Automobile steel wire quality prediction, regulation and control method based on digital twinning

Abstract

The invention relates to the technical field of data processing, in particular to an automobile steel wire quality prediction, regulation and control method based on digital twinning, which comprises the following steps: acquiring multi-source sensing data and drawing speed in the drawing process of an automobile steel wire in real time; determining a preliminary time attention factor of each historical moment in the time period to which the current moment belongs; determining a space-time mapping deviation of each historical moment in the time period to which the current moment belongs; determining an optimized forgetting gate bias item of the Att-BiLSTM at the current moment; and predicting the residual stress of the steel wire at the current moment through the self-adaptive Att-BiLSTM, and realizing real-time regulation and control of the quality of the automobile steel wire according to a prediction result. According to the method, Att-BiLSTM is optimized through the multi-source data and the space-time mapping deviation, parameters are adjusted in a self-adaptive mode, and therefore the digital twinning prediction precision and the steel wire quality control stability under variable-speed drawing are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a method for predicting and controlling the quality of automotive steel wires based on digital twins. Background Technology

[0002] The automotive industry has extremely high requirements for the fatigue strength and toughness of steel wire used in key components such as suspension springs and valve springs. The core of automotive steel wire production lies in the multi-pass straight-in high-speed drawing process. In this process, the wire rod is continuously drawn through a series of dies with decreasing apertures under the traction of the main motor, undergoing intense plastic deformation.

[0003] Currently, the quality control of automotive steel wire mainly relies on offline sampling inspection, which has a significant lag. With the development of industrial internet technology, time series prediction algorithms based on deep learning, such as Attention Bidirectional Long Short-Term Memory Network (Att-BiLSTM), are being applied to online quality prediction. Att-BiLSTM collects data from multiple sensors during the production process, extracts features using an attention mechanism, and makes predictions.

[0004] However, when traditional Att-BiLSTM is applied to automotive wire drawing scenarios, it suffers from several drawbacks. Traditional Att-BiLSTM is based on discrete, equally spaced time steps for feature extraction. It assumes that the weight of information contained in each fixed time sampling interval is determined solely by the numerical fluctuations of the data itself. However, actual wire drawing is not a constant-speed process. To accommodate wire weight changes, weld seam passage, or equipment thermal equilibrium, the drawing speed fluctuates. These speed fluctuations lead to spatiotemporal misalignment: at high speeds, the wire may be drawn to a longer length, and this physical length may contain a significant amount of information related to lattice deformation, frictional heat generation, and stress accumulation. However, since it only occupies one point on the time axis, traditional Att-BiLSTM often underestimates its contribution to the final residual stress due to sparse sampling. Under low-speed conditions, the steel wire may be pulled for a shorter length. Although the physical deformation is small, it has a long duration on the time axis and dense sampling points. Traditional Att-BiLSTM tends to give it excessive attention, leading to overfitting of the model. This results in a mismatch between the time domain and the physical space domain due to speed fluctuations. As a result, the digital twin model cannot correctly restore the true accumulation path of residual stress, causing the prediction accuracy to decrease during the speed change stage, which in turn affects the reliability of subsequent control over the quality of automotive steel wire. Summary of the Invention

[0005] To address the technical problem of low residual stress prediction accuracy caused by information mismatch between the time domain and physical space domain due to drawing speed fluctuations in traditional Att-BiLSTM, which in turn affects the reliability of subsequent quality control of automotive steel wire, this invention proposes a digital twin-based method for predicting and controlling the quality of automotive steel wire. This method includes the following steps: Multi-source sensor data and drawing speed are collected in real time during the drawing process of automotive steel wire. Based on the multi-source sensor data, the attention weights of each historical moment are obtained using an Att-BiLSTM trained on a model. According to the attention weights, the preliminary time attention factors of each historical moment within the current time period are determined. Based on the preliminary time attention factors, the difference between the drawing speed of each historical moment within the current time period and the set drawing speed of the corresponding process segment, and the cumulative distance the steel wire has moved from each historical moment within the current time period to the current moment, the spatiotemporal mapping deviation of each historical moment within the current time period is determined. Based on the baseline bias term of Att-BiLSTM, the spatiotemporal mapping deviation, and the drawing speed of each historical moment within the current time period, the optimized forget gate bias term of Att-BiLSTM at the current moment is determined. Based on the optimized forget gate bias term, Att-BiLSTM is dynamically updated. The updated Att-BiLSTM is used to predict the residual stress of the steel wire at the current moment, and the quality of automotive steel wire is controlled in real time based on the prediction results.

