-making systems, defining different response ranges through threshold definitions. Confidence thresholds are typically set to 0.7-0.8, representing the minimum confidence level the system considers reliable; low-confidence duration thresholds are typically set to 0.3-0.5 seconds, defining the maximum time the system can tolerate temporary low-confidence states. These two thresholds together form the basic framework of the system's decision-making, enabling rapid response to obvious anomalies while tolerating short-term fluctuations and avoiding overreaction. The specific threshold settings need to be optimized based on the application scenario and product characteristics, achieving a balance between stability and response speed. Reasonable threshold settings are crucial for the robust operation of the system and directly affect the effectiveness of the control strategy.

[0040] When the confidence level is higher than the confidence threshold, the target torque correction is calculated using the current tightening stiffness value. High-confidence processing is the system's primary operating mode, achieving active control by directly adopting the current estimate. In a high-confidence state, the system considers the current tightening stiffness value to have sufficient reliability and directly uses it for parameter calculation. The calculation of the target torque correction is based on the relationship between the expected target torque and the current stiffness: ; in, The target torque correction amount, For the expected target torque, This is the current tightening stiffness value. To predict the completion angle, this stiffness-based torque prediction method can adjust tightening parameters in advance, achieving precise target torque control, reducing the risk of over-tightening or under-tightening, and improving the consistency and reliability of tightening quality.

[0041] When the confidence level is below the confidence threshold but the duration does not exceed the low confidence duration threshold, the stiffness value from the previous sampling time is retained for calculation. Temporary low confidence handling is a buffer mechanism of the system, using a temporary holding strategy to cope with short-term fluctuations. When the system detects a decrease in confidence but it has not yet lasted for a long time, a conservative strategy is adopted, continuing to use the most recent reliable stiffness value for parameter calculation, rather than immediately adopting the current low confidence estimate. This buffer mechanism can effectively filter out short-term disturbances and fluctuations, avoid frequent changes in system parameters, and maintain control stability. The buffer period is set based on the time characteristics of the tightening process and the dynamic response of signal processing, providing the system with appropriate reaction time and reducing misjudgments and over-adjustments.

[0042] When the duration of a low-confidence state exceeds a low-confidence duration threshold, the tightening speed adjustment is reduced by a preset ratio to suppress increased oscillation. Continuous low-confidence handling is the system's protection mechanism, addressing persistent anomalies through speed adjustment. When the duration of a low-confidence state exceeds a threshold, the system considers a potential substantial problem in the current tightening process, requiring proactive intervention. The speed adjustment employs a proportional deceleration strategy. ; in, The adjusted tightening speed, The current tightening speed, This is the speed adjustment coefficient (usually set to 0.3-0.7). This represents the current confidence level. This dynamic deceleration strategy decelerates more when the confidence level is lower, effectively suppressing oscillations and providing the system with an opportunity to recover stability. Speed ​​adjustment is an active intervention method that reduces oscillations by decreasing the system's kinetic energy, creating conditions for subsequent precise control.

[0043] The target torque correction and tightening speed adjustment are combined to output the corrected tightening process parameters. Parameter combination is the final step in the control system, generating a comprehensive control command through multi-dimensional adjustments. The combination process packages the torque correction and speed adjustment according to the format required by the control system, forming a complete process parameter correction command. In practical applications, depending on the specific tightening equipment and control interface, auxiliary parameters such as acceleration limits and tightening termination conditions may also be needed to form a complete control strategy. The corrected process parameters are directly sent to the tightening execution system to achieve real-time control of the tightening process, ensuring that the bolt tightening quality meets design requirements. This adaptive control strategy based on multi-level decision-making can effectively cope with various complex working conditions and improve the quality stability and efficiency of bolt tightening.

