A Torque Testing Method Based on Kalman Filter Algorithm

Abstract

This invention discloses a torque testing method based on the Kalman filter algorithm, belonging to the field of torque testing technology. The combination of the dynamic noise model and the material feature library in this invention enables the filter to automatically adapt to the acoustic characteristics of different substrates, overcoming the misjudgment problem caused by material parameter fluctuations in traditional methods. Secondly, the frequency band processing mechanism completely preserves the transient characteristics before material fracture while suppressing high-frequency noise, avoiding the loss of effective signals. The transfer learning model eliminates batch differences through adversarial training, ensuring the generalization ability of dynamic threshold judgment. The real-time compensation strategy generates a re-tightening scheme that takes into account both mechanical safety and assembly accuracy in emergency situations.

Description

Technical Field

[0001] This invention relates to the field of torque testing technology, and in particular to a torque testing method based on the Kalman filter algorithm. Background Technology

[0002] The current assembly of microelectronic components is evolving towards ultra-precision and heterogeneous material integration. In fields such as smart wearable devices and micro medical devices, brittle substrates such as aluminum nitride ceramics and ultra-thin glass are widely used. In such scenarios, the tightening torque of M0.6 level micro screws often needs to be controlled in the range of 10-50 mN·m, which is equivalent to one percent of the precision range of human finger pinching force.

[0003] Although the emerging miniature servo electric screwdriver system can achieve a resolution of 0.1 mN·m, its torque detection module still faces fundamental challenges when dealing with brittle materials: the critical difference between the material fracture strength and the optimal tightening torque is close to the sensor noise margin, and the traditional signal processing method based on time-domain averaging is prone to misjudgment.

[0004] Current solutions mainly focus on two dimensions: On the one hand, some manufacturers introduce deep learning models, such as LSTM time-series prediction, to identify torque curve features in real time. However, in the early stages of screwing in micro screws, non-stationary noise caused by debris accumulation can cause network weights to oscillate violently. On the other hand, resonance suppression algorithms based on frequency domain analysis, such as wavelet threshold denoising, can weaken high-frequency interference but will erase the transient change features before material fracture. These features often last only 2-5ms and their frequency bands overlap with noise. More seriously, existing systems generally rely on a pre-set material parameter library, while dynamic changes within ±15% of parameters such as substrate thickness and sintering density in actual production will directly lead to mismatch in the pre-trained model. Therefore, a torque testing method is urgently needed to solve these problems. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a torsion testing method based on the Kalman filter algorithm to solve the problems of noise interference and dynamic parameter mismatch in the fastening of micro brittle materials, and the fact that existing filtering methods are prone to losing key transient features and causing misjudgment.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a torque testing method based on the Kalman filter algorithm, which includes:

[0009] Step S1: The original torque signal during the screw insertion process is acquired in real time using a miniature torque sensor;

[0010] Step S2: Dynamically adjust the noise covariance matrix parameters of the Kalman filter based on the material feature library of the current assembly substrate;

[0011] Step S3: Perform wavelet transform frequency segmentation processing on the original torque signal to extract sub-signals that contain at least low-frequency, mid-frequency and high-frequency bands;

[0012] Step S4: Construct independent Kalman filter channels for each frequency band sub-signal, with a dynamic notch filter operator introduced into the high-frequency band channel;

[0013] Step S5: After reconstructing the multi-channel filtering results, input them into the transfer learning model to match the optimal torque decision threshold for the current material in real time.

[0014] Step S6: When the reconstructed signal exceeds the decision threshold, an emergency stop command is triggered and a compensation torque recommendation value is output.

[0015] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, step S2, which involves dynamically adjusting the noise covariance matrix parameters, includes:

[0016] Based on the offline established database of acoustic emission spectrum characteristics of substrate materials and mapping relationship between optimal filtering parameters, the filtering parameters include at least the process noise covariance matrix and the measurement noise covariance matrix. Similarity matching is performed by the acoustic emission signal spectrum characteristics collected online, and the corresponding initial noise covariance matrix is ​​automatically loaded.

[0017] During the filtering process, based on the statistical characteristics of the signal residuals, the exponentially weighted moving average algorithm is used to update the process noise covariance matrix in real time.

