Automobile part manufacturing process data processing method and system based on artificial intelligence

Abstract

The invention discloses an automobile part manufacturing process data processing method and system based on artificial intelligence, and relates to the technical field of data processing. The automobile part manufacturing process data processing method and system based on artificial intelligence comprises the steps of S1, collecting process record data and process characteristic data of automobile part manufacturing, preprocessing the process record data and the process characteristic data, and constructing a part manufacturing database; s2, fluctuation sensitivity judgment is carried out through the process characteristic data, and signal quality grading, enhancement parameter adjustment and noise suppression processing operation are executed according to the judgment result; s3, integrating the amplitude quantity and the information correlation quantity to construct a cross-channel causal coupling characterization quantity, and triggering the generation of a causal link; and S4, realizing aging state identification by fusing amplitude offset, an attenuation baseline and causal coupling features and combining offset grading analysis and link traceability processing. The problems of synchronous dislocation of multi-channel signals, insufficient perturbation detection precision, difficulty in causal chain reconstruction and difficulty in aging drift identification and tracking are solved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data processing, in particular to an automobile part manufacturing process data processing method and system based on artificial intelligence. BACKGROUND

[0002] For example, the invention with the publication number CN118673579A discloses an automobile part stamping process optimization method and system, relating to the technical field of data processing. The model restoration parameter set is obtained by performing curve parameter extraction on the digital model, the first restoration parameter group is input into the model simulation restoration module to obtain the first simulation restoration model, and the restoration parameter optimization rule is used as a constraint to optimize the restoration parameters of the part to obtain the part parameter optimization result. The raw material size parameters and the part parameter optimization result are analyzed to determine the stamping positioning optimization result. The technical problems that the existing automobile part stamping process optimization relies on multiple rounds of repeated experiments of technical personnel, resulting in a long process optimization time, and the actual performance of the automobile part produced based on the process optimization result is not well adapted to the production demand are solved. The technical effects of improving the automobile part stamping process optimization efficiency and effectiveness, and improving the adaptation of the actual performance of the automobile part to the automobile function demand are achieved.

[0003] However, the above optimization effect often depends on the initial parameters and simulation models input, and it is difficult to fully adapt to the actual demand in the production process, resulting in insufficient adaptation between the actual performance and the production demand. And most of them only focus on single-dimensional monitoring and processing, lacking real-time processing and comprehensive analysis of multi-channel and multi-dimensional data, especially when facing details such as perturbation and lag response, it is difficult to accurately capture and adjust, thereby affecting the optimization effect of the process and the quality stability of the part.

[0004] Therefore, in view of the above problems, an automobile part manufacturing process data processing method and system based on artificial intelligence are urgently needed. SUMMARY

[0005] Technical problems to be solved In view of the deficiencies of the prior art, the application provides an automobile part manufacturing process data processing method and system based on artificial intelligence, which solves the problems of multi-channel signal synchronization misplacement, causality chain reconstruction difficulty and aging drift identification.

[0006] Technical scheme To achieve the above object, the application is implemented by the following technical solutions: the automobile parts manufacturing process data processing method and system based on artificial intelligence, comprising S1: collecting process record data and process characteristic data of automobile parts manufacturing for preprocessing, and storing the process record data and process characteristic data to construct a parts manufacturing database; S2: carrying out fluctuation sensitivity discrimination through the process characteristic data, and performing signal quality grading, enhancement parameter adjustment and noise suppression processing operation according to the discrimination result; S3: integrating amplitude quantity and information correlation quantity to construct cross-channel causal coupling representation, triggering causal link generation, lag response correction and link cascade adjustment; S4: through fusion of amplitude offset, decay baseline and causal coupling characteristics, combined with offset grading analysis and link tracing processing, realizing aging state recognition, offset observation screening and drift risk tracking.

[0007] Further, the specific steps of collecting and preprocessing the process record data and process characteristic data of automobile parts manufacturing are as follows: obtaining process record data and process characteristic data, the process record data including temperature change data, pressure change data, flow change data, torque change data, and multi-channel synchronous recording content; the process characteristic data including channel original amplitude, signal enhancement amplitude, noise power, lag response amplitude, target end receiving amplitude, information correlation, aging baseline amplitude value, and aging fluctuation scale; obtaining temperature change data through a heat-sensitive acquisition channel, obtaining pressure change data through a pressure sensing channel, obtaining flow change data through a mass flow recording channel, obtaining torque change data through a torque monitoring channel, and obtaining multi-channel synchronous recording content through a synchronous trigger signal of a unified acquisition host; performing point value extraction, peak value identification, and energy envelope calculation on the temperature, pressure, flow, and torque recording content to obtain the amplitude main value of each channel at the recording time and obtain the channel original amplitude; performing multi-scale decomposition on the channel original amplitude through wavelet packet transform to extract high-frequency components to obtain the signal enhancement amplitude; performing frequency band decomposition and energy statistics on the fast fluctuation components in a continuous time period, calculating the energy proportion of the background disturbance in the medium frequency band and the high frequency band, and repeatedly sampling in the steady state stage and the no-load stage to estimate the disturbance energy in the recording content to obtain the noise power; performing time series difference analysis and delay characteristic identification on each channel record to calculate the response delay time on the signal transmission path, and then extracting the amplitude main value at the time to obtain the lag response amplitude; performing peak value search and main frequency energy identification on the record curve of the affected channel to extract the amplitude center value at the recording time to obtain the target end receiving amplitude; estimating the joint distribution, edge distribution, and time consistency of the multi-channel recording content, performing entropy calculation and information comparison on the recording content to obtain the information correlation between the records; performing trend fitting, moving smoothing, and curve convergence calculation in a long period sampling sequence to obtain a decay curve that gradually decreases with running time to obtain the aging baseline amplitude value; performing variance statistics, fluctuation amplitude measurement, and energy distribution extraction on the time period of the decay curve to obtain the amplitude range of normal fluctuation in the natural decay process, thereby obtaining the aging fluctuation scale.