[0006] The beneficial effects are as follows: By analyzing multi-source sensor data to obtain attention weights, and combining the differences in drawing speed and the cumulative movement distance of the steel wire, an evaluation model for spatiotemporal mapping deviation is constructed, which effectively solves the spatiotemporal misalignment problem of traditional Att-BiLSTM under variable speed conditions and achieves accurate mapping between the time domain and the physical space domain; by dynamically adjusting the forget gate bias term based on the spatiotemporal mapping deviation, adaptive optimization of the Att-BiLSTM model parameters is achieved, ensuring that physical deformation information under high-speed conditions is fully considered and redundant information under low-speed conditions is effectively suppressed; this invention effectively improves the prediction accuracy of the digital twin model in the variable speed drawing process, accurately restores the true accumulation path of residual stress, and improves the reliability and stability of automotive steel wire quality control through real-time prediction and quality control.

[0007] Furthermore, the multi-source sensing data includes: vibration data collected by a high-frequency vibration sensor deployed at the final die box; load current data collected by a Hall current sensor deployed at the drum drive motor end; and wire exit temperature data collected using a dual-wavelength infrared thermometer.

[0008] Furthermore, the multi-source sensor data and pull-out speed are preprocessed multi-source sensor data and pull-out speed. The preprocessing includes: performing timestamp alignment processing on the multi-source sensor data and pull-out speed; using a Kalman filter to denoise the timestamp-aligned data; and performing Z-score normalization processing on the denoised data to obtain preprocessed standard time series multi-source sensor data and pull-out speed.

[0009] Furthermore, the Att-BiLSTM trained by the model also includes: periodically collecting residual stress fingerprint signals on the surface of the steel wire using an online magnetoelastic instrument as preliminary data for training labels; performing offline calibration on the preliminary data, including cutting out steel wire segments containing residual stress fingerprint signals, using X-ray diffraction to determine the true value of residual stress inside the steel wire segments, and constructing a surface and internal stress mapping database based on the true value of residual stress and the residual stress fingerprint signals, and using the preliminary data of training labels for model training after correction.

[0010] Furthermore, the preliminary time focus factor satisfies: In the formula, For the current moment Within the time period Initial time-related factors for a historical moment. For the current moment Within the time period The attention weight of each historical moment For the current moment Within the time period The attention weight of each historical moment The set length of the local mutation analysis time window. For the maximum-minimum normalization function, It is the absolute value symbol.

[0011] The beneficial effects are as follows: by constructing a composite function that includes attention weights and local change rates, a preliminary assessment of time attention factors is achieved. It comprehensively considers the attention intensity and local attention change characteristics at the current moment, and can more accurately reflect the importance of historical moments in the time series. The normalization process ensures the reasonable range of factor values, and provides a reliable weight basis for subsequent calculation of spatiotemporal mapping deviation.

[0012] Furthermore, the spatiotemporal mapping deviation satisfies: In the formula, For the current moment Within the time period Spatiotemporal mapping deviation at a historical moment For the current moment Within the time period Initial time-related factors for a historical moment. For the current moment Within the time period The pulling speed at a historical moment For the current moment Within the time period The set drawing speed for the process segment corresponding to each historical moment. For the current moment Within the time period From a historical moment to the present moment The cumulative distance the steel wire moves over time. To prevent hyperparameters with a denominator of 0, It is a natural exponential function.