[0018] The similarity matching specifically includes:

[0019] Mel-spectral coefficients of the acoustic emission signal in the 50-150kHz frequency band are extracted as feature vectors;

[0020] The cosine similarity algorithm is used to calculate the similarity value between the current feature and each sample in the material library;

[0021] When the maximum similarity value is lower than the preset threshold, the parameter interpolation mode is activated to generate transition parameters.

[0022] In a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, in step S2, the material feature library is constructed as follows:

[0023] Acoustic emission signals and corresponding fracture torque data were collected from materials including at least aluminum nitride ceramics, ultrathin glass, and silicon-based composites.

[0024] The material characteristics are divided into several subclasses using a spectral clustering algorithm;

[0025] Each material type is associated with a set of particle swarm optimized Kalman filter parameters and a safety torque boundary condition.

[0026] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, in step S2, a residual-driven exponential weighting strategy is used to dynamically correct the process noise covariance matrix, enabling the filter to maintain adaptability in a multi-material assembly environment. The update formula is as follows:

[0027]

[0028] Among them, Q k+1 Let α represent the process noise covariance matrix at discrete time k+1. k Q is the exponential decay coefficient at the current time step k. k Let r be the process noise covariance matrix at time step k. k The filter residual vector, For r k The transpose of the matrix;

[0029] The exponential decay coefficient is adaptively adjusted based on the residual energy ratio:

[0030]

[0031] Where β is the smoothing adjustment coefficient, ζ k The residual energy ratio is γ, and the threshold constant is γ.

[0032] The residual energy ratio is defined as:

[0033]

[0034] Where, ε k For instantaneous residual energy, The mean energy of the exponentially weighted residuals;

[0035] Recursive calculation of the mean energy of exponentially weighted residuals:

[0036]

[0037] Where ρ is the residual energy smoothing coefficient. This represents the exponentially weighted mean residual energy of the previous time step;

[0038] Source of residual vector: Among them, z k Let H be the observation vector and H be the observation matrix. Let be the state prediction vector at time step k.

[0039] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, the frequency band processing in step S3 specifically includes:

[0040] A stationary wavelet transform is performed using the db4 wavelet basis to decompose the signal into three levels of detail coefficients and approximation coefficients.

[0041] The pre-trained frequency band identification model is used to determine whether each sub-band belongs to the noise-dominated frequency band.

[0042] An adaptive notch filter is applied to frequency bands identified as noise-dominant, while preserving their phase information;

[0043] The frequency band identification model is constructed as follows:

[0044] Collect a noise sample library containing metal scraps, ceramic fragments, and plastic debris;

[0045] The wavelet packet energy distribution features of each sample are extracted as training data.

[0046] Supervised training is performed using a one-dimensional convolutional neural network classifier.

[0047] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, the high-frequency channel processing in step S4 further includes:

[0048] In the Kalman filter prediction stage, controllable white noise is injected, and the resonant frequency is shifted by adjusting the noise intensity.

[0049] The signal kurtosis index is calculated using a sliding time window, and the system automatically switches to a robust gain matrix when a transient pulse is detected.

[0050] The controllable white noise injection method is as follows:

[0051] A Gaussian white noise sequence is superimposed on the state variables of the Kalman filter prediction equation;

[0052] The noise intensity is proportional to the derivative of the current signal amplitude;

[0053] The noise generator is automatically reset when a resonant frequency deviation of more than 10% is detected.

[0054] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, in step S4, during the process of calculating the signal kurtosis index using a sliding time window, the sliding window method is used to calculate the signal kurtosis and drive gain switching.

[0055]

[0056] Among them, κ mLet s be the kurtosis of the m-th sliding window, L be the window length, and s be the kurtosis. i For the i-th signal sample within the window, μ m This is the mean of the window, and the constant 3 is used to align with the normal distribution baseline;

[0057] The window mean is defined as:

[0058]

[0059] To reduce computational complexity, a central moment recursion is introduced:

[0060]

[0061] M p,m+1 =M p,m +(s m+L -μ m+1 ) p -(s m -μ m ) p ,

[0062]

[0063] Among them, M p,m Let s be the p-th order central moment of the m-th window, where m is the window number. m+L This indicates the newly entered sample when the window moves to the right, s m μ represents the sample that was removed. m+1 This is the average value for new windows.