[0008] Further, the specific steps of storing the process record data and the process characteristic data to build the part manufacturing database are as follows: after the acquisition of the process record data, map the timestamps of all channels to a unified millisecond time axis, fill in the missing points and eliminate the sampling misplacement by interpolation, do not interpolate and replace the abnormal interval marked by the pre-warning link, only mark and keep the original value; according to the channel and the process batch, calculate the mean value and standard deviation, mark the abnormal points that obviously deviate from the statistical range, and replace the abnormal points with the interpolated values of the adjacent time points before and after, to form smooth and continuous input data; according to the channel, perform standardization processing on the temperature, pressure, flow, torque and aging related quantities, take the mean value of each type of quantity as the center value, and take the standard deviation as the scale, to convert the original value to a dimensionless order of magnitude with zero mean value; then construct the normalization coefficient according to the minimum and maximum values of the channel, linearly map the standardization result to the interval of zero to one, complete the standardization and normalization steps of the process record data; cut a fixed length of data sequence from the continuous record, which is used to calculate the noise energy, amplitude fluctuation and response characteristics, and move it backward by a fixed length to cover the new data sequence, which is a sliding window; define the time from the start of a complete data collection to the end of the disturbance identification, causal reconstruction and aging deviation calculation and all signal processing steps as a detection period; store the process record data and the process characteristic data to build the part manufacturing database, and set the signal record table, the aging record table and the drift record table in the part manufacturing database.

[0009] Further, the specific steps of carrying out fluctuation sensitivity discrimination through the process characteristic data are as follows: obtain the channel original amplitude, signal enhancement amplitude and noise power; multiply the channel original amplitude and the signal enhancement amplitude to obtain the enhanced signal energy, and then sum the noise power and a minimum constant to obtain the effective noise energy, wherein the minimum constant is given as ten to the minus sixth power through offline numerical stability analysis, to avoid the denominator approaching zero in the case that the noise power approaches zero; calculate the perturbation significance value by dividing the enhanced signal energy by the effective noise energy.

[0010] Further, the specific steps of performing signal quality classification, enhancement parameter adjustment and noise suppression processing operation according to the discrimination result are as follows: real-time comparison of perturbation significance value and perturbation threshold value, when the perturbation significance value is less than the perturbation threshold value, continue to collect and record signal data, monitor signal fluctuation, maintain signal sampling frequency, mark qualified signal and archive to signal record table of part manufacturing database; when the perturbation significance value is greater than or equal to the perturbation threshold value, enhance the resolution of the signal to improve the detection accuracy, enhance the definition of the signal by wavelet packet transform; expand the sliding window to reevaluate the noise intensity of the signal, fuse the data through the redundant signal channel, reduce the influence of single channel noise; recalculate the perturbation significance value, when the perturbation significance value is greater than or equal to the perturbation threshold value for continuous n signal sampling periods, generate a warning report, mark the abnormal signal, store the abnormal data with related time stamp, disturbance amplitude, influence process parameter to signal record table of part manufacturing database.

[0011] Further, the specific steps of integrating amplitude and information correlation to construct cross-channel causal coupling representation are as follows: obtaining lag response amplitude, perturbation significance value, target end receiving amplitude and information correlation; calculating the product of lag response coefficient, lag response amplitude and perturbation significance value and taking the absolute value to get the comprehensive influence quantity; calculating the reference amplitude quantity by taking the target end receiving amplitude and the minimum constant, wherein the minimum constant is used to stabilize the denominator scale when the target end reference amplitude approaches zero; calculating the causal correlation strength value by dividing the comprehensive influence quantity by the reference amplitude quantity and multiplying the information correlation.

[0012] Further, the specific steps of triggering causal link generation, lag response correction and link cascade adjustment are as follows: real-time comparison of causal correlation strength value and correlation threshold value, when the causal correlation strength value is less than the correlation threshold value, there is no causal relationship between signals, no intervention is needed, and the current control strategy is maintained; when the causal correlation strength value is greater than or equal to the correlation threshold value, the causal relationship chain between signals is constructed: performing causal test on each signal sequence by lag graph convolution based on AI prior knowledge, obtaining the influence direction between different signals, then drawing the influence direction as a one-way line from the source signal to the target signal to form a clear structure of directed link and construct causal directed graph; based on the causal directed graph, the time series analysis method is used to calculate the lag time between signals to get the time difference value that best reflects the causal transmission, then the signal sequence is time-shifted to generate lag correction sequence, and lag time correction instruction, response amplitude compensation instruction and chain adjustment instruction are obtained; after the lag time is adjusted, output the new causal chain directed link structure, the adjusted time sequence of each signal and the lag response adjustment scheme and archive to the part manufacturing database.

[0013] Further, by fusing the amplitude offset, the attenuation baseline and the causal coupling feature, the specific steps of combining the offset grading analysis and the link traceability processing are as follows: obtaining the channel original amplitude, the aging baseline amplitude value, the causal correlation strength value and the aging fluctuation scale; calculating the difference between the channel original amplitude and the aging baseline amplitude value and performing square operation to obtain the aging offset square term, calculating the product of the absolute size of the causal correlation strength value and the aging fluctuation scale to obtain the causal amplification weight term, adding the aging offset square term and the causal amplification weight term and then taking square root to obtain the aging deviation degree value.

[0014] Further, the specific steps of realizing the aging state recognition, the offset observation screening and the drift risk tracking are as follows: comparing the aging deviation degree value with the deviation threshold value in real time, the deviation threshold value includes T1 and T2; when the aging deviation degree value is less than T1, the signal is marked as an aging label, the aging baseline curve is updated, and the record content is archived to the aging record table of the part manufacturing database for providing long-term stable samples for the perturbation detection link and the causal reconstruction link in the next detection period; when T1 is less than or equal to the aging deviation degree value and less than T2, the current signal is marked as an aging deviation observation object, the aging analysis time window is expanded, the lag correction sequence output by the causal reconstruction is refitted, the aging fluctuation scale is updated, the amplitude tracking is performed along the causal chain, the direction recognition and the upstream correlation degree comparison are performed on the offset interval, the tracking result is marked as a content needing review and is archived to the part manufacturing database; when the aging deviation degree value is greater than or equal to T2, the signal is marked as an abnormal drift risk label, the source traceability is performed along the causal chain, the difference analysis is performed on the propagation path, the amplitude change and the direction relationship of the offset point and the upstream signal, the aging drift abnormal report is generated, and the offset amplitude, the offset time and the related signal are archived to the drift record table of the part manufacturing database.