[0013] The beneficial effects are as follows: by constructing a composite function that includes a time concern factor, a speed difference ratio, and a distance exponent, the accurate assessment of spatiotemporal mapping deviation is achieved. The speed difference ratio accurately reflects the degree of deviation between the actual pulling speed and the set speed, and the distance exponent reflects the influence of cumulative distance on the deviation. Thus, it is possible to effectively correct the problem of mismatch between the time domain and the physical space domain caused by speed fluctuations.

[0014] Furthermore, the optimized forget gate bias term satisfies: In the formula, For Att-BiLSTM at the current time Optimize the forget gate bias term. For the reference bias term of Att-BiLSTM, The preset adjustment gain, For the current moment The length of the time period, For the current moment Within the time period Spatiotemporal mapping deviation at a historical moment For the current moment Within the time period The pulling speed at a historical moment It is the maximum-minimum normalization function.

[0015] The beneficial effects are as follows: by constructing an optimization model that includes a weighted sum of baseline bias and spatiotemporal mapping deviation, the adaptive adjustment of the forget gate bias term is achieved, the logarithmic term of the pulling speed ensures the reasonable weighting of the speed influence, and the normalization process ensures the numerical stability of the deviation term. Thus, the model parameters can be dynamically optimized according to the historical spatiotemporal mapping deviation, thereby improving the adaptability and prediction accuracy of Att-BiLSTM under variable speed conditions.

[0016] Furthermore, the real-time control of the quality of automotive steel wire includes: in response to the residual stress of the steel wire being greater than a preset threshold, adjusting the pressing amount of the straightening roller to apply reverse stress; if the pressing amount reaches the upper limit and the residual stress of the steel wire at the corresponding moment is still greater than the preset threshold, then reducing the drawing speed during the drawing process of automotive steel wire.

[0017] The beneficial effects are as follows: by constructing a graded control strategy, the residual stress of the steel wire is precisely controlled. When the residual stress exceeds the preset threshold, the amount of the straightening roller reduction is adjusted first to apply reverse stress. This method can respond quickly and effectively reduce the residual stress. When the reduction reaches the upper limit and still cannot meet the requirements, the stress accumulation is fundamentally reduced by reducing the drawing speed, thus ensuring the effectiveness of quality control and the rationality of process parameters.

[0018] Furthermore, the preset threshold is 200 MPa.

[0019] Furthermore, the reduction of the drawing speed during the drawing process of automotive steel wire includes: reducing the drawing speed at a rate of 5% per second until the residual stress of the steel wire at the corresponding moment is not greater than the preset threshold.

[0020] The present invention has the following beneficial effects: (1) To address the problems of sparse sampling and underestimation of contribution in high-speed drawing and dense sampling and overemphasis in low-speed drawing caused by long time intervals in traditional Att-BiLSTM, this invention calculates the spatiotemporal mapping deviation by integrating drawing speed, cumulative wire movement distance and preliminary time attention factor, and more accurately assesses the mismatch between time sampling interval and physical deformation. In addition, the forget gate bias term is dynamically optimized by drawing speed. Under high-speed conditions, the optimized bias term will strengthen the feature weights of the corresponding historical time, avoiding the underestimation of key information such as lattice deformation and stress accumulation due to few sampling points. Under low-speed conditions, the bias term will weaken the interference of redundant sampling points, prevent overemphasis on non-critical information, effectively solve the adaptation contradiction between the time domain and the physical space domain, and enable the model to more accurately match the real physical process of wire drawing.

[0021] (2) By optimizing the forget gate bias term to dynamically update Att-BiLSTM, the model is no longer limited by a fixed time step. It can autonomously adjust the memory and forgetting strategies for different historical moments according to the fluctuation of the drawing speed, capture the coupling relationship between speed, deformation and stress, and completely restore the real accumulation path of residual stress in the variable speed stages such as wire rod weight change and weld seam passage. It avoids the prediction deviation caused by the spatiotemporal misalignment of the traditional model, provides reliable data support for the digital twin model, and ensures that the digital twin can more accurately map the physical drawing process and improve the consistency between virtual simulation and actual working conditions.