[0064] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, the transfer learning model in step S5 includes:

[0065] The material feature extraction module is built based on a convolutional neural network, and its input is the time-frequency plot of the wavelet reconstructed signal;

[0066] A cross-material domain adversarial training module is used to eliminate feature distribution differences between different substrate batches;

[0067] The output layer generates dynamic torque thresholds and confidence assessment values.

[0068] As a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, the method further includes a compensation strategy generation step:

[0069] When an emergency stop command is triggered, the optimal torque path is deduced by reversing the historical trajectory of the current filtered state variables.

[0070] A torque compensation curve is generated by combining a screw insertion depth prediction model.

[0071] The output includes a control parameter package containing the compensation torque value, re-tightening angle, and recommended speed.

[0072] In a preferred embodiment of the torque testing method based on the Kalman filter algorithm described in this invention, in the compensation strategy generation step, to provide a feasible and safe re-tightening scheme after triggering an emergency stop, the filter state history before time T is used as a reference. Back-engineering an optimal torque path The steps include:

[0073] First, perform Rauch-Tung-Striebel smoothing on the filtered trajectory:

[0074]

[0075] in, This represents the state vector at step t after smoothing. For filtering estimation, This represents the smoothed state vector at step t+1. To predict the state in one step, A t The smoothing gain matrix;

[0076] The smoothing gain matrix is ​​defined as:

[0077]

[0078] Among them, P t Let F be the filtered covariance, and F be the state transition matrix. To predict covariance in one step;

[0079] Smoothed torque component Based on this, a quadratic objective is constructed that includes tracking error, depth error, and smoothness:

[0080]

[0081] Where J is the cost function, τ t For torque to be optimized, For a smooth torque reference, d t d represents the real-time predicted depth of the screw. T Depth at trigger time These are the first and second order differences of torque, respectively; Δt is the sampling period; and w1, w2, and w3 are weighting coefficients that are applied to torque tracking, depth error, and smoothness terms, respectively.

[0082] Weight adaptive definition:

[0083]

[0084] Where ξ and η are constant coefficients, cT The confidence score output by the transfer learning model;

[0085] Minimize the discrete objective J such that:

[0086]

[0087] The torque optimality condition in differential form is obtained as follows:

[0088]

[0089] in, Let the gradient of the tracking error term be defined, and the boundary conditions be defined. Recursively calculating from t=T towards 0, we obtain:

[0090]

[0091] in, This is the value of the previous moment in the optimal torque sequence. Let it be its first-order difference, denoted by (·). ★ Represents the optimal solution, initial value Get the real-time torque at the time of triggering;

[0092] If the depth deviation at the very end exceeds the limit after reverse calculation, use linear correction:

[0093]

[0094] in, The corrected torque is d0, which is the initial depth. This correction ensures that the compensation curve meets the depth constraint at the end of the path.

[0095] The beneficial effects of this invention are as follows: First, the combination of the dynamic noise model and the material feature library enables the filter to automatically adapt to the acoustic characteristics of different substrates, overcoming the misjudgment problem caused by material parameter fluctuations in traditional methods. Second, the frequency band processing mechanism completely preserves the transient characteristics before material fracture while suppressing high-frequency noise, avoiding the loss of effective signals. Third, the transfer learning model eliminates batch differences through adversarial training, ensuring the generalization ability of dynamic threshold judgment. Fourth, the real-time compensation strategy generates a re-tightening scheme that balances mechanical safety and assembly accuracy in emergency situations. Attached Figure Description

[0096] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0097] Figure 1This is a flowchart illustrating a torque testing method based on the Kalman filter algorithm in Example 1. Detailed Implementation

[0098] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0099] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0100] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0101] Example 1, referring to Figure 1 This embodiment provides a torque testing method based on the Kalman filter algorithm, including the following steps:

[0102] Step S1: The original torque signal during the screw insertion process is acquired in real time using a miniature torque sensor;

[0103] Step S2: Dynamically adjust the noise covariance matrix parameters of the Kalman filter based on the material feature library of the current assembly substrate;

[0104] Step S2 involves dynamically adjusting the noise covariance matrix parameters, including:

[0105] Based on the offline established database of mapping relationship between acoustic emission spectrum characteristics of substrate materials and optimal filtering parameters, the filtering parameters include at least the process noise covariance matrix and the measurement noise covariance matrix. Similarity matching is performed by the acoustic emission signal spectrum characteristics collected online, and the corresponding initial noise covariance matrix is ​​automatically loaded.