[0015] Further, the second aspect of the present application provides an automobile part manufacturing process data processing system based on artificial intelligence, which is applied to an automobile part manufacturing process data processing method based on artificial intelligence, and includes: a data acquisition and preprocessing module, which is used for acquiring and preprocessing process record data and process feature data of automobile part manufacturing, and storing the process record data and the process feature data to construct a part manufacturing database; a signal fluctuation recognition enhancement module, which is used for performing fluctuation sensitivity discrimination through the process feature data, and performing signal quality grading, enhancement parameter adjustment and noise suppression processing operations according to a discrimination result; a causal chain construction correction module, which is used for integrating amplitude and information correlation to construct a cross-channel causal coupling representation, and triggering causal link generation, lag response correction and link cascade adjustment; and an aging drift layered recognition module, which is used for fusing the amplitude offset, the attenuation baseline and the causal coupling feature, combining the offset grading analysis and the link traceability processing, and realizing the aging state recognition, the offset observation screening and the drift risk tracking.

[0016] Beneficial effects The present application has the following advantages: (1) The present application can realize cross-channel signal coupling and lag response correction by comprehensive processing and causal chain construction of multi-channel signal data, and further realize automatic identification and correction of complex causal relationship between signals, effectively solving the problem of difficult process adjustment caused by signal delay and error transmission in the prior art.

[0017] (2) The present application realizes early warning and dynamic management of aging and drift risk in the manufacturing process of parts by fusing amplitude offset, attenuation baseline and causal coupling characteristics, combining real-time identification of aging state and drift risk tracking, effectively solving the problem of unable to accurately track and identify aging influence in the prior art.

[0018] (3) The present application realizes efficient identification and optimization of perturbation signals in the manufacturing process by intelligent perturbation analysis and signal quality grading technology, effectively solving the technical bottleneck of being unable to accurately identify and adjust perturbation in complex process environment in traditional technology.

[0019] (4) The present application can greatly improve the quality and reliability of process data by fusion and lag correction based on multi-channel synchronous data, and further realizes accurate synchronization and consistency analysis effect of cross-channel data, effectively solving the problem of unable to effectively integrate multi-channel data in the prior art.

[0020] Of course, any product implementing the present application does not necessarily need to achieve all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The present application is based on artificial intelligence automobile parts manufacturing process data processing method flow chart; Figure 2 The present application is based on artificial intelligence automobile parts manufacturing process data processing system structure diagram; Figure 3 The present application is based on artificial intelligence automobile parts manufacturing process data processing system and method of multi-channel process signal link causal directed graph; Figure 4 The present application is automobile parts aging deviation state classification and drift disposal flow chart; DETAILED DESCRIPTION The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] Please see Figures 1-4 This invention provides a technical solution: an artificial intelligence-based method and system for processing automotive parts manufacturing process data, comprising: S1: collecting process record data and process feature data of automotive parts manufacturing, preprocessing them, and storing the process record data and process feature data to construct a parts manufacturing database; S2: performing fluctuation sensitivity discrimination through process feature data, and performing signal quality grading, enhancement parameter adjustment, and noise suppression processing operations based on the discrimination results; S3: integrating amplitude quantities and information correlation quantities to construct cross-channel causal coupling characterization quantities, triggering causal link generation, hysteresis response correction, and link cascade adjustment; S4: by fusing amplitude offset, attenuation baseline, and causal coupling features, combined with offset grading analysis and link tracing processing, realizing aging state identification, offset observation and screening, and drift risk tracking.

[0023] Specifically, the preprocessing steps for acquiring process record data and process characteristic data in automotive parts manufacturing are as follows: The process of acquiring process record data and process characteristic data is based on precise data acquisition and processing. Process record data includes temperature change data, pressure change data, flow rate change data, torque change data, and multi-channel synchronous recordings. These data reflect the changes of various parameters over time during the process. Temperature change data is obtained through a thermal acquisition channel, pressure change data through a pressure sensing channel, flow rate change data through a mass flow recording channel, and torque change data through a torque monitoring channel. Synchronous recordings from multiple channels are acquired using a unified acquisition host's synchronous trigger signal, ensuring the temporal consistency of data across all channels. Based on the acquired data, the processing of process characteristic data is crucial for subsequent analysis. This includes several key data indicators such as channel original amplitude, noise power, hysteresis response amplitude, target end received amplitude, information correlation, aging baseline amplitude value, and aging fluctuation scale. By performing point value extraction, peak identification, and energy envelope calculation on the temperature, pressure, flow rate, and torque recordings, the principal amplitude values ​​of each channel at the recording time can be obtained, thus deriving the channel's original amplitude. Wavelet packet transform technology is used to decompose the original amplitude of the channels into multiple scales, further extracting high-frequency components to obtain signal enhancement amplitude, thereby improving signal accuracy. To effectively extract noise information, frequency band decomposition and energy statistics are performed on rapidly fluctuating components within continuous time periods. The energy proportion of background disturbances in the mid-frequency and high-frequency bands is calculated, and repeated sampling is performed during steady-state and idle phases to accurately estimate the disturbance energy in the recorded content, thus obtaining the noise power. Timing difference analysis and delay feature identification are performed to determine the response delay time of the signal on the transmission path. The hysteresis response amplitude is then extracted based on the principal amplitude value at that time, enhancing the reliability of signal analysis. By performing peak search and principal frequency energy identification on the recording curves of the affected channels, the center value of the amplitude at the recording time is extracted, thus obtaining the received amplitude at the target end and accurately quantifying the amplitude change of the target signal. By estimating the joint distribution, edge distribution, and temporal consistency of the multi-channel recorded content, entropy calculation and information comparison are performed to finally obtain the information correlation between records, thereby accurately judging the correlation and transmission impact between different signals. To capture the long-term variation trend of the signal, the system performs trend fitting, moving average smoothing, and curve convergence calculations on long-term sampling sequences to obtain an attenuation curve that gradually decreases over time, thereby deriving the aging baseline amplitude value. By performing variance statistics, fluctuation amplitude measurement, and energy distribution extraction on the time period of the attenuation curve, the amplitude range of normal fluctuations during natural attenuation can be obtained, thus obtaining the aging fluctuation scale.