[0044] In this embodiment of the invention, generating a weighted coefficient vector of corresponding length based on the filter window width includes: Within the viscous-slip oscillation region, sampling points are divided into viscous phase and slip phase sub-segments based on the sign of the torque change rate. Phase classification is a crucial step in understanding the oscillation characteristics, identifying different stages of oscillation through sign analysis. The classification process checks the sign of the torque change rate within the oscillation segment: a positive rate indicates increasing torque, corresponding to the viscous phase; a negative rate indicates decreasing torque, corresponding to the slip phase. This classification method based on physical characteristics directly reflects the fundamental mechanism of viscous-slip oscillation: in the viscous phase, contact surface adhesion leads to torque accumulation; in the slip phase, contact surface sliding leads to torque release. By statistically analyzing the time distribution of each phase, the dynamic characteristics of the oscillation can be deeply understood, providing a physical basis for window design. In practical applications, to reduce the impact of noise, a small threshold is usually introduced to distinguish between true phase changes and noise fluctuations, improving the stability and reliability of the classification.

[0045] The average number of sustained sampling points for both the viscous and sliding phase segments is counted separately, and the ratio of these two values ​​is denoted as the phase duration ratio. Calculating the duration ratio is a crucial step in quantifying oscillation asymmetry, revealing the temporal structure characteristics of the oscillations through statistical analysis. The calculation process first involves counting the duration (number of sampling points) of each viscous and sliding phase, then calculating their respective averages, and finally taking the ratio to obtain the phase duration ratio. The phase duration ratio intuitively reflects the asymmetric characteristics of oscillations. In standard symmetrical oscillations, this ratio is close to 1; in typical viscous-slip oscillations, the duration of the viscous phase is usually longer than that of the slip phase, and the ratio is greater than 1. This time-structure-based characteristic analysis can capture the physical essence of oscillations, providing a more accurate basis for window design and achieving a fine match with actual oscillation characteristics.

[0046] Using the filter window width as the total length, the window is divided into two asymmetric intervals based on the phase duration ratio. The ratio of the length of the first interval to the length of the second interval is equal to the phase duration ratio. Window segmentation is a key design feature to adapt to oscillation asymmetry, and non-uniform division is used to match time structure characteristics. The segmentation process divides the window of total length L into two intervals, with the length of the first interval being... and the length of the interval satisfy: ; ; Solving for: ; ; This asymmetric partitioning based on phase duration ratio allows the window structure to match the actual oscillation time characteristics, with the first interval corresponding to a longer viscous phase and the second interval corresponding to a shorter sliding phase. This asymmetric window design is an innovative extension of the traditional symmetric window, enabling more precise adaptation to complex oscillation characteristics, improving filtering performance and phase fidelity, and providing more accurate input for subsequent signal processing.

[0047] Independent half-Gaussian decay curves are generated for the first and second intervals respectively. The peaks of the two curves are aligned and spliced ​​at the center of the window to form an asymmetric basic weight curve. Weight curve design is the core step in constructing the window function, using an asymmetric function to match the characteristic structure of the oscillation. The design process applies a half-Gaussian function to the first and second intervals respectively: ; in, This represents the weight coefficient value of the first interval in the window function. This represents the weight coefficient value of the latter interval in the window function. Index for the center of the window, and These are the standard deviation parameters for the preceding and following intervals, typically set to 1 / 4 to 1 / 6 of the corresponding interval length. The two half-Gaussian curves are aligned at the center point and smoothly joined to form a complete asymmetric window function. This asymmetric design allows the window function to better adapt to the time structure characteristics of oscillations, providing appropriate filtering responses at different phases. Compared to traditional symmetric windows, it can more accurately preserve signal characteristics and reduce phase distortion.

[0048] The mean amplitude difference between the sequences of local maxima and local minima of the torque change rate within the viscous-slip oscillation range is calculated and denoted as the mean oscillation amplitude. Calculating the mean amplitude is a crucial step in assessing the intensity of the oscillation, quantifying its severity through extreme value analysis. The calculation process first identifies all local maxima and minima of the torque change rate within the oscillation range; then, it calculates the amplitude difference between adjacent maxima and minima, forming an amplitude difference sequence; finally, it calculates the mean of this sequence to obtain the mean oscillation amplitude. The oscillation amplitude directly reflects the severity of torque changes and is an important basis for adjusting the window weight distribution. Strong oscillations require a smoother window function to provide a stronger filtering effect, while weak oscillations can use a steeper window function to retain more signal details.