[0106] During the filtering process, based on the statistical characteristics of the signal residuals, the exponentially weighted moving average algorithm is used to update the process noise covariance matrix in real time.

[0107] Similarity matching specifically includes:

[0108] Mel-spectral coefficients of the acoustic emission signal in the 50-150kHz frequency band are extracted as feature vectors;

[0109] The cosine similarity algorithm is used to calculate the similarity value between the current feature and each sample in the material library;

[0110] When the maximum similarity value is lower than the preset threshold, the parameter interpolation mode is activated to generate transition parameters;

[0111] In step S2, the material feature library is constructed as follows:

[0112] Acoustic emission signals and corresponding fracture torque data were collected from materials including at least aluminum nitride ceramics, ultrathin glass, and silicon-based composites.

[0113] The material characteristics are divided into several subclasses using a spectral clustering algorithm;

[0114] Each material type is associated with a set of particle swarm optimized Kalman filter parameters and a safety torque boundary condition;

[0115] In step S2, a residual-driven exponential weighting strategy is used to dynamically correct the process noise covariance matrix, enabling the filter to remain adaptive in a multi-material assembly environment. The update formula is as follows:

[0116]

[0117] Among them, Q k+1 Let α represent the process noise covariance matrix at discrete time k+1. k Q is the exponential decay coefficient at the current time step k. k Let r be the process noise covariance matrix at time step k. k The filter residual vector, For r k The transpose of the matrix;

[0118] The exponential decay coefficient is adaptively adjusted based on the residual energy ratio:

[0119]

[0120] Where β is the smoothing adjustment coefficient, ζ k The residual energy ratio is γ, and the threshold constant is γ.

[0121] The residual energy ratio is defined as:

[0122]

[0123] Where, ε k For instantaneous residual energy, The mean energy of the exponentially weighted residuals;

[0124] Recursive calculation of the mean energy of exponentially weighted residuals:

[0125]

[0126] Where ρ is the residual energy smoothing coefficient. This represents the exponentially weighted mean residual energy of the previous time step;

[0127] Source of residual vector: Among them, z k Let H be the observation vector and H be the observation matrix. Let k be the state prediction vector at time step k;

[0128] Specifically, the update formula projects the residual energy into the covariance space, through adaptive α k Control the matrix expansion rate when a sudden change in operating conditions causes ε k During a surge, ζ k The rise causes α k Reduce, introduce larger Weights and covariance broaden rapidly, filtering retains sensitivity to anomalies, and as the system stabilizes, ε k and Approaching, α k Approaching 1, matrix regression is smoothed to avoid gain oscillations, exponential mean suppresses isolated spikes to ensure continuous adjustment process, and overall balances fast response and stable convergence.

[0129] Step S3: Perform wavelet transform frequency segmentation processing on the original torque signal to extract sub-signals that contain at least low-frequency, mid-frequency and high-frequency bands;

[0130] The frequency band division process in step S3 is as follows:

[0131] A stationary wavelet transform is performed using the db4 wavelet basis to decompose the signal into three levels of detail coefficients and approximation coefficients.

[0132] The pre-trained frequency band identification model is used to determine whether each sub-band belongs to the noise-dominated frequency band.

[0133] An adaptive notch filter is applied to frequency bands identified as noise-dominant, while preserving their phase information;

[0134] The frequency band identification model is constructed as follows:

[0135] Collect a noise sample library containing metal scraps, ceramic fragments, and plastic debris;

[0136] The wavelet packet energy distribution features of each sample are extracted as training data.

[0137] Supervised training is performed using a one-dimensional convolutional neural network classifier;

[0138] Step S4: Construct independent Kalman filter channels for each frequency band sub-signal, with a dynamic notch filter operator introduced into the high-frequency band channel;

[0139] The high-frequency channel processing in step S4 further includes:

[0140] In the Kalman filter prediction stage, controllable white noise is injected, and the resonant frequency is shifted by adjusting the noise intensity.

[0141] The signal kurtosis index is calculated using a sliding time window, and the system automatically switches to a robust gain matrix when a transient pulse is detected.

[0142] The controllable white noise injection method is as follows:

[0143] A Gaussian white noise sequence is superimposed on the state variables of the Kalman filter prediction equation;

[0144] The noise intensity is proportional to the derivative of the current signal amplitude;

[0145] The noise generator is automatically reset when a resonant frequency deviation exceeding 10% is detected.

