Stamping depth dynamic compensation control method based on press slider displacement closed-loop feedback

Abstract

This invention relates to a dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement, aiming to solve the technical problem of difficult accurate identification and fine compensation of multi-source disturbances during the operation of stamping equipment. Its core solution includes: acquiring slide motion data through a high-frequency displacement sensor; generating a two-dimensional spectrum of multi-source disturbances using a combination of short-time Fourier transform and continuous wavelet transform; constructing a time-frequency semantic label system characterizing typical process anomalies to achieve automated identification of disturbance types; dynamically generating a multi-dimensional disturbance confidence vector by combining contextual process metadata and a historical knowledge base, mapping it to a response matrix of the compensation strategy, and then calculating and outputting safe and compatible multi-strategy dynamic compensation control commands. This method improves the interpretability and accuracy of slide displacement anomaly identification, achieves fractional compensation for disturbances such as springback and material thickness fluctuations, and enhances the adaptive robustness of the stamping process and the quality control capability of the production process.

Description

Technical Field

[0001] This invention relates to the field of stamping manufacturing process control technology, and in particular to a dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement. Background Technology

[0002] Current deep compensation control systems in the home appliance stamping manufacturing industry generally adopt a compensation strategy selection mechanism based on preset priorities or empirical rules, combined with real-time feedback of slider displacement to achieve closed-loop fine-tuning. Mainstream compensation systems typically configure multiple compensation strategies, including thickness compensation for material thickness fluctuations, springback compensation for die springback, and thermal deformation / impact compensation for equipment thermal deformation and vibration. These strategies often have their triggering order and activation priority set through fixed rules or empirical thresholds, such as combining outputs according to a pre-defined primary and secondary strategy weight table. Current industry technology development trends focus more on refining compensation methods, improving signal feedback resolution, and enhancing the ability to execute strategies in parallel. Some advanced systems have already achieved multi-strategy parallel compensation centered on slider displacement, significantly improving the forming accuracy and process robustness of stamped parts.

[0003] In existing methods, the priority of compensation strategies based on slider displacement feedback is mostly statically set, lacking the ability to dynamically adapt in real time to multiple factors such as disturbance dynamics within the stamping cycle, material batches, and die wear. When sudden and severe disturbances occur in the stamping environment, material batches are switched, or the die condition changes (such as local fatigue wear), the original fixed priority mechanism cannot adjust the compensation strategy combination in time, often leading to attenuation of compensation effect, decreased system stability, or even amplification of impact error. For example, the thickness compensation strategy may maintain a high weight in the latter half of the springback-dominated process, potentially causing mechanical conflicts with springback compensation and preventing the compensation command from being executed correctly. Furthermore, under special conditions such as cold material entering the die or insufficient die preheating, the frequency domain disturbance structure is prominent, but the system struggles to dynamically identify the dominant disturbance and make priority adjustments.

[0004] Most existing representative technologies rely on manually set rule logic, using empirical parameters / process thresholds to trigger strategies and allocate weights, or manually adjusting offline calibrated expert knowledge bases to adapt to different process adjustment scenarios. These methods have certain applicability under standardized, mass production conditions, but their adaptability to complex and changing environments is limited: they cannot effectively determine the real-time transformation of the dominant dimension of disturbance energy during the stamping cycle, nor can they dynamically adjust the strategy response weights according to the disturbance intensity, which easily leads to problems such as overlapping compensation strategies and control command out-of-bounds issues, effectively restricting the overall responsiveness of the system and stable operation under process limits.

[0005] Furthermore, existing compensation control modules generally suffer from the following technical deficiencies in dynamic strategy sequencing: It lacks the ability to perform time-frequency analysis on the structural characteristics of multi-source disturbance signals and cannot accurately distinguish the dominant disturbance type based on the real-time energy distribution in the slider displacement feedback.

[0006] The priority and triggering conditions of the compensation strategy are statically set, making it impossible to perceive the evolution trend of disturbances online, nor can it achieve adaptive compensation decision-making by leveraging real-time process context (such as mold number, material batch number, etc.).