[0024] In this implementation plan, through comprehensive acquisition and processing of process record data and process characteristic data, including real-time monitoring and analysis of key parameters such as temperature, pressure, flow rate, and torque, this step achieves accurate signal extraction, enhancement, and noise suppression. Using techniques such as wavelet packet transform, multi-scale decomposition, and frequency band decomposition, not only are the signal enhancement amplitude, hysteresis response amplitude, and target end received amplitude obtained, but also disturbance components are effectively identified through precise calculation of noise power. Combined with time-series differential analysis and delay feature identification, this step achieves precise extraction of the signal's timing during transmission. Furthermore, the correlation between signals is evaluated through joint analysis of multi-channel data, ensuring efficient signal data processing and accurate analysis, providing reliable data support for subsequent process optimization and risk assessment.

[0025] Specifically, the steps for storing process record data and process feature data to construct a parts manufacturing database are as follows: After acquiring the process record data, a unified millisecond-level time axis mapping is performed on the timestamps of all channels to ensure time synchronization of each channel. Interpolation algorithms are used to fill in missing points and eliminate sampling misalignments, ensuring data continuity and consistency. For warning links marked as abnormal, the original data is retained, and only the abnormal intervals are marked to avoid misoperation. By statistically analyzing the mean and standard deviation of each channel and process batch, abnormal data points that significantly deviate from the statistical range are identified and marked. These abnormal points are replaced with interpolated values ​​from adjacent time points to form smooth and continuous signal input data. For temperature, pressure, flow rate, torque, and aging-related quantities, standardization is performed on each channel, using the mean of each data type as the center value and the standard deviation as the scale to convert the original values ​​into dimensionless quantities with a zero mean. By calculating the minimum and maximum values ​​of each channel, normalization coefficients are constructed, and the standardization results are linearly mapped to the zero-to-one interval, thus completing the standardization and normalization steps of the process record data. A fixed-length data sequence is selected, and noise energy statistics, amplitude fluctuation, and response characteristic analysis are performed. This is then continuously moved forward using a sliding window to cover new data sequences. This data sequence is used for real-time monitoring of signal changes. A detection cycle is defined, starting from the beginning of a complete data acquisition and ending at the completion of disturbance identification, causal reconstruction, aging deviation calculation, and all signal processing steps. All process record data and process characteristic data are stored to construct a part manufacturing database, within which signal recording tables, aging recording tables, and drift recording tables are set up to provide a foundation for subsequent data analysis and anomaly detection.

[0026] In this implementation plan, this step ensures the consistency and integrity of the process record data by performing time synchronization, interpolation, and outlier handling. Standardization and normalization convert the raw signals from each channel into dimensionless standardized data, eliminating dimensional differences and facilitating subsequent analysis. The sliding window technique further aids in identifying noise energy, amplitude fluctuations, and response characteristics, providing stable input data for disturbance identification and signal analysis. The processed process records and process characteristic data are stored in a database, providing data support for subsequent detection, analysis, and anomaly monitoring.

[0027] Specifically, the steps for sensitive fluctuation discrimination using process characteristic data are as follows: First, obtain the original channel amplitude, signal enhancement amplitude, and noise power. Second, select labeled perturbation samples from historical batches. By traversing different decomposition levels and reconstruction weight combinations, compare the consistency between the perturbation significance value and the actual quality fluctuation. Select a weight configuration that balances perturbation amplification and noise suppression capabilities, and label the corresponding weights as wavelet packet transform enhancement factors. Third, multiply the original channel amplitude by the signal enhancement amplitude to obtain the enhanced signal energy, reflecting the change in signal energy after enhancement. Fourth, add the noise power to a minimum constant to obtain the effective noise energy, avoiding computational instability caused by noise power approaching zero. This minimum constant is a power of 10-6, obtained through offline numerical stability analysis. Finally, divide the enhanced signal energy by the effective noise energy to obtain the perturbation significance value, representing the degree of influence of small perturbations in the signal.

[0028] The specific formula for calculating the significance value of perturbation is as follows: ; in, The value represents the significance of the perturbation, indicating the significance of the perturbation between signals and reflecting the degree of influence of small perturbations in the signal; This represents the original amplitude value of the channel, indicating the amplitude of changes in temperature, pressure, flow rate, and torque signals; This indicates the signal enhancement amplitude, representing the signal amplitude enhanced by wavelet packet transform. This represents noise power, indicating the noise intensity in the signal; This represents the wavelet packet transform enhancement factor, indicating the degree of signal enhancement. This represents a very small constant used to avoid division by zero errors, and its value is... .

[0029] In this implementation scheme, the significance value of the perturbation is obtained by calculating the original channel amplitude, signal enhancement amplitude, and noise power, combined with the stable value of the minimum constant. This process effectively enhances the clarity of the signal and ensures that the effective energy of the signal can still be calculated stably even when the noise power is low, thus providing a more accurate basis for signal quality assessment and subsequent processing.

[0030] Specifically, the steps for signal quality grading, enhancement parameter adjustment, and noise suppression based on the discrimination results are as follows: Real-time comparison of the perturbation significance value with the perturbation threshold. When the perturbation significance value is less than the perturbation threshold, signal data continues to be collected and recorded, signal fluctuations are monitored, a stable signal sampling frequency is maintained, and the signal is marked as qualified and archived in the signal record table of the parts manufacturing database to ensure long-term signal stability and traceability. The data will continue to be used for subsequent normal monitoring and analysis without further processing. When the perturbation significance value is greater than or equal to the perturbation threshold, the signal resolution is first enhanced to improve the detection accuracy of the perturbation. Wavelet packet transform is used to enhance the signal and optimize its clarity. The sliding window is expanded to re-evaluate the noise intensity of the signal, and data is fused through redundant signal channels to reduce the interference of single-channel noise on the results. The perturbation significance value is recalculated. If the perturbation significance value is still greater than or equal to the perturbation threshold within n consecutive signal sampling periods, an early warning report is generated and marked as an abnormal signal. n represents the number of consecutive periods, with a value range of [3, 10]. The value of n can be optimized and adjusted according to the sensitivity requirements for perturbation and the fluctuation characteristics of the sampling data to ensure the accuracy and stability of the detection. The relevant abnormal data, timestamps, perturbation amplitudes, and potentially affected process parameters are stored and archived in the signal record table of the part manufacturing database to ensure that abnormal signals are identified and processed in a timely manner, providing data support for subsequent analysis and decision-making.