[0022] (3) Accurate prediction results of residual stress in steel wire provide a scientific basis for real-time control of automotive steel wire quality. When the model predicts that the residual stress is close to the threshold, the process parameters can be adjusted in time to avoid quality problems such as excessive residual stress and insufficient steel wire toughness caused by prediction deviation during the speed change stage. Compared with the traditional method, which is delayed or misadjusted due to insufficient prediction accuracy, this invention can stably output reliable prediction results under speed change conditions, ensure the timeliness and accuracy of quality control, reduce the risk of use such as fracture and fatigue caused by abnormal residual stress in steel wire, and protect the core performance of automotive steel wire. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the steps of a method for predicting and controlling the quality of automotive steel wires based on digital twins, according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the prediction results of residual stress in steel wire using the traditional Att-BiLSTM method, which is a digital twin-based method for predicting and controlling the quality of automotive steel wires according to an embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram of the prediction results of residual stress in automotive steel wire using an adaptive Att-BiLSTM method based on digital twin for predicting and controlling the quality of automotive steel wire, according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below. The described embodiments are only a part of the embodiments of the present invention. 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.

[0027] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] Please see Figure 1 The diagram illustrates a flowchart of a method for predicting and controlling the quality of automotive steel wires based on digital twins, according to an embodiment of the present invention. The method includes the following steps: S001: Real-time acquisition of multi-source sensor data and pulling speed during the steel wire pulling process of automobiles.

[0029] Specifically, a high-frequency data acquisition network is constructed on the straight-line wire drawing unit, including: installing a high-frequency vibration sensor at the final die box to capture the frictional contact state between the die and the steel wire at a sampling rate of 10kHz; deploying a Hall current sensor at the drum drive motor end to collect the load current; using a dual-wavelength infrared thermometer to monitor the steel wire exit temperature in real time, obtaining multi-source sensor data during the automotive steel wire drawing process; and installing a 24-bit high-precision absolute encoder at the traction wheel axle end to collect drawing speed data in real time with a time resolution of 1ms. Simultaneously, the multi-source sensor data and drawing speed are timestamped, and the timestamped data is denoised using a Kalman filter. The denoised data is then Z-score normalized to obtain pre-processed standard time-series multi-source sensor data and drawing speed.

[0030] S002: Determine the preliminary time focus factors for each historical moment within the time period to which the current moment belongs.

[0031] It should be noted that this step evaluates the importance of each historical moment from a data statistics perspective. In the attention mechanism, the magnitude of the weight value reflects the contribution of the information at that moment to the prediction result. However, in a continuous production process such as wire drawing, a high weight at a single moment may only be a manifestation of instantaneous sensor noise and has no process guidance significance. Truly valuable process events, such as the lubrication failure initiation point or the temperature rise mutation point, are usually manifested as local drastic fluctuations in weight. That is, the weight at that moment is not only high in absolute value, but also undergoes a significant mutation relative to the time steps before and after it.

[0032] Based on multi-source sensor data, the attention weights of each historical moment are obtained using an Att-BiLSTM trained by the model. Based on the attention weights, the preliminary time attention factors of each historical moment within the time period to which the current moment belongs are determined.

[0033] Implementers can set the length of the time period to which the current moment belongs, for example, 50 historical moments, depending on the specific implementation situation.

[0034] Specifically, the Att-BiLSTM trained by the model further includes: The residual stress fingerprint signal on the surface of the steel wire was periodically collected using an online magnetoelastic instrument as preliminary data for training labels; The preliminary data is calibrated offline, including extracting a steel wire segment containing residual stress fingerprint signals, using X-ray diffraction to determine the true value of residual stress inside the steel wire segment, and constructing a surface and internal stress mapping database based on the true value of residual stress and the residual stress fingerprint signals. The preliminary data of the training labels is then corrected and used for model training.