[0007] When multiple strategies are output in parallel, the lack of a unified reference coordinate system for conflict detection and coordination can easily lead to overshoot, compensation command out-of-bounds errors, and even equipment safety hazards.

[0008] Under complex operating conditions, strategy switching suffers from lag and jitter, lacks smooth order reduction processing and instruction transition filtering, resulting in uncontrollable compensation actions or slow response.

[0009] In practical applications, these shortcomings cause the compensation system to lag in recognizing complex disturbances and switch strategies abruptly, affecting the stability and yield of stamped products. As stamping processes place higher demands on precision and response speed, the industry urgently needs a deep compensation control method that can perceive the dominant distribution of disturbance energy in real time, intelligently prioritize and dynamically adjust the compensation strategy based on interpretable signal characteristics.

[0010] Based on the above analysis, to improve the dynamic response and robustness of the stamping compensation system under complex and variable disturbance environments, and to promote the evolution of compensation control technology towards higher intelligence and adaptability, it is urgent to study and implement a self-organizing selection method for compensation strategies based on the dynamic disturbance spectrum of slider displacement. By analyzing the time-frequency characteristics of the slider displacement signal online, a mapping relationship between disturbance and strategy is dynamically constructed. Without manual intervention or fixed priority rules, the system can autonomously activate the most suitable combination of compensation strategies and adjust their weights based on the current disturbance energy distribution, evolution trend, and process stage status. This effectively eliminates conflicts and boundary overflows during parallel compensation of multiple strategies, ensuring the accuracy of the tightening depth and equipment safety, and meeting the actual needs of high-quality stamping manufacturing. Summary of the Invention

[0011] This application provides a dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement, which aims to solve one of the problems or issues of the prior art mentioned in the background.

[0012] The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement provided in this application specifically includes: S1: Obtain the high sampling rate slider displacement deviation sequence within the stamping cycle, and perform short-time Fourier transform and continuous wavelet transform on the slider displacement deviation sequence to generate a two-dimensional perturbation spectrum characterizing the energy of multi-source perturbation. S2: Based on the time-frequency energy distribution characteristics and process stage semantic information of the two-dimensional perturbation spectrum, construct a perturbation spectrum semantic tag system that includes typical forms such as cold material entering the mold, insufficient mold preheating, and springback dominance. S3: Utilize the semantic tagging system of the disturbance spectrum to perform interpretable template matching on the current two-dimensional disturbance spectrum, and combine the context metadata of the stamping part model and mold number to generate a multi-dimensional disturbance confidence vector reflecting the confidence of each disturbance source; S4: Based on the energy distribution and evolution trend of the multidimensional perturbation confidence vector, and following the three-layer logic of perturbation-dominant dimension, strategy-sensitive dimension, and process stage constraints, dynamically generate a strategy spectrum response matrix that defines the strategy activation weight; S5: Based on the strategy spectrum response matrix, automatically activate the most suitable compensation strategy combination, and calculate the initial displacement correction amount based on the compensation strategy combination; S6: Project the initial displacement correction amount onto a unified reference coordinate system to synthesize a compensation command vector, and determine whether the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system. Based on the determination result, realize dynamic compensation for the stamping depth. S7: If it is determined that the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system, the order reduction coordination process is started. By freezing the principal components of the low confidence strategy and superimposing a smooth transition filter, the final compensation control command that satisfies the geometric compatibility constraint is generated. S8: Based on the final compensation control command, drive the press slide to perform displacement compensation action, and feed back the actual slide position after execution to the closed-loop control system to update the slide displacement deviation sequence for the next cycle.