[0031] In this implementation plan, effective monitoring and anomaly identification of signal data are ensured by comparing the significance value of perturbations with the perturbation threshold in real time. When the signal is normal, data is continuously collected and marked as qualified signals to ensure signal stability and accuracy. For cases where the significance value of perturbations exceeds the perturbation threshold, the perturbation detection accuracy is effectively improved and noise interference is reduced by enhancing signal resolution, expanding the sliding window, reassessing noise intensity, and fusing data through redundant signal channels. By generating early warning reports and storing abnormal data, strong data support is provided for subsequent analysis, decision-making, and risk management.

[0032] Specifically, the steps for constructing a cross-channel causal coupling representation by integrating amplitude and information correlation quantities are as follows: After obtaining the hysteresis response amplitude, perturbation significance value, target-end received amplitude, and information correlation quantity; historical batch records containing multi-channel synchronous records such as temperature, pressure, flow rate, and torque are selected from the existing parts manufacturing database, and least squares regression fitting is performed. The hysteresis response coefficient is obtained based on the relationship between the product of the hysteresis response amplitude and the perturbation significance value and the target-end received amplitude; the comprehensive influence quantity is obtained by multiplying the hysteresis response coefficient, hysteresis response amplitude, and perturbation significance value and taking the absolute value. The reference amplitude quantity is calculated by combining the target-end received amplitude and the minimum constant to ensure that the denominator avoids approaching zero when the target-end reference amplitude is close to zero. The comprehensive influence quantity is divided by the reference amplitude quantity and then multiplied by the information correlation quantity to finally obtain the causal correlation strength value, which provides a quantitative basis for subsequent causal chain construction and signal adjustment.

[0033] The specific formula for calculating the causal relationship strength value is as follows: ; In the formula, Indicates signal and The strength of the causal relationship between signals is measured by the strength of the causal relationship between them. The hysteresis coefficient represents the signal lag response coefficient, reflecting the signal lag response coefficient. For signal The hysteresis response strength; It represents the hysteresis response magnitude, used to reflect the strength of the hysteresis response of a signal along the causal propagation path; Indicates the significance value of the perturbation, representing the signal. and The degree of influence between them; Indicates the received amplitude at the target end, indicating the signal. In time The amplitude at that time reflects the overall strength of the signal; Denotes a minimal constant, taking the value of ; Represents the amount of information associated, and represents the signal. and The dependencies between them.

[0034] The signals from the first detection round (temperature and pressure) were obtained, with a hysteresis response coefficient of 0.85, a hysteresis response amplitude of 2.40, a perturbation significance value of 1.30, a target-end received amplitude of 2.90, and an information correlation value of 0.82. The calculated causal correlation strength was 0.750. The signals from the second detection round (temperature and flow rate) were obtained, with a hysteresis response coefficient of 1.10, a hysteresis response amplitude of 1.90, a perturbation significance value of 0.90, a target-end received amplitude of 2.50, and an information correlation value of 0.60. The calculated causal correlation strength was 0.451. The signals from the third detection round (pressure and flow rate) were obtained, with a hysteresis response coefficient of 1.25, a hysteresis response amplitude of 2.80, a perturbation significance value of 1.60, a target-end received amplitude of 3.00, and an information correlation value of 0.88. The calculated causal correlation strength was 1.643. The signals from the fourth detection wheel are flow rate and torque, with a hysteresis response coefficient of 0.95, a hysteresis response amplitude of 2.10, a perturbation significance value of 1.10, a target-end received amplitude of 2.20, and an information correlation value of 0.57. The calculated causal correlation strength value is 0.569. The signals from the fifth detection wheel are pressure and torque, with a hysteresis response coefficient of 1.30, a hysteresis response amplitude of 2.60, a perturbation significance value of 1.70, a target-end received amplitude of 2.70, and an information correlation value of 0.79. The calculated causal correlation strength value is 1.681. The signals from the sixth detection wheel are temperature and torque, with a hysteresis response coefficient of 0.75, a hysteresis response amplitude of 1.70, a perturbation significance value of 0.80, a target-end received amplitude of 2.00, and an information correlation value of 0.45. The calculated causal correlation strength value is 0.229.

[0035] Table 1. Example data table of causal correlation strength of multi-channel process signals like Figure 3 The diagram shows a causal directed graph of the multi-channel process signal links provided in this embodiment. Table 1 and the graph reveal the mutual influence and transmission paths between different process signals. For example, in detection round 1, the causal correlation strength between temperature and pressure is 0.750, indicating a strong influence of temperature on pressure changes. In round 3, the influence of pressure on flow rate is most significant, with a causal correlation strength of 1.643, indicating a strong driving effect of pressure changes on flow rate changes. Furthermore, the influence of flow rate on torque is relatively weak, with a causal correlation strength of 0.569, indicating a small influence of flow rate on torque changes. The arrow directions and values ​​in the graph illustrate the dependencies and strengths between these signals, clearly revealing how different process parameters interact, thus providing a scientific basis for subsequent process optimization and signal control.

[0036] In this implementation scheme, by acquiring the hysteresis response amplitude, perturbation significance value, target end received amplitude, and information correlation quantity, and based on the calculations of these signal characteristics, the causal correlation strength value is finally obtained. This process effectively assesses the strength of the causal relationship between signals by quantifying the hysteresis response, perturbation impact, and information sharing degree among the signals, providing accurate numerical basis for subsequent causal chain construction, signal optimization, and adjustment, thereby improving the accuracy and reliability of signal processing.