[0049] The average oscillation amplitude is compared with a preset baseline oscillation amplitude to obtain the amplitude scaling factor. Amplitude scaling is a key step in adaptive window adjustment, achieving dynamic optimization of the window shape through amplitude comparison. The scaling factor is calculated using a ratio method: ; in, This is the amplitude scaling factor. This represents the average current fluctuation range. This is a preset baseline oscillation amplitude. When the actual oscillation is stronger than the baseline, the scaling factor is greater than 1, indicating that the filtering effect needs to be enhanced; when the actual oscillation is weaker than the baseline, the scaling factor is less than 1, indicating that the filtering intensity can be reduced. This adaptive adjustment mechanism based on oscillation characteristics enables the window function to automatically optimize according to the actual oscillation situation, minimizing signal distortion while ensuring the filtering effect, and improving the system's adaptability and robustness.

[0050] Multiplying the peak height of the asymmetric basic weight curve by the reciprocal of the amplitude scaling factor reduces the advantage of the weights at the window center relative to the edges as oscillations become more severe. Weight adjustment is a specific operation of window shape optimization, achieving adaptive control of the filtering intensity through amplitude correction. The adjustment process modifies the peak height of the window function: ; in, The adjusted center point weights, The original center point weights, This is the amplitude scaling factor. Simultaneously, the weights at the window edges remain unchanged, forming a new window shape. This adjustment strategy reduces the relative advantage of the center weights during periods of severe oscillation, making the window function smoother and providing stronger filtering; conversely, it strengthens the center weights during periods of weak oscillation, making the window function steeper and preserving more signal details. This adaptive shape adjustment based on oscillation characteristics can provide optimal filtering response under different operating conditions, balancing the requirements of filtering effect and signal fidelity.

[0051] A cosine decay function is applied to the regions at both ends of the adjusted weight curve, with a preset proportion of length, to eliminate truncation and abrupt changes at the window boundaries. Edge processing is the final refinement step in window design, eliminating boundary effects through smooth transitions. The processing involves taking 10%-15% of the total length at both ends of the window and applying a cosine decay function. ; in, The weights after edge processing. The original weights after adjustment. This is the starting position of the transition zone. The length of the transition region is given. Cosine gradient ensures the window function smoothly decays to zero at the boundaries, eliminating spectral leakage caused by truncation. This edge smoothing is standard practice in high-quality window design, significantly improving the filter's frequency response characteristics, reducing spurious signals during filtering, and enhancing signal processing accuracy.

[0052] The weighted curve, after boundary gradation processing, is integrally normalized to ensure the sum of all weighted coefficients equals a unit value, generating a weighted coefficient vector. Normalization is a necessary step in the final shaping of the window function, ensuring energy conservation through coefficient adjustment. The process calculates the sum of all weights and then divides each weight by this sum. Normalization ensures that the sum of all weighted coefficients equals 1, guaranteeing that the filtering process does not change the average energy of the signal and avoids introducing system gain bias. The final weighted coefficient vector integrates oscillation period characteristics, phase structure features, and amplitude intensity information, forming a fully adaptive filtering window capable of providing optimal filtering effects for different oscillation characteristics, providing a high-quality fundamental tool for signal processing.

[0053] In this embodiment of the invention, linear regression is performed on the equivalent stiffness values ​​of multiple recent consecutive segmented intervals to obtain the slope of stiffness change, including: The equivalent stiffness values ​​of a predetermined number of recent consecutive segmented intervals are selected as regression samples. Sample selection is a fundamental step in regression analysis, ensuring representativeness and timeliness by choosing appropriate data points. The selection process begins with the most recently completed segmented interval and proceeds backwards through N consecutive intervals (typically N=3-5), collecting their equivalent stiffness values ​​as the basis for regression analysis. Determining the number of selected intervals requires balancing two requirements: a sufficiently large sample size to provide statistical stability, yet not so large as to reflect recent system behavior. During tightening, stiffness characteristics may change significantly with each stage; an excessively long historical window may introduce irrelevant historical data, affecting the accuracy of trend prediction. This sliding window analysis method based on recent data enables real-time tracking of the system's dynamic characteristics, providing the most relevant information for stiffness trend prediction.