[0035] Specifically, the initial time focus factor satisfies: ; In the formula, For the current moment Within the time period Initial time-related factors for a historical moment. For the current moment Within the time period The attention weight of each historical moment For the current moment Within the time period The attention weight of each historical moment The length of the set local mutation analysis time window, for example, , For the maximum-minimum normalization function, It is the absolute value symbol.

[0036] in, This reflects the fundamental importance of the moment of selection; Reflects the first The mutation energy of the weight of the historical moment; the larger this value, the more significant the change. The greater the probability that a technological event disrupting the stable state occurred at a historical moment, the greater its value in time series analysis, and the greater the corresponding initial time focus factor.

[0037] S003: Determine the spatiotemporal mapping deviation of each historical moment within the time period to which the current moment belongs.

[0038] It should be noted that during high-speed drawing, the lattice slip inside the steel wire is intense, resulting in extremely high deformation energy. However, due to the sparse sampling points, the initial time attention factor is often too small and may be underestimated. During low-speed drawing, the physical deformation is weak, but due to the dense sampling points, the initial time attention factor is often too high and may be overestimated. Therefore, the physical dimension can be corrected by introducing the drawing speed. Since the plastic deformation work of the steel wire is highly correlated with the square of the speed, i.e., the kinetic energy, if the speed at a certain moment is much higher than the set speed of the corresponding process segment, it indicates that the possibility of physical deformation information contained at that moment is greater, and its weight should be greater. At the same time, the greater the physical distance between the deformation history and the current moment, the smaller the impact on the current residual stress.

[0039] Based on the preliminary time focus factor, the difference between the drawing speed of each historical moment within the current time period and the set drawing speed of the corresponding process segment, and the cumulative distance the steel wire has moved from each historical moment within the current time period to the current moment, the spatiotemporal mapping deviation of each historical moment within the current time period is determined.

[0040] Specifically, the spatiotemporal mapping deviation satisfies: ; In the formula, For the current moment Within the time period Spatiotemporal mapping deviation at a historical moment For the current moment Within the time period Initial time-related factors for a historical moment. For the current moment Within the time period The pulling speed at a historical moment For the current moment Within the time period The set drawing speed for the process segment corresponding to each historical moment. The current moment after Z-score normalization. Within the time period From a historical moment to the present moment The cumulative distance the steel wire moves over time. For example, to prevent hyperparameters with a denominator of 0, , It is a natural exponential function.

[0041] in, The larger the value, the more likely it is that the time-domain analysis based on pure data-driven approaches considers the first... The more important the information at a historical moment, the greater the corresponding spatiotemporal mapping deviation. The larger the value, the more likely it is to be the first. The faster the real-time drawing speed at a given historical moment is relative to the average speed, the greater the physical deformation kinetic energy experienced by the steel wire within that time step and the higher the density of physical information contained therein. In order to correct the bias caused by the sparse sampling points under high-speed conditions, its weight needs to be amplified, which will result in a larger spatiotemporal mapping deviation. The larger the value, the more likely it is to be the first. The further the cumulative physical deformation path is from the current moment in the historical time, the greater the spatial memory decay characteristic during the residual stress generation process. The weaker the residual influence of deformation effects generated at a historical moment on the current internal stress state, the smaller the corresponding spatiotemporal mapping deviation.

[0042] S004: Determine the optimized forget gate bias term for Att-BiLSTM at the current time.

[0043] It should be noted that after obtaining the spatiotemporal mapping deviation for each historical time point, this step calculates the forget gate bias term of the Att-BiLSTM at the current time based on the spatiotemporal mapping deviation. In Att-BiLSTM, the forget gate determines how much historical information is retained. If the cumulative spatiotemporal mapping deviation within the recent time window is large, it indicates that the steel wire has likely just undergone a high-speed, drastic deformation process with critical technological abrupt changes. To accurately calculate the cumulative residual stress generated by this process, the model needs to remember this historical period and cannot allow it to decay naturally over time. Therefore, a positive bias needs to be added to the forget gate for optimization to keep it open, i.e., the output should be close to 1.