[0013] The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement provided in this application has the following beneficial effects: (1) To address the problem that the selection of compensation control strategies in the traditional manufacturing of stamped parts for home appliances relies on fixed thresholds or manual experience and lacks a dynamic priority ranking mechanism, this solution mines the hidden disturbance structure information from the closed-loop feedback signal of the slider displacement, and realizes adaptive strategy scheduling driven by the actual process state. Compared with the existing technology that generally adopts static rules to trigger compensation actions, this solution first performs time-frequency joint analysis on the high sampling rate deviation sequence to construct a two-dimensional disturbance spectrum with clear physical meaning, and couples the travel stage in the time dimension with the cause of disturbance in the frequency dimension, so that the multiple disturbances originally superimposed in the time domain signal can be decoupled in layers; further, it establishes a semantic labeling system for the spectrum by combining process knowledge, so that the typical disturbance mode has an interpretable process attribution ability, thereby providing a reliable basis for subsequent strategy decision-making. This design significantly improves the system's accuracy in identifying complex disturbance conditions and its response sensitivity. It effectively overcomes the technical defects of traditional methods in the face of complex disturbances, such as mis-activation, missed activation, or multiple activation of strategies. In particular, it demonstrates stronger robustness and adaptability in typical unsteady-state scenarios such as cold material entering the mold, insufficient mold preheating, springback dominance, and clamping force attenuation.

[0014] (2) To achieve efficient and feasible online decision-making, this solution abandons the computationally intensive path of relying on deep learning model training and offline optimization, and innovatively designs a lightweight spectrum matching engine and dynamic response matrix mechanism to ensure that the selection process of compensation strategy is both real-time and logically transparent. By using template matching combined with contextual metadata (such as stamping part model, mold number, and equipment runtime) to generate a multi-dimensional disturbance confidence vector, the unexplainable risks brought by the black box model are avoided. Based on this, the strategy spectrum response matrix is ​​constructed, and priority order is dynamically generated according to the three-layer logic of "disturbance dominant dimension → strategy sensitive dimension → process stage constraint", which realizes the synergistic suppression and conditional activation between compensation strategies. For example, springback compensation is enhanced in the late forming stage, while dual-strategy synergy is implemented in the holding stage, but is limited by geometric compatibility, which ensures the process rationality of control commands. Simultaneously, a strategy execution consistency verification mechanism is introduced, uniformly projecting the output corrections of each strategy onto the kinematic space and detecting whether the synthesized vector exceeds the mechanical limit or the dynamic boundary of the servo system. If it exceeds the boundary, a reduction-order coordination process is initiated, freezing low-confidence strategies and retaining the principal component output, and superimposing a smoothing filter to ensure command continuity and safety. The entire closed-loop control process does not require offline modeling or parameter tuning, and feature extraction, matching judgment, and command generation can be completed within a single stamping cycle, exhibiting excellent real-time response performance and plug-and-play capability for production lines.

[0015] The aforementioned technical methods collectively construct a closed-loop control system for dynamic disturbance identification and autonomous ranking of compensation strategies. This system not only significantly improves the accuracy and stability of stamping process control but also solves the deployment challenges of existing technologies caused by strong model dependence, complex debugging, and poor generalization ability through interpretable semantic spectrogram analysis and lightweight logical reasoning mechanisms. Without introducing predictive models, reinforcement learning, or multi-objective optimization, the system achieves rapid identification and differentiated responses to various typical disturbance conditions, balancing optimal control performance with safety during execution. It is particularly suitable for intelligent manufacturing scenarios such as home appliance stamping, which involve frequent product model switching and drastic fluctuations in operating conditions, providing a practical technical path for achieving high-quality, highly flexible, and low-intervention automated production. Attached Figure Description

[0016] Figure 1 This is the main flowchart of the dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement.

[0017] Figure 2 This is a sub-flowchart of the dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement.