[0146] In step S4, during the process of calculating the signal kurtosis index using a sliding time window, the sliding window method is used to calculate the signal kurtosis and drive gain switching:

[0147]

[0148] Among them, κ m Let s be the kurtosis of the m-th sliding window, L be the window length, and s be the kurtosis. i For the i-th signal sample within the window, μ m This is the mean of the window, and the constant 3 is used to align with the normal distribution baseline;

[0149] The window mean is defined as:

[0150]

[0151] To reduce computational complexity, a central moment recursion is introduced:

[0152]

[0153] M p,m+1 =M p,m +(s m+L -μ m+1 ) p -(s m -μ m ) p ,

[0154]

[0155] Among them, M p,m Let s be the p-th order central moment of the m-th window, where m is the window number. m+L This indicates the newly entered sample when the window moves to the right, s m μ represents the sample that was removed. m+1 This is the average value for new windows;

[0156] Specifically, kurtosis measures fourth-order statistical properties and is highly sensitive to high-amplitude spikes. The recursive formula only requires updating two samples and the mean to complete the central moment iteration. The computational load scales linearly with the sampling rate, satisfying hard real-time constraints. The sliding window length is 3-5 times the resonant principal period, balancing transient capture and background stabilization, enabling real-time κ... m After entering the high-frequency channel logic, when the value exceeds the set threshold, the robust gain matrix is ​​triggered. The filter responds to the impact interference in milliseconds, maintains the integrity of high-frequency information, and provides reliable input for subsequent torque threshold judgment.

[0157] Step S5: After reconstructing the multi-channel filtering results, input them into the transfer learning model to match the optimal torque decision threshold for the current material in real time.

[0158] The transfer learning model in step S5 includes:

[0159] The material feature extraction module is built based on a convolutional neural network, and its input is the time-frequency plot of the wavelet reconstructed signal;

[0160] A cross-material domain adversarial training module is used to eliminate feature distribution differences between different substrate batches;

[0161] The output layer generates dynamic torque thresholds and confidence assessment values.

[0162] Step S6: When the reconstructed signal exceeds the decision threshold, an emergency stop command is triggered and a compensation torque recommendation value is output.

[0163] Torque testing methods also include a compensation strategy generation step:

[0164] When an emergency stop command is triggered, the optimal torque path is deduced by reversing the historical trajectory of the current filtered state variables.

[0165] A torque compensation curve is generated by combining a screw insertion depth prediction model.

[0166] The output includes a control parameter package containing the compensation torque value, the re-tightening angle, and the recommended speed.

[0167] In the compensation strategy generation step, to provide a feasible and safe re-tightening scheme after triggering an emergency stop, the filter state history before time T is used as a basis. Back-engineering an optimal torque path The steps include:

[0168] First, perform Rauch-Tung-Striebel smoothing on the filtered trajectory:

[0169]

[0170] in, This represents the state vector at step t after smoothing. For filtering estimation, This represents the smoothed state vector at step t+1. To predict the state in one step, A t The smoothing gain matrix;

[0171] The smoothing gain matrix is ​​defined as:

[0172]

[0173] Among them, P t Let F be the filtered covariance, and F be the state transition matrix. To predict covariance in one step;

[0174] Smoothed torque component Based on this, a quadratic objective is constructed that includes tracking error, depth error, and smoothness:

[0175]

[0176] Where J is the cost function, τ t For torque to be optimized, For a smooth torque reference, d t d represents the real-time predicted depth of the screw. T Depth at trigger time These are the first and second order differences of torque, respectively; Δt is the sampling period; and w1, w2, and w3 are weighting coefficients that are applied to torque tracking, depth error, and smoothness terms, respectively.

[0177] Weight adaptive definition:

[0178] w1=1+ξc T ,

[0179] Where ξ and η are constant coefficients, c T The confidence score output by the transfer learning model;

[0180] Minimize the discrete objective J such that:

[0181]

[0182] The torque optimality condition in differential form is obtained as follows:

[0183]

[0184] in, Let the gradient of the tracking error term be defined, and the boundary conditions be defined. Recursively calculating from t=R towards 0, we obtain:

[0185]

[0186] in, This is the value of the previous moment in the optimal torque sequence. Let it be its first-order difference, denoted by (·). ★ Represents the optimal solution, initial value Get the real-time torque at the time of triggering;

[0187] If the depth deviation at the very end exceeds the limit after reverse calculation, use linear correction:

[0188]

[0189] in, The corrected torque is d0, which is the initial depth. This correction ensures that the compensation curve meets the depth constraint at the end of the path.