[0054] Using the center corner position of each segmented interval as the independent variable and the corresponding equivalent stiffness value as the dependent variable, a linear regression equation is constructed. Constructing the regression equation is the core step in trend analysis, quantifying the relationship between variables through a mathematical model. The construction process first clarifies the dependent and independent variables: the equivalent stiffness value. As the dependent variable, the corresponding interval center angle As the independent variable; then establish the assumption of a linear relationship:

[0055] in, For angle Stiffness prediction at the location, Let be the slope of the stiffness change to be determined. The value is the fitting intercept. This linear model assumes that stiffness changes linearly within a local range, which is a reasonable simplification of complex nonlinear systems. It can provide sufficiently accurate local trend predictions while being computationally simple. At different stages of the tightening process, other forms of regression models (such as quadratic or exponential models) may need to be selected based on physical characteristics, but in most practical applications, the local linear model can provide satisfactory prediction accuracy.

[0056] The slope parameter of the linear regression equation is solved using the least squares method, denoted as the stiffness change slope. Least squares fitting is a standard method for parameter solving, obtaining the optimal parameter estimate by minimizing the sum of squared errors. The solution process is based on the classic least squares formula for linear regression: ; ; in, For the first The corner position of each sample This corresponds to the equivalent stiffness value. and These are the average values ​​of the rotation angle and stiffness, respectively, and the summation covers all selected sample points. The slope is calculated in this way. The slope directly reflects the trend of system stiffness change with rotation angle and is a key parameter for predicting future stiffness changes. During the tightening process, the sign and magnitude of the slope have important physical meanings: a positive slope indicates that the system is entering the solid phase and the stiffness increases with rotation angle; a negative slope may indicate that the system is approaching or has exceeded the yield point and the stiffness has begun to decrease; the absolute value of the slope reflects the rate of stiffness change and directly affects the prediction time window and accuracy.

[0057] Calculate the sum of squared residuals for each regression sample point relative to the regression line. When the sum of squared residuals exceeds a preset residual threshold, remove the sample point with the largest deviation and regress. Outlier handling is a crucial step in improving regression stability; by screening outliers, the representativeness of the model is enhanced. The processing first calculates the residual for each sample point: ; Then, the residual sum of squares (RSS) is calculated. When the RSS exceeds a preset threshold (usually set based on the statistical characteristics of historical data), the sample point with the largest absolute residual value is identified and removed from the dataset. The remaining samples are then used to perform linear regression again. This outlier handling based on residual analysis effectively reduces the impact of outliers and improves the stability and representativeness of the regression model. In complex tightening environments, factors such as material inhomogeneity and changes in contact conditions may cause individual data points to deviate from the normal trend. Through reasonable outlier handling, more reliable trend predictions can be obtained, providing a more accurate basis for subsequent stiffness estimation and parameter adjustment.

[0058] In this embodiment of the invention, the oscillation intensity coefficient is calculated based on the peak-to-peak value of the torque change rate within the current viscous-slip oscillation range, including: Extract the maximum and minimum values ​​of the torque change rate sequence within the current viscous-slip oscillation range. Extreme value extraction is a fundamental step in assessing oscillation amplitude, quantifying the intensity of fluctuations by identifying the range of variation. The extraction process examines all data points of the torque change rate within the currently active oscillation range to identify the maximum and minimum values. Extreme value detection needs to consider signal noise and short-term fluctuations, typically employing a local extremum search algorithm. This requires extreme points to remain at a local maximum or minimum within a certain window, avoiding the misclassification of noise points as extreme values. Accurate extreme value extraction is fundamental to oscillation intensity assessment and directly affects the reliability of subsequent calculations. In practical applications, it may be necessary to combine techniques such as filtering and thresholding to further improve the robustness of extreme value detection, ensuring stable and reliable extreme value estimates under various noise conditions.