[0044] Based on the baseline bias term of Att-BiLSTM, the spatiotemporal mapping deviation, and the pull-out speed of each historical time within the current time period, the optimized forget gate bias term of Att-BiLSTM at the current time is determined.

[0045] Specifically, the optimized forget gate bias term satisfies: ; In the formula, For Att-BiLSTM at the current time Optimize the forget gate bias term. For example, the reference bias term for Att-BiLSTM. , For example, the preset adjustable gain. , For the current moment The length of the time period, For the current moment Within the time period Spatiotemporal mapping deviation at a historical moment For the current moment Within the time period The pulling speed at a historical moment It is the maximum-minimum normalization function.

[0046] in, The larger the value, the higher the value after spacetime mapping correction. The more significant the contribution of the physical deformation information carried by each historical moment to the cumulative residual stress, the larger the corresponding optimized forget gate bias term at the current moment should be to ensure that the model can capture and retain key physical evolution features. The larger the value, the more likely it is to be the first. The higher the deformation kinetic energy and the more active the physical processes at each historical moment, the more intense the deformation process. The greater the weight of a historical moment in the memory retention decision, the larger the corresponding optimization forget gate bias term for the current moment. The larger the value, the more intense and effective the physical deformation process that occurred within the recent historical window. In order to prevent the model from erroneously truncating important physical accumulation processes due to the natural passage of time steps, it is necessary to force the forget gate to remain open. Therefore, the larger the corresponding optimized forget gate bias term at the current moment, the larger the value.

[0047] S005: The residual stress of the steel wire at the current moment is predicted by adaptive Att-BiLSTM, and the quality of the automotive steel wire is controlled in real time based on the prediction results.

[0048] Based on dynamically updating the Att-BiLSTM by optimizing the forget gate bias term, the updated Att-BiLSTM is used to predict the residual stress of the wire at the current moment, such as... Figure 2 and Figure 3 As shown, this paper compares the prediction capabilities of traditional Att-BiLSTM and adaptive Att-BiLSTM for residual stress in steel wire under the same high-speed, sudden-change operating conditions. The actual stress line represents the real stress value in actual production. It can be seen that the actual stress line fluctuates violently, containing multiple large peaks and troughs, which shows the complexity of the operating conditions. Because traditional Att-BiLSTM often outputs a smooth average value in order to minimize the overall error when facing rapidly changing signals, it lags in responding to sudden-change signals, making the predicted line of the traditional algorithm almost a straight line with a large prediction error. In contrast, the adaptive Att-BiLSTM of this invention dynamically adjusts the forget gate bias, enabling the model to accurately capture every detail and quickly output the predicted value. This makes the predicted line of the adaptive algorithm almost completely overlap with the actual stress line, thus verifying the effectiveness of this invention.

[0049] Specifically, the real-time control of the quality of automotive steel wire includes: In response to the residual stress of the steel wire being greater than a preset threshold, the amount of pressure applied by the straightening roller is adjusted to apply reverse stress. If the residual stress of the steel wire is still greater than the preset threshold when the amount of pressure is increased to the upper limit, the drawing speed during the drawing process of the automotive steel wire is reduced.

[0050] Specifically, the preset threshold is 200 MPa.

[0051] Specifically, reducing the drawing speed during the drawing process of automotive steel wire includes: The drawing speed is reduced by 5% per second until the residual stress of the steel wire at the corresponding moment is no greater than the preset threshold.