[0037] Specifically, the steps for triggering causal link generation, hysteresis response correction, and link cascading adjustment are as follows: First, by comparing the causal correlation strength value with the correlation threshold in real time, it is determined whether a causal relationship exists between the signals. When the causal correlation strength value is less than the correlation threshold, it indicates that no obvious causal relationship has formed between the signals, therefore no intervention is needed, and the current control strategy is maintained to ensure the natural evolution of the signals. When the causal correlation strength value is greater than or equal to the correlation threshold, causal testing is performed on each signal sequence using hysteresis graph convolution based on AI prior knowledge to identify and determine the causal relationship between different signals and its direction of influence. The direction of influence is drawn as a unidirectional line from the source signal to the target signal, thus constructing a clear causal directed graph. Based on this causal directed graph, time series analysis is used to calculate the hysteresis time between each signal to obtain the time difference value that best reflects the causal transmission. Next, the signal sequence is time-shifted to generate a hysteresis correction sequence. The hysteresis time correction command is written to the controller's time base in milliseconds to ensure synchronous processing. The response amplitude compensation command will be applied to key parameters such as heating temperature, press pressure, and servo motor speed to adjust the behavior of the control system and achieve precise control. By using a chain of adjustment instructions, the coordination between signals is achieved. The adjusted causal chain directed link structure, the adjusted time series of each signal, and the hysteresis response adjustment scheme are output, and the relevant data are archived into the parts manufacturing database to provide a basis for subsequent processing.

[0038] In this implementation scheme, the significance of the causal relationship between signals is determined by comparing the causal association strength value with the association threshold in real time. When the causal association strength value is greater than or equal to the threshold, a causal relationship chain between signals is constructed using AI prior and hysteresis graph convolution techniques to identify the direction of influence between signals and draw a clear causal directed graph. Based on the graph structure, the lag time between each signal is calculated and the signal sequence is adjusted to achieve accurate lag correction and response amplitude compensation. Finally, the adjusted causal chain structure, time series, and response adjustment scheme are output, and the results are archived in a database to provide data support for subsequent analysis and optimization.

[0039] In this implementation, the aging deviation value is obtained by calculating and squaring the difference between the original channel amplitude and the aging baseline amplitude, and combining this with the causal correlation strength value and the aging fluctuation scale. This process effectively quantifies the degree of signal deviation relative to the baseline during the aging process, while considering the strength of the causal relationship between signals, thereby accurately assessing the aging state of the signal. This method provides an important basis for subsequent aging state identification, risk prediction, and maintenance decisions, and helps to improve stability and prediction accuracy.

[0040] Specifically, by fusing amplitude offset, attenuation baseline, and causal coupling characteristics, and combining offset grading analysis with link tracing, the specific steps are as follows: After obtaining the original channel amplitude, aging baseline amplitude value, causal correlation strength value, and aging fluctuation scale, the difference between the original channel amplitude and the aging baseline amplitude value is calculated and squared to obtain the aging offset square term; the absolute value of the causal correlation strength value is multiplied by the aging fluctuation scale to obtain the causal amplification weight term; the aging offset square term and the causal amplification weight term are added together and the square root is taken to obtain the aging deviation value. This process, by quantifying the degree of deviation of the signal from the baseline during the aging process and combining it with the causal relationship strength, can more accurately assess the aging state of the signal, providing an effective basis for subsequent optimization and risk warning.

[0041] The specific formula for calculating the aging deviation value is as follows: ; In the formula, This represents the aging deviation value, reflecting the strength of the signal's offset from the aging baseline at the current moment; This represents the original amplitude of the channel, reflecting the intensity of changes in temperature, pressure, flow rate, and torque signals at time t. The aging baseline amplitude value reflects the natural decay trajectory of the signal over time, and is obtained by trend fitting and smoothing modeling of the long-term sequence. Indicates signal and The strength of the causal relationship between signals is measured by the strength of the causal relationship between them. It represents the scale of aging fluctuations, reflecting the range of fluctuations in the aging trajectory itself.

[0042] Specifically, the steps for achieving aging state identification, offset observation and screening, and drift risk tracking are as follows: By comparing the aging deviation value with the deviation threshold in real time, the aging state of the signal is effectively classified and managed, such as... Figure 4This is a flowchart illustrating the state classification and drift handling of aging deviation in automotive parts according to the present invention. Deviation thresholds include T1 and T2. When the aging deviation value < T1, the signal is marked as an aging label, the aging baseline curve is updated, and the recorded content is archived in the aging record table of the part manufacturing database to provide long-term stable samples for the perturbation detection and causal reconstruction stages in the next detection cycle. When T1 ≤ aging deviation value < T2, the current signal is marked as an aging deviation observation object, the aging analysis time window is expanded, and the hysteresis correction sequence output by causal reconstruction is refitted, the aging fluctuation scale is updated, and amplitude tracking is performed along the causal chain. The direction of the deviation interval is identified and compared with the upstream correlation. The tracking results are marked as content requiring review and archived in the part manufacturing database. When the aging deviation value ≥ T2, the signal is marked with an abnormal drift risk label. By performing topological sorting on each node in the causal chain, the causal relationship transmission direction of the signal is determined, thereby ensuring the clarity of the signal flow. At each causal chain edge, the amplitude gain of the signal path is recorded, representing the increase in signal strength as it passes through the path; simultaneously, the propagation delay is recorded to indicate the time delay for the signal to travel from one node to the next, generating an aging drift anomaly report, and the offset amplitude, offset time, and associated signal are archived together in the drift record table of the part manufacturing database.

[0043] In this implementation scheme, real-time comparison of aging deviation values ​​with deviation thresholds enables the identification and labeling of different signal states. When the aging deviation value is less than T1, the signal is labeled as an aging tag, and the aging baseline is updated, providing a long-term stable sample for future perturbation detection and causal reconstruction. When the deviation value is between T1 and T2, the signal is labeled as an aging deviation observation object. By expanding the analysis time window, the hysteresis correction sequence is refitted, the aging fluctuation scale is updated, and causal chain amplitude tracking and upstream correlation comparison are performed. When the deviation value is greater than or equal to T2, the signal is labeled as an abnormal drift risk tag, and the source is traced along the causal chain to analyze the propagation path, amplitude changes, and direction of signal deviation. Finally, a drift anomaly report is generated, and the relevant data is archived in the drift record table. This process effectively improves the accuracy and reliability of aging state identification, deviation observation, and drift risk tracking.