[0059] The difference between the maximum and minimum values ​​is calculated and denoted as the peak-to-peak rate of change. Peak-to-peak value calculation is a direct method for quantifying oscillation amplitude, assessing the fluctuation range through range analysis. The calculation uses simple algebraic difference. The peak-to-peak value intuitively reflects the fluctuation range of the torque change rate and is a fundamental quantitative indicator of oscillation intensity. Compared to statistical indicators such as standard deviation, peak-to-peak value more directly reflects the amplitude of extreme fluctuations, making it particularly suitable for assessing viscous-slip oscillations with distinct peaks and troughs. In signal processing, peak-to-peak value is a commonly used parameter characterizing the dynamic range of a signal; its physical meaning is clear, and its calculation is simple, making it an ideal choice for oscillation intensity assessment.

[0060] The relative oscillation amplitude is obtained by normalizing the rate of change by dividing the peak-to-peak value of the rate of change by the mean of the current torque signal. Normalization is a crucial step for fair comparison, eliminating the influence of absolute values ​​through relative expression. The process first calculates the mean of the torque signal within the current time period, and then divides the peak-to-peak value of the rate of change by this mean.

[0061] This normalization process considers the comparability of oscillation assessments under different torque levels, converting absolute oscillation amplitude into a relative intensity index. At low torque levels, even small absolute oscillation amplitudes can have a significant impact relative to the base torque; while at high torque levels, larger absolute oscillations may have a relatively smaller impact. Through normalization, the system can provide a consistent oscillation assessment standard across different tightening stages and different bolt types, improving the universality and comparability of the assessment.

[0062] The ratio of the relative oscillation amplitude to a preset oscillation benchmark value is used as the oscillation intensity coefficient. Calculating the intensity coefficient is the final step in standardizing the evaluation results, generating a normalized index through benchmark comparison. The calculation uses a simple ratio: ; in, This is the oscillation intensity coefficient. The relative oscillation range, This is a preset oscillation baseline value. The oscillation baseline value represents the oscillation level that the system considers "normal," and is usually determined through historical data analysis or theoretical models. When When the value is close to 1, it indicates that the current fluctuation is at a normal level; when... A value much greater than 1 indicates abnormally strong oscillations, which may require intervention; when A value much less than 1 indicates an abnormally weak oscillation, suggesting the system may be in an atypical state. This strength coefficient, based on benchmark comparison, provides a direct input for subsequent confidence level adjustments, enabling the system to adjust its decision-making strategy according to the oscillation state, thereby improving the adaptability and robustness of control.

[0063] This invention achieves adaptive correction of bolt tightening process parameters by performing local rate-of-change sign analysis on the torque signal during bolt tightening, extracting oscillation characteristics, constructing an adaptive filtering window, performing phase compensation, calculating stiffness through piecewise energy integration, constructing a statistical reference distribution, and dynamically adjusting process parameters. The method of this invention can effectively identify and address viscous-slip oscillations during tightening, improve the accuracy and reliability of stiffness estimation, and significantly improve the consistency and stability of bolt tightening quality. It is particularly suitable for bolt connection applications requiring high precision and high reliability.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0065] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

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

Claims

1. An adaptive correction method for bolt tightening process parameters based on algorithm analysis, characterized in that, include: Acquire real-time sampling data of torque and rotation signals during the bolt tightening process; Local rate of change sign analysis is performed on the torque signal to identify segments with high-frequency alternation of rate of change signs, which are marked as viscous-sliding oscillation segments; Based on the zero-crossing time-series characteristics of the viscous-slip oscillation segment, the main oscillation period is extracted, and an adaptive filtering window model is constructed. The torque signal is weighted and smoothed using the adaptive filtering window model, and the smoothing result is compensated for phase lead to generate a de-oscillation torque trend signal. The piecewise energy integration operation is performed on the de-oscillation torque trend signal and the rotation angle signal to obtain the equivalent stiffness value of each segment interval, and the current tightening stiffness value is obtained by extrapolation through the stiffness change trend. Construct a stiffness statistical reference distribution based on historical tightening data, calculate the standardized deviation of the current tightening stiffness value relative to the stiffness statistical reference distribution, and generate a confidence index. Based on the confidence index, a stiffness acceptance strategy is set, and the target torque correction amount and tightening speed adjustment amount are dynamically calculated in combination with the current tightening stiffness value, and the corrected tightening process parameters are output.

2. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 1, characterized in that, The step of performing local rate-of-change sign analysis on the torque signal to identify segments with high-frequency alternation of rate-of-change signs, and marking them as viscous-slip oscillation segments, includes: Perform a differential operation on adjacent sampling points of the torque signal to obtain a torque change rate sequence; Extract the positive and negative signs of each element in the torque change rate sequence to generate a sign sequence; A detection window of fixed length is set along the symbol sequence, and the number of flips of adjacent symbols that are different in each window is counted. When the ratio of the number of flips to the length of the sliding window exceeds a preset flip density threshold, the section covered by the sliding window is marked as the viscous-sliding oscillation section. Merge adjacent viscous-sliding oscillation segments and record the start and end sampling indices of each oscillation segment after merging.

3. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 1, characterized in that, The step of extracting the main oscillation period based on the zero-crossing time-series characteristics of the viscous-slip oscillation segment and constructing an adaptive filtering window model includes: Within the viscous-slip oscillation zone, the zero-crossing points of the torque change rate sequence, where it changes from positive to negative or from negative to positive, are detected to form a zero-crossing index sequence. Calculate the index difference between adjacent zero-crossing points in the zero-crossing index sequence, and denote it as the half-cycle sampling number sequence; The median of the half-cycle sampled data sequence is calculated, and twice the median is converted into a time value, which is denoted as the main oscillation period. Multiply the oscillation period by a preset window multiplier to obtain the filter window width; The adaptive filtering window model is constructed by generating a weighted coefficient vector of corresponding length based on the width of the filtering window.

4. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 1, characterized in that, The step of using the adaptive filtering window model to perform weighted smoothing on the torque signal and performing phase lead compensation on the smoothing result to generate a de-oscillation torque trend signal includes: The weighting coefficient vector of the adaptive filtering window model is slid along the torque signal point by point, and the dot product of the sampled value in the window and the weighting coefficient at each position is calculated to obtain the preliminary smoothed signal. The number of sampling points delayed relative to the current time point from the center of the window is calculated based on the width of the filter window, and is denoted as the group delay. The preliminary smoothed signal is shifted forward by the number of points corresponding to the group delay in the sampling index direction to obtain the phase-aligned signal; The missing region caused by the forward shift at the end of the phase alignment signal is filled by linear extrapolation using the end slope to generate the de-oscillation torque trend signal.

5. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 1, characterized in that, The step of performing piecewise energy integration on the de-oscillation torque trend signal and the rotation angle signal to obtain the equivalent stiffness value of each segment interval, and extrapolating the current tightening stiffness value through the stiffness change trend, includes: The angle signal is divided into multiple continuous segmented intervals according to a preset angle step size; The de-oscillation torque trend signal within each segment interval is numerically integrated to obtain the cumulative torque energy value of that segment interval. Divide the cumulative torque energy value of each segment interval by the corresponding angle step to obtain the equivalent stiffness value of each segment interval. Linear regression is performed on the equivalent stiffness values ​​of the most recent consecutive segmented intervals to obtain the slope of stiffness change; The current tightening stiffness value is obtained by multiplying the stiffness change slope by the extrapolation step size and superimposing it on the equivalent stiffness value of the latest segmented interval.

6. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 1, characterized in that, The process of constructing a stiffness statistical reference distribution based on historical tightening data, calculating the standardized deviation of the current tightening stiffness value relative to the stiffness statistical reference distribution, and generating a confidence index includes: Extract a set of stiffness values ​​at the same corner positions from historical tightening records, calculate the mean and standard deviation of the set of values, and construct the stiffness statistical reference distribution. Calculate the absolute value of the difference between the current tightening stiffness value and the mean of the stiffness statistical reference distribution, and divide it by the standard deviation of the stiffness statistical reference distribution to obtain the standardized deviation value; The standardized deviation value is mapped to a preset confidence decay curve to obtain the initial confidence level; The oscillation intensity coefficient is calculated based on the peak-to-peak value of the torque change rate within the viscous-slip oscillation zone described above. The confidence index is generated by multiplying the initial confidence level by a reduction factor that is negatively correlated with the oscillation intensity coefficient.

7. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 1, characterized in that, The step of setting a stiffness acceptance strategy based on the confidence index, dynamically calculating the target torque correction and tightening speed adjustment based on the current tightening stiffness value, and outputting the corrected tightening process parameters includes: Set confidence thresholds and low confidence duration thresholds; When the confidence index is higher than the confidence threshold, the target torque correction is calculated using the current tightening stiffness value; When the confidence index is lower than the confidence threshold but the duration does not exceed the low confidence duration threshold, the stiffness value of the previous sampling time is maintained. When the duration of the low confidence state exceeds the low confidence duration threshold, the tightening speed adjustment amount is reduced by a preset ratio. The target torque correction amount is combined with the tightening speed adjustment amount to output the corrected tightening process parameters.

8. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 3, characterized in that, The step of generating a weighted coefficient vector of corresponding length based on the width of the filtering window includes: Within the viscous-sliding oscillation region, the sampling points are divided into viscous phase sub-segments and sliding phase sub-segments based on the positive or negative sign of the torque change rate. The average number of continuous sampling points for the viscous phase segment and the sliding phase segment are counted respectively, and the ratio between the two is recorded as the phase duration ratio; Using the width of the filter window as the total length, the window is divided into two asymmetric intervals, front and back, according to the phase duration ratio. The ratio of the length of the front interval to the length of the back interval is equal to the phase duration ratio. Independent half-Gaussian decay curves are generated for the first interval and the second interval respectively. The peaks of the two curves are aligned and spliced ​​at the center of the window to form an asymmetric basic weight curve. The mean difference between the amplitudes of the local maxima and local minima of the torque change rate within the viscous-sliding oscillation range is calculated and denoted as the mean oscillation amplitude. The average oscillation amplitude is compared with a preset benchmark oscillation amplitude to obtain the amplitude scaling factor; Multiply the peak height of the asymmetric basic weight curve by the reciprocal of the amplitude scaling factor to make the advantage of the weight center of the window relative to the edge smaller when the oscillation is more violent; Apply a cosine gradient decay to the regions at both ends of the adjusted weight curve with a preset proportion length to eliminate the truncation jump at the window boundary. The weighted curve after boundary gradient processing is integrally normalized to generate the weighted coefficient vector.

9. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 5, characterized in that, The step of performing linear regression on the equivalent stiffness values ​​of the most recent consecutive segmented intervals to obtain the slope of stiffness change includes: Select the equivalent stiffness values ​​of the most recent consecutive preset number of segmented intervals as regression samples; A linear regression equation is constructed with the center corner position of each segment interval as the independent variable and the corresponding equivalent stiffness value as the dependent variable. The slope parameter of the linear regression equation is solved using the least squares method and denoted as the stiffness change slope. Calculate the sum of squared residuals of each regression sample point relative to the regression line. When the sum of squared residuals exceeds a preset residual threshold, remove the sample point with the largest deviation and regress.

10. The adaptive correction method for bolt tightening process parameters based on algorithm analysis according to claim 6, characterized in that, The calculation of the oscillation intensity coefficient based on the peak-to-peak value of the torque change rate within the current viscous-slip oscillation range includes: Extract the maximum and minimum values ​​of the torque change rate sequence within the current viscous-slip oscillation range; Calculate the difference between the maximum value and the minimum value, and record it as the peak-to-peak value of the rate of change; The relative oscillation amplitude is obtained by normalizing the peak-to-peak value of the rate of change by the mean value of the current torque signal. The ratio of the relative oscillation amplitude to the preset oscillation benchmark value is used as the oscillation intensity coefficient.

Images