[0190] Specifically, noise is first estimated using the smoother suppression state to ensure the continuity of the subsequent optimization benchmark. Then, torque tracking, depth preservation, and curve smoothing are unified into a single objective function, with weights automatically adjusted according to confidence level, achieving robust and delicate control priority allocation. The recursive formula derived by Euler-Lagrange only involves local gradient calculation and constant coefficient update, making it suitable for real-time reverse stepping. If there is a systematic deviation in the final depth, the linear correction term can fine-tune the torque across the entire segment without destroying the curve smoothness, thereby simultaneously meeting mechanical safety and assembly geometry requirements.

[0191] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A torque testing method based on the Kalman filter algorithm, characterized in that, include, Step S1: The original torque signal during the screw insertion process is acquired in real time using a miniature torque sensor; Step S2: Dynamically adjust the noise covariance matrix parameters of the Kalman filter based on the material feature library of the current assembly substrate; Step S3: Perform wavelet transform frequency segmentation processing on the original torque signal to extract sub-signals that contain at least low-frequency, mid-frequency and high-frequency bands; Step S4: Construct independent Kalman filter channels for each frequency band sub-signal, with a dynamic notch filter operator introduced into the high-frequency band channel; Step S5: After reconstructing the multi-channel filtering results, input them into the transfer learning model to match the optimal torque decision threshold for the current material in real time. Step S6: When the reconstructed signal exceeds the decision threshold, an emergency stop command is triggered and a compensation torque recommendation value is output. It also includes the compensation strategy generation step: To provide a feasible and safe re-tightening solution after an emergency stop is triggered, based on the time... Previous filter state history Back-engineering an optimal torque path The steps include: First, perform Rauch-Tung-Striebel smoothing on the filtered trajectory: ， in, Indicates the smoothed first... Step state vector, For filtering estimation, Indicates the first The smooth state vector of the step, To predict the state in one step, The smoothing gain matrix; The smoothing gain matrix is ​​defined as: ， in, The covariance after filtering. Here is the state transition matrix. To predict covariance in one step; Smoothed torque component Based on this, a quadratic objective is constructed that includes tracking error, depth error, and smoothness: ， in, Let cost function be For torque to be optimized, For smooth torque reference, This is the predicted real-time depth of the screw. Depth at trigger time , These are the first and second order differences of torque, respectively. The sampling period is These are weighting coefficients, which are applied to the torque tracking, depth error, and smoothness terms, respectively. Weight adaptive definition: ， in, and The constant coefficients, The confidence score output by the transfer learning model; For discrete targets Find the minimum such that: ， The torque optimality condition in differential form is obtained as follows: ， in, Let the gradient of the tracking error term be defined, and the boundary conditions be defined. ,since Towards By recursion, we obtain: ， in, This is the value of the previous moment in the optimal torque sequence. Its first difference, sign Represents the optimal solution, initial value Get the real-time torque at the time of triggering; If the depth deviation at the very end exceeds the limit after reverse calculation, use linear correction: ， in, To correct the torque, The initial depth is used as the correction, which ensures that the compensation curve meets the depth constraint at the end of the path.

2. The torque testing method based on the Kalman filter algorithm as described in claim 1, characterized in that, Step S2 involves dynamically adjusting the noise covariance matrix parameters, including: Based on the offline established database of acoustic emission spectrum characteristics of substrate materials and mapping relationship between optimal filtering parameters, the filtering parameters include at least the process noise covariance matrix and the measurement noise covariance matrix. Similarity matching is performed by the acoustic emission signal spectrum characteristics collected online, and the corresponding initial noise covariance matrix is ​​automatically loaded. During the filtering process, based on the statistical characteristics of the signal residuals, the exponentially weighted moving average algorithm is used to update the process noise covariance matrix in real time. The similarity matching specifically includes: Mel-spectral coefficients of the acoustic emission signal in the 50-150kHz frequency band are extracted as feature vectors; The cosine similarity algorithm is used to calculate the similarity value between the current feature and each sample in the material library; When the maximum similarity value is lower than the preset threshold, the parameter interpolation mode is activated to generate transition parameters.