[0018] Figure 3 This is another sub-flowchart of the dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0021] like Figure 1 As shown, this application provides a dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement, specifically including: S1: Obtain the high sampling rate slider displacement deviation sequence within the stamping cycle, and perform short-time Fourier transform and continuous wavelet transform on the slider displacement deviation sequence to generate a two-dimensional perturbation spectrum characterizing the energy of multi-source perturbation. S2: Based on the time-frequency energy distribution characteristics and process stage semantic information of the two-dimensional perturbation spectrum, construct a perturbation spectrum semantic tag system that includes typical forms such as cold material entering the mold, insufficient mold preheating, and springback dominance. S3: Utilize the semantic tagging system of the disturbance spectrum to perform interpretable template matching on the current two-dimensional disturbance spectrum, and combine the context metadata of the stamping part model and mold number to generate a multi-dimensional disturbance confidence vector reflecting the confidence of each disturbance source; S4: Based on the energy distribution and evolution trend of the multidimensional perturbation confidence vector, and following the three-layer logic of perturbation-dominant dimension, strategy-sensitive dimension, and process stage constraints, dynamically generate a strategy spectrum response matrix that defines the strategy activation weight; S5: Based on the strategy spectrum response matrix, automatically activate the most suitable compensation strategy combination, and calculate the initial displacement correction amount based on the compensation strategy combination; S6: Project the initial displacement correction amount onto a unified reference coordinate system to synthesize a compensation command vector, and determine whether the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system. Based on the determination result, realize dynamic compensation for the stamping depth. S7: If it is determined that the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system, the order reduction coordination process is started. By freezing the principal components of the low confidence strategy and superimposing a smooth transition filter, the final compensation control command that satisfies the geometric compatibility constraint is generated. S8: Based on the final compensation control command, drive the press slide to perform displacement compensation action, and feed back the actual slide position after execution to the closed-loop control system to update the slide displacement deviation sequence for the next cycle.

[0022] Step S1: Obtain the high-sampling-rate slider displacement deviation sequence within the stamping cycle, and perform joint processing of the slider displacement deviation sequence with short-time Fourier transform and continuous wavelet transform to generate a two-dimensional perturbation spectrum characterizing the energy of multi-source perturbations. Specifically, this includes: S1.1: Obtain the original sampling data of the high-frequency displacement sensor of the press slide within a single stamping cycle, and perform real-time differential calculation between the original sampling data of the high-frequency displacement sensor and the preset ideal motion trajectory curve to generate a slide displacement deviation sequence that reflects the actual execution error.

[0023] When acquiring the raw sampling data of the high-frequency displacement sensor of the press slide within a single stamping cycle, a high-resolution optical grating ruler or laser displacement sensor mounted on the slide guide column is used. At key process nodes of the press, the data acquisition module is triggered to achieve a continuous time-series data stream with a sampling frequency higher than 10kHz. The raw sampling data of the high-frequency displacement sensor is input to a real-time processing unit based on a buffer queue to ensure no data loss and that the timestamp accuracy is not lower than the microsecond level, for phase alignment in subsequent differential operations. A preset ideal motion trajectory curve is invoked. This curve is provided by the press manufacturer or generated based on equipment calibration tests, using a unified reference coordinate system to characterize the target displacement value throughout the stroke. The curve resolution is consistent with the sensor sampling frequency.

[0024] For example, in the production of stamped parts for a certain type of household appliance, a laser displacement sensor with a sampling frequency of 12500Hz and a resolution of 1μm is used to acquire the original displacement data of the slider within one stamping cycle. The ideal motion trajectory curve is a stroke displacement target curve generated based on equipment calibration tests, with 12500 sampling points. A point-by-point difference calculation is performed between the measured value and the ideal value at each sampling point. According to the above formula, if the measured displacement at a certain time point is 40.000mm and the ideal displacement is 39.995mm, the difference result is 0.005mm, indicating that the slider position deviation at that time point is 5μm. This deviation value is written into the slider displacement deviation sequence cache, with the sequence length consistent with the number of sampling points in the cycle. After calculating the deviation for the entire punch sequence, the maximum deviation value in the sequence reached 0.015mm, and the minimum deviation value was -0.004mm. This data is the direct input for subsequent Fourier and wavelet joint analysis. In actual production line verification, the deviation sequence can accurately capture the small displacement dynamic changes caused by mold springback and material thickness fluctuations, significantly improving the construction accuracy of the perturbation spectrum and the reliability of strategy selection.

[0025] S1.2: Apply a Hanning window function to the slider displacement deviation sequence for segmentation and truncation, and perform a fast Fourier transform on the windowed data segment to extract a set of discrete spectral components that reflect the periodic disturbance characteristics.