[0052] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting and controlling the quality of automotive steel wire based on digital twins, characterized in that, include: Real-time acquisition of multi-source sensor data and drawing speed during the steel wire drawing process of automobiles; Based on multi-source sensor data, the attention weights of each historical moment are obtained using the Att-BiLSTM trained by the model. Based on the attention weights, the preliminary time attention factors of each historical moment within the time period to which the current moment belongs are determined. Based on the preliminary time focus factor, the difference between the drawing speed of each historical moment within the current time period and the set drawing speed of the corresponding process segment, and the cumulative distance the steel wire has moved from each historical moment within the current time period to the current moment, the spatiotemporal mapping deviation of each historical moment within the current time period is determined. Based on the baseline bias term of Att-BiLSTM, the spatiotemporal mapping deviation, and the pull-out speed of each historical time within the current time period, determine the optimized forget gate bias term of Att-BiLSTM at the current time. By dynamically updating the Att-BiLSTM based on the optimized forget gate bias term, the updated Att-BiLSTM is used to predict the residual stress of the steel wire at the current moment, and the quality of the automotive steel wire is controlled in real time based on the prediction results.

2. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The multi-source sensing data includes: Vibration data is collected by deploying a high-frequency vibration sensor at the last pass mold box; Load current data is collected by deploying a Hall current sensor at the end of the drum drive motor; The temperature data of the steel wire exiting the mold was collected using a dual-wavelength infrared thermometer.

3. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The multi-source sensor data and pull-out speed are pre-processed multi-source sensor data and pull-out speed, and the pre-processing includes: Timestamp alignment is performed on multi-source sensor data and pull-out speed; Use a Kalman filter to denoise the timestamp-aligned data; The denoised data is Z-score normalized to obtain preprocessed standard time series multi-source sensor data and pull-out speed.

4. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The Att-BiLSTM trained by the model also includes: The residual stress fingerprint signal on the surface of the steel wire was periodically collected using an online magnetoelastic instrument as preliminary data for training labels; The preliminary data is calibrated offline, including extracting a steel wire segment containing residual stress fingerprint signals, using X-ray diffraction to determine the true value of residual stress inside the steel wire segment, and constructing a surface and internal stress mapping database based on the true value of residual stress and the residual stress fingerprint signals. The preliminary data of the training labels is then corrected and used for model training.

5. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The preliminary time focus factor satisfies: ； In the formula, For the current moment Within the time period Initial time-related factors for a historical moment. For the current moment Within the time period The attention weight of each historical moment For the current moment Within the time period The attention weight of each historical moment The set length of the local mutation analysis time window. For the maximum-minimum normalization function, It is the absolute value symbol.

6. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The spatiotemporal mapping deviation satisfies: ； In the formula, For the current moment Within the time period Spatiotemporal mapping deviation at a historical moment For the current moment Within the time period Initial time-related factors for a historical moment. For the current moment Within the time period The pulling speed at a historical moment For the current moment Within the time period The set drawing speed for the process segment corresponding to each historical moment. For the current moment Within the time period From a historical moment to the present moment The cumulative distance the steel wire moves over time. To prevent hyperparameters with a denominator of 0, It is a natural exponential function.

7. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The optimized forget gate bias term satisfies: ； In the formula, For Att-BiLSTM at the current time Optimize the forget gate bias term. For the reference bias term of Att-BiLSTM, The preset adjustment gain, For the current moment The length of the time period, For the current moment Within the time period Spatiotemporal mapping deviation at a historical moment For the current moment Within the time period The pulling speed at a historical moment It is the maximum-minimum normalization function.

8. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 1, characterized in that, The real-time control of automotive steel wire quality includes: In response to the residual stress of the steel wire being greater than a preset threshold, the amount of pressure applied by the straightening roller is adjusted to apply reverse stress. If the residual stress of the steel wire is still greater than the preset threshold when the amount of pressure is increased to the upper limit, the drawing speed during the drawing process of the automotive steel wire is reduced.

9. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 8, characterized in that, The preset threshold is 200 MPa.

10. The method for predicting and controlling the quality of automotive steel wire based on digital twins according to claim 8, characterized in that, The reduction of the drawing speed during the automotive steel wire drawing process includes: The drawing speed is reduced by 5% per second until the residual stress of the steel wire at the corresponding moment is no greater than the preset threshold.

Images