3. The torque testing method based on the Kalman filter algorithm as described in claim 2, characterized in that, In step S2, the material feature library is constructed as follows: Acoustic emission signals and corresponding fracture torque data were collected from materials including at least aluminum nitride ceramics, ultrathin glass, and silicon-based composites. The material characteristics are divided into several subclasses using a spectral clustering algorithm; Each material type is associated with a set of particle swarm optimized Kalman filter parameters and a safety torque boundary condition.

4. The torque testing method based on the Kalman filter algorithm as described in claim 3, characterized in that, In step S2, a residual-driven exponential weighting strategy is used to dynamically correct the process noise covariance matrix, enabling the filter to remain adaptive in a multi-material assembly environment. The update formula is as follows: ， in, Representing discrete time The process noise covariance matrix, For the current time step The exponential decay coefficient, For time step The process noise covariance matrix, The filter residual vector, for The transpose of the matrix; The exponential decay coefficient is adaptively adjusted based on the residual energy ratio: ， in, For smoothing adjustment coefficient, The residual energy ratio, It is a threshold constant; The residual energy ratio is defined as: ， in, For instantaneous residual energy, The mean energy of the exponentially weighted residuals; Recursive calculation of the mean energy of exponentially weighted residuals: ， in, The residual energy smoothing coefficient is... This represents the exponentially weighted mean residual energy of the previous time step; Source of residual vector: ,in, For the observation vector, For the observation matrix, For time step The state prediction vector.

5. The torque testing method based on the Kalman filter algorithm as described in claim 1, characterized in that, The frequency band division process in step S3 is as follows: A stationary wavelet transform is performed using the db4 wavelet basis to decompose the signal into three levels of detail coefficients and approximation coefficients. The pre-trained frequency band identification model is used to determine whether each sub-band belongs to the noise-dominated frequency band. An adaptive notch filter is applied to frequency bands identified as noise-dominant, while preserving their phase information; The frequency band identification model is constructed as follows: Collect a noise sample library containing metal scraps, ceramic fragments, and plastic debris; The wavelet packet energy distribution features of each sample are extracted as training data. Supervised training is performed using a one-dimensional convolutional neural network classifier.

6. The torque testing method based on the Kalman filter algorithm as described in claim 1, characterized in that, The high-frequency channel processing in step S4 further includes: In the Kalman filter prediction stage, controllable white noise is injected, and the resonant frequency is shifted by adjusting the noise intensity. The signal kurtosis index is calculated using a sliding time window, and the system automatically switches to a robust gain matrix when a transient pulse is detected. The controllable white noise injection method is as follows: A Gaussian white noise sequence is superimposed on the state variables of the Kalman filter prediction equation; The noise intensity is proportional to the derivative of the current signal amplitude; The noise generator is automatically reset when a resonant frequency deviation of more than 10% is detected.

7. The torque testing method based on the Kalman filter algorithm as described in claim 6, characterized in that, In step S4, during the process of calculating the signal kurtosis index using a sliding time window, the sliding window method is used to calculate the signal kurtosis and drive gain switching: ， in, For the first The kurtosis of a sliding window, For the length of the window, For the first one inside the window One signal sample, This is the mean of the window, and the constant 3 is used to align with the normal distribution baseline; The window mean is defined as: ， To reduce computational complexity, a central moment recursion is introduced: ， ， ， in, For the first A window The central moment of the first order, For window number, This indicates the newly entered sample when the window moves to the right. Indicates the sample that was removed. This is the average value for new windows.

8. The torque testing method based on the Kalman filter algorithm as described in claim 1, characterized in that, The transfer learning model in step S5 includes: The material feature extraction module is built based on a convolutional neural network, and its input is the time-frequency plot of the wavelet reconstructed signal; A cross-material domain adversarial training module is used to eliminate feature distribution differences between different substrate batches; The output layer generates dynamic torque thresholds and confidence assessment values.

9. The torque testing method based on the Kalman filter algorithm as described in claim 1, characterized in that, In the compensation strategy generation step: When an emergency stop command is triggered, the optimal torque path is deduced by reversing the historical trajectory of the current filtered state variables. A torque compensation curve is generated by combining a screw insertion depth prediction model. The output includes a control parameter package containing the compensation torque value, re-tightening angle, and recommended speed.

Images