[0026] Set the processing window parameters for the slider displacement deviation sequence output from the preceding step S1.1, and pass the sequence as an input object to the window function processing module to establish a segmented truncated time-domain data structure.

[0027] Based on the sampling length of the stamping cycle, the Hanning window function is selected and the corresponding sampling point coefficient matrix is ​​loaded. The data segments in the slider displacement deviation sequence are multiplied and weighted to form windowed data segments that have been suppressed by the Hanning window to suppress spectral leakage.

[0028] The windowed data segments are fed into the Fast Fourier Transform (FFT) computation unit, and a recursive divide-and-conquer spectral decomposition based on the segment length is performed to obtain the complex spectral representation matrix of each segment.

[0029] The real and imaginary parts of the complex spectrum representation matrix are subjected to modulus squaring and normalization to extract the corresponding amplitude spectrum components and construct a set of discrete spectrum components that reflect the characteristics of periodic perturbations.

[0030] The set of discrete spectral components is stored in the frequency domain buffer module as the reference frequency domain data for subsequent continuous wavelet transforms, thus achieving a closed-loop connection between the data flow and the next sub-step.

[0031] By using Hanning window weighting and FFT spectrum extraction, the slider displacement deviation sequence generated in the previous step is transformed into a set of discrete spectral components with periodic disturbance characteristics, thereby achieving accurate separation and energy quantization of the main frequency disturbance components in the stamping process.

[0032] For example, in the stamping process of household appliance stamping parts, the sampling frequency of the slider displacement deviation sequence is set to 20000Hz, and the total sampling length for a single stamping cycle is 40000 points. The Hanning window function window length is set to 1024 points, and the window coefficient is... According to the formula

[0033] In this diagram, n represents the sample point index, and N represents the window length. The data segments of the slider displacement deviation sequence are weighted by multiplying the window coefficients to obtain windowed data segments that eliminate the abrupt boundary effects. A 1024-point FFT is then performed to output a complex spectrum matrix. After modulus squaring, the amplitude spectrum is obtained. A significant peak in the low-frequency band corresponds to material thickness fluctuations, while a stable, smaller peak in the mid-frequency band corresponds to mold elastic deformation disturbances. This set of discrete spectral components is used as the frequency domain reference input in subsequent step S1.3. In the application effect verification, the relative amplitude stability of the low-frequency and mid-frequency peaks is significantly improved, effectively supporting the generation of dynamic priority strategies.

[0034] S1.3: Using the set of discrete spectral components as the frequency domain reference, apply the Morlet wavelet basis function to the slider displacement deviation sequence to perform multi-scale continuous wavelet transform processing to generate a wavelet coefficient matrix that characterizes transient impacts and non-stationary signal details.

[0035] S1.4: Based on the set of discrete spectral components and the modulus squared of the wavelet coefficient matrix, calculate the frequency band energy density for each time slice, and map the frequency band energy density to grayscale intensity values ​​to construct a preliminary time-frequency energy distribution heatmap.

[0036] The set of discrete spectral components obtained by short-time Fourier transform and the wavelet coefficient matrix generated by continuous wavelet transform are respectively subjected to modulus square values ​​to form two datasets representing energy in the frequency domain and time domain. The modulus square values ​​of the discrete spectral component set are divided into frequency bands according to time slice indices, and the energy values ​​of each frequency component within each time slice are obtained as frequency domain energy parameters. The modulus square values ​​of the wavelet coefficient matrix are mapped back to the corresponding frequency bands according to scale, and the energy values ​​of transient impacts and non-stationary fluctuations are extracted as wavelet domain energy parameters. A weighted average operation is performed on the frequency domain energy parameters and the wavelet domain energy parameters under the same time slice. The weighting coefficients are preset according to the physical characteristics of the interference source and the sensitivity of the detection target to achieve time-frequency energy fusion. The energy density of a given frequency band under each time slice is calculated using the following formula:

[0037] in, This represents the amplitude value of the corresponding frequency component or wavelet coefficient. This represents the number of sampling points within the frequency band. Gray-scale mapping is performed on the calculated energy density values, assigning deeper gray intensities to higher energy values ​​and lighter gray intensities to lower energy values, ensuring the time-frequency structure is visually represented in a two-dimensional coordinate system. The mapped gray values ​​are then reorganized into a matrix according to time slices and frequency axis positions, generating a preliminary time-frequency energy distribution heatmap. This processing method transforms the frequency and time domain analysis results from the previous step into structural data where the distribution of multi-source disturbances can be directly observed on a two-dimensional plane, enabling the time-frequency information required for subsequent feature extraction and semantic label construction.

[0038] For example, in the stamping process of manufacturing stamped parts for home appliances, a high-sampling-rate displacement sensor acquires 1024 sampling points within one cycle. FFT analysis yields discrete spectral component sets for low-frequency, mid-frequency, and high-frequency bands. The amplitude range for the low-frequency band is 0.8–1.2 mm, for the mid-frequency band it is 0.3–0.5 mm, and for the high-frequency band it is 0.05–0.1 mm. Morlet wavelet transform is used to extract coefficient matrices at five scales, with the second scale corresponding to the mid-frequency band and the fourth scale corresponding to the high-frequency band. Modulus squared values ​​are calculated for each component. After categorizing by frequency band, the number of sampling points in each time slice is set to 256. These are then used to calculate the energy density. For instance, the energy density for the low-frequency band reaches 0.95 mm² in the first half of the stamping process, the mid-frequency band remains at 0.45 mm², and the high-frequency band suddenly increases to 0.09 mm² at the end of the forming process. The energy density is mapped to gray values ​​from 0 to 255. The high-density area in the low-frequency band corresponds to a gray value of 210, the mid-frequency band corresponds to a gray value of 150, and the high-frequency band surge area corresponds to a gray value of 240. The generated preliminary heat map can clearly show the interference pattern of "low-frequency surge in the first half + continuous rise in the mid-frequency band + high-frequency spikes in the last half", which provides accurate time-frequency feature input for subsequent perturbation semantic label matching.

[0039] S1.5: Perform bilinear interpolation smoothing and dynamic range compression on the preliminary time-frequency energy distribution thermogram to eliminate high-frequency noise interference from the sensor and enhance the recognizability of low-frequency material thickness fluctuation characteristics, ultimately generating a standardized two-dimensional perturbation spectrum.

[0040] Step S2: Based on the time-frequency energy distribution characteristics and process stage semantic information of the two-dimensional perturbation spectrum, a perturbation spectrum semantic tagging system is constructed, including typical forms such as cold material entering the mold, insufficient mold preheating, and springback dominance. Specifically, this includes: S2.1: The time-frequency energy distribution matrix of the two-dimensional perturbation spectrum is divided into frequency bands. Based on the physical causes of disturbances in the field of home appliance stamping parts manufacturing, the frequency axis is divided into a low-frequency band that represents the overall thickness shift of the material, a mid-frequency band that represents the elastic deformation of the local mold, and a high-frequency band that represents transient impact, so as to generate a set of frequency band energy sub-graphs with clear physical meaning.

[0041] A frequency axis scan operation is performed on the standardized two-dimensional perturbation spectrum energy distribution matrix to read the frequency value and energy amplitude corresponding to each column in the matrix. The frequency value is then matched with the perturbation physical cause mapping rules summarized in the field of home appliance stamping parts manufacturing to identify the perturbation cause category to which each frequency value belongs.

[0042] The classification boundary determination operation is performed on the identified disturbance cause categories. According to the physical characteristics of low-frequency band, mid-frequency band local mold elastic deformation, and high-frequency band transient impact of the overall material thickness offset, the frequency axis is divided into three non-overlapping intervals that cover the entire frequency domain. Each interval is assigned a physical meaning label and an energy characteristic index number.

[0043] After the frequency axis is divided, the energy matrix of the two-dimensional perturbation spectrum is subjected to interval mask extraction processing. The frequency columns corresponding to the low frequency band, mid frequency band and high frequency band are extracted and saved as independent energy matrix subsets in the order of index number, forming a preliminary data framework of frequency band energy sub-map set.

[0044] Based on the evolution relationship of the energy matrix subsets on the time axis, time slice recombination processing is performed on the energy sub-maps of each frequency band. The energy values ​​of different time slices are sorted and stored according to the interval segment number to ensure that the timing information of each process stage can be accurately aligned when extracting morphological features later.

[0045] Through the above-mentioned frequency axis division, energy mask extraction and time slice recombination, the standardized two-dimensional perturbation spectrum is transformed into a set of low-frequency, mid-frequency and high-frequency energy sub-graphs with clear physical meaning and independent analysis, thereby realizing the transformation of the time-frequency energy distribution matrix into a segmented data structure based on physical causes.

[0046] S2.2: Based on the time evolution trajectory of the frequency band energy sub-graph set, extract key morphological feature parameters for different stages of the stamping stroke, including the slope of the low-frequency energy surge in the first half, the amplitude of the continuous rise in the mid-frequency, and the density of high-frequency burrs in the second half, so as to generate a multi-dimensional morphological feature vector describing the spatiotemporal distribution characteristics of the disturbance.

[0047] S2.3: Using the multi-dimensional morphological feature vector and the preset process knowledge base, a mapping and association operation is performed to define specific feature combination patterns as standard perturbation semantic prototypes. Specifically, this includes binding the combination of low-frequency energy surge and mid-frequency continuous rise in the first half of the process as a cold material entering the mold composite perturbation prototype, and binding the combination of high-frequency burr density and low-frequency baseline drift in the second half of the process as a springback-dominated coupling state prototype, so as to generate a perturbation semantic prototype library containing a variety of typical perturbation types.

[0048] Using the multi-dimensional morphological feature vector output from step S2.2 as input, the pre-defined disturbance pattern definitions and morphological combination rule tables in the process knowledge base are invoked. Pattern matching operations are performed on each component of the morphological feature vector, mapping the frequency and time-based dual-dimensional features to disturbance category indices with process semantics. A two-dimensional combined feature group is constructed using the low-frequency surge slope and mid-frequency sustained rise amplitude of the morphological feature vector. Rule retrieval is performed on this group to determine its coupling relationship with the cold material entry mold composite disturbance prototype. Another combined feature group is constructed by combining the high-frequency burr density and low-frequency baseline drift amplitude in the latter half of the morphological feature vector. Rule retrieval is performed on this group to determine its mapping to the springback-dominated coupling state prototype. For other disturbance patterns, the morphological feature components of the corresponding frequency band and time period are extracted sequentially, and matrix matching is performed to bind the successfully matched feature combinations to the standard disturbance semantic prototype entries defined in the knowledge base. All matching results are indexed and organized according to perturbation type number and morphological group feature uniqueness to generate a perturbation semantic prototype library containing cold material entering the mold, insufficient mold preheating, springback dominance, and other typical perturbation types. Through pattern matching and binding processing, the morphological feature vector results from the previous step are transformed into perturbation semantic prototype data with process semantics that can be used for tag template construction, achieving the technical effect of standardized perturbation morphological definition.

[0049] For example, in a home appliance stamping parts manufacturing line, the low-frequency surge slope was measured to be 2.5 units amplitude / ms, the mid-frequency sustained rise amplitude was measured to be 1.8 units amplitude, the high-frequency burr density in the latter half was 35 peaks per millisecond, and the low-frequency baseline drift amplitude was 0.6 units amplitude. The low-frequency surge slope and the mid-frequency sustained rise amplitude are combined to form a feature set input to the rule matching function.

[0050] Wherein, CF is the combined feature value, which is confirmed to fall within the composite disturbance threshold range of cold material entering the mold through knowledge base retrieval. The high-frequency burr density and the low-frequency baseline drift amplitude are used to construct the combined feature group input rule matching function:

[0051] Here, CB is a combined feature value, which is confirmed to fall within the threshold range of the springback-dominated coupling