[0044] Specifically, this embodiment provides an AI-based automotive parts manufacturing process data processing system, applied to an AI-based automotive parts manufacturing process data processing method. It includes: a data acquisition and preprocessing module, a signal fluctuation identification and enhancement module, a causal chain construction and correction module, and an aging drift stratification identification module. The data acquisition and preprocessing module is responsible for collecting process record data and process feature data during the automotive parts manufacturing process, preprocessing this data, and ensuring its storage and management. By processing the process records and process feature data, a parts manufacturing database is constructed to facilitate subsequent data analysis and application. The signal fluctuation identification and enhancement module identifies signal quality by sensitively judging fluctuations in process feature data, performs signal quality grading based on the judgment results, adjusts enhancement parameters, and performs noise suppression to improve signal quality and ensure data accuracy. The causal chain construction and correction module integrates amplitude and information correlation quantities between cross-channel signals to construct causal coupling representation quantities, providing support for subsequent causal link generation and hysteresis response correction. This module triggers the establishment of causal relationships between signals and optimizes the accuracy of signal transmission through link cascading adjustments. The aging drift hierarchical identification module combines amplitude offset, attenuation baseline and causal coupling features to perform offset hierarchical analysis and link tracing, thereby realizing aging state identification. It can not only screen the offset observation objects, but also effectively track drift risks, ensure timely identification of abnormal signal drift, and help avoid potential risks in advance.

[0045] This implementation scheme achieves end-to-end optimization from data acquisition to anomaly drift identification through the collaborative work of multiple modules. The data acquisition and preprocessing module ensures efficient acquisition and storage of process records and feature data, providing a stable data foundation for subsequent analysis. The signal fluctuation identification and enhancement module improves signal quality and reduces noise interference by sensitively identifying fluctuations in process feature data, ensuring data accuracy. The causal chain construction and correction module successfully constructs causal chains by integrating amplitude and information correlation quantities across channels, accurately correcting signal hysteresis responses and providing precise adjustments for signal transmission. The aging drift stratification identification module accurately identifies aging states, filters deviation observation objects, and tracks drift risks by combining amplitude offset and attenuation baseline, ensuring timely detection and resolution of potential signal drift problems. This system efficiently and accurately processes and optimizes process data, ensuring signal quality and stability.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0047] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A data processing method for automotive parts manufacturing processes based on artificial intelligence, characterized in that, Includes the following steps: S1: Collect process record data and process feature data of automotive parts manufacturing, preprocess them, and store the process record data and process feature data to build a parts manufacturing database; S2: Perform fluctuation sensitivity discrimination through process characteristic data, and perform signal quality classification, enhancement parameter adjustment and noise suppression processing based on the discrimination results; S3: Integrate amplitude and information correlation quantities to construct cross-channel causal coupling representation quantities, triggering causal link generation, hysteresis response correction and link cascade adjustment; S4: By integrating amplitude offset, attenuation baseline and causal coupling features, and combining offset hierarchical analysis and link tracing, we can achieve aging state identification, offset observation and screening and drift risk tracking.

2. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for preprocessing the collected process record data and process feature data of automotive parts manufacturing are as follows: Acquire process record data and process characteristic data. Process record data includes temperature change data, pressure change data, flow change data, torque change data, and multi-channel synchronous recording content. The process characteristic data includes the original channel amplitude, signal enhancement amplitude, noise power, hysteresis response amplitude, target end received amplitude, information correlation, aging baseline amplitude value, and aging fluctuation scale. Temperature change data is obtained through the thermal acquisition channel, pressure change data is obtained through the pressure sensing channel, flow change data is obtained through the mass flow recording channel, torque change data is obtained through the torque monitoring channel, and multi-channel synchronous recording content is obtained through the synchronous trigger signal of the unified acquisition host. By performing point value extraction, peak identification, and energy envelope calculation on the recorded data of temperature, pressure, flow, and torque, the principal amplitude values ​​of each channel at the recording time are obtained, thus obtaining the original amplitude of the channel. The original amplitude of the channel is then decomposed into multiple scales by wavelet packet transform to extract high-frequency components and obtain the signal enhancement amplitude. By performing frequency band decomposition and energy statistics on the rapidly fluctuating components over a continuous time period, the energy ratio of background disturbances in the mid-frequency and high-frequency bands is calculated. The noise power is obtained by repeatedly sampling during the steady-state and idle phases and estimating the disturbance energy in the recorded content. By performing timing difference analysis and delay feature identification on each channel record, the response delay time on the signal transmission path is calculated, and then the principal value of the amplitude at that time is extracted to obtain the hysteresis response amplitude. By performing peak search and main frequency energy identification on the recording curves of the affected channels, the amplitude center value at the recording time is extracted to obtain the received amplitude at the target end; By estimating the joint distribution, marginal distribution, and temporal consistency of multi-channel recorded content, entropy calculation and information content comparison are performed on the recorded content to obtain the information correlation between records; By performing trend fitting, moving smoothing and curve convergence calculations on long-term sampling sequences, a decay curve that gradually decreases with running time is obtained, and the aging baseline amplitude value is obtained. By performing variance statistics, fluctuation amplitude measurement, and energy distribution extraction on the time period of the decay curve, the amplitude range of normal fluctuations during natural decay is obtained, thereby obtaining the aging fluctuation scale.

3. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for storing process record data and process feature data to construct a parts manufacturing database are as follows: After acquiring the process record data, the timestamps of all channels are mapped to a unified millisecond-level time axis. Missing points are filled in and sampling misalignments are eliminated through interpolation. Abnormal intervals marked by the warning links are not replaced by interpolation; they are simply marked and the original values ​​are retained. The mean and standard deviation are calculated by channel and process batch. Abnormal points that deviate significantly from the statistical range are marked, and the interpolated values ​​of adjacent time points are used to replace the abnormal points, forming smooth and continuous input data. Temperature, pressure, flow rate, torque, and aging-related quantities are standardized by channel. The mean of each quantity is used as the center value, and the standard deviation is used as the scale to convert the original values ​​into dimensionless quantities with zero mean. Then, normalization coefficients are constructed according to the minimum and maximum values ​​of the channels, and the standardization results are linearly mapped to the zero-to-one interval, completing the standardization and normalization steps of the process record data. A fixed-length data sequence is extracted from continuous records to statistically analyze noise energy, amplitude fluctuations, and response characteristics. This data sequence is then moved forward by a fixed length to cover new data sequences; this data sequence is called a sliding window. A detection cycle is defined as the time from the start of a complete data acquisition to the end of the process of disturbance identification, causal reconstruction, aging deviation calculation, and all signal processing steps. The process record data and process feature data are stored to build a part manufacturing database, and signal record table, aging record table, and drift record table are set in the part manufacturing database.

4. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for conducting fluctuation sensitivity discrimination using process feature data are as follows: Obtain the original channel amplitude, signal enhancement amplitude, and noise power; The enhanced signal energy is obtained by multiplying the original channel amplitude by the signal enhancement amplitude. The effective noise energy is obtained by summing the noise power and the minimum constant. The minimum constant is given by offline numerical stability analysis to be 10 to the power of negative 6, so as to avoid the denominator approaching zero when the noise power is close to zero. The significance value of the perturbation is obtained by calculating the enhanced signal energy and dividing it by the effective noise energy.

5. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for performing signal quality grading, enhancement parameter adjustment, and noise suppression processing based on the discrimination result are as follows: The perturbation significance value and the perturbation threshold are compared in real time. When the perturbation significance value is less than the perturbation threshold, signal data continues to be collected and recorded, signal fluctuations are monitored, the signal sampling frequency is maintained, and the signal is marked as a qualified signal and archived to the signal record table of the part manufacturing database. When the significance value of the perturbation is greater than or equal to the perturbation threshold, the resolution of the enhanced signal is improved to increase the detection accuracy, and the clarity of the signal is enhanced by wavelet packet transform. Expand the sliding window to reassess the noise intensity of the signal, and fuse the data through redundant signal channels to reduce the impact of noise from a single channel; The perturbation significance value is recalculated. When the perturbation significance value is greater than or equal to the perturbation threshold for n consecutive signal sampling periods, an early warning report is generated and marked as an abnormal signal. The abnormal data, along with the relevant timestamp, perturbation amplitude, and affected process parameters, are stored and archived in the signal record table of the part manufacturing database.

6. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for constructing a cross-channel causal coupling representation quantity by integrating the amplitude quantity and the information correlation quantity are as follows: Obtain the hysteresis response amplitude, perturbation significance value, target reception amplitude, and information correlation quantity; The comprehensive impact is obtained by multiplying the lag response coefficient, lag response amplitude, and perturbation significance value and taking the absolute value; the reference amplitude is obtained by calculating the target end reception amplitude and the minimum constant, wherein the aforementioned minimum constant is used to stabilize the denominator when the target end reference amplitude is close to zero; the comprehensive impact is calculated by subtracting the reference amplitude and multiplying it by the information correlation value to obtain the causal correlation strength value.

7. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for triggering causal link generation, hysteresis response correction, and link cascading adjustment are as follows: By comparing the causal correlation strength value and the correlation threshold in real time, when the causal correlation strength value is less than the correlation threshold, it indicates that there is no causal relationship between the signals, no intervention is required, and the current control strategy is maintained. When the causal association strength value is greater than or equal to the association threshold, a causal relationship chain between signals is constructed: causal tests are performed on each signal sequence by hysteresis graph convolution based on AI prior knowledge to obtain the influence direction between different signals, and then the influence direction is drawn as a unidirectional connection from the source signal to the target signal to form a well-structured directed link to construct a causal directed graph. Based on the causal directed graph, the lag time between each signal is calculated using time series analysis to obtain the time difference that best reflects the causal transmission. Then, the signal sequence is time-shifted to generate a lag correction sequence, and lag time correction instructions, response amplitude compensation instructions, and chain adjustment instructions are obtained. After the lag time is adjusted, the new causal chain directed link structure, the adjusted time series of each signal, and the lag response adjustment scheme are output and archived to the parts manufacturing database.

8. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for fusing amplitude offset, attenuation baseline, and causal coupling features, combined with offset hierarchical analysis and link tracing processing, are as follows: Obtain the original channel amplitude, aging baseline amplitude value, causal correlation strength value, and aging fluctuation scale; The difference between the original amplitude of the channel and the amplitude of the aging baseline is calculated and squared to obtain the aging offset square term. The product of the absolute magnitude of the causal correlation strength value and the aging fluctuation scale is calculated to obtain the causal amplification weight term. The aging offset square term and the causal amplification weight term are added together and the square root is taken to obtain the aging deviation value.

9. The method for processing automotive parts manufacturing process data based on artificial intelligence according to claim 1, characterized in that: The specific steps for achieving aging state identification, offset observation and screening, and drift risk tracking are as follows: By comparing the aging deviation value with the deviation threshold in real time, the deviation threshold includes T1 and T2; when the aging deviation value < T1, the signal is marked as an aging label, the aging baseline curve is updated, and the recorded content is archived to the aging record table of the part manufacturing database, which provides long-term stable samples for the next detection cycle's perturbation detection and causal reconstruction stages. When T1≤Aging Deviation Value<T2, the current signal is marked as the object of aging deviation observation, the aging analysis time window is expanded, the hysteresis correction sequence output by the causal reconstruction is refitted, the aging fluctuation scale is updated, and amplitude tracking is performed along the causal chain. The direction of the offset interval is identified and compared with the upstream correlation. The tracking results are marked as the content to be reviewed and archived to the parts manufacturing database. When the aging deviation value is ≥ T2, the signal is marked with an abnormal drift risk label, the source is traced along the causal chain, differential analysis is performed on the propagation path, amplitude change and direction relationship between the offset point and the upstream signal, an aging drift anomaly report is generated, and the offset amplitude, offset time and associated signal are archived together in the drift record table of the part manufacturing database.

10. An AI-based automotive parts manufacturing process data processing system, employing the AI-based automotive parts manufacturing process data processing method according to any one of claims 1-9, comprising: The data acquisition and preprocessing module is used to collect process record data and process feature data of automotive parts manufacturing, preprocess them, and store the process record data and process feature data to build a parts manufacturing database. The signal fluctuation identification and enhancement module is used to perform fluctuation sensitivity discrimination through process feature data, and to perform signal quality classification, enhancement parameter adjustment and noise suppression processing based on the discrimination results. The causal chain construction and correction module is used to integrate amplitude and information correlation quantities to construct cross-channel causal coupling representation quantities, triggering causal link generation, hysteresis response correction and link cascade adjustment. The aging drift stratification identification module is used to identify aging status, observe and screen drift, and track drift risks by fusing amplitude offset, attenuation baseline, and causal coupling features, combined with offset grading analysis and link tracing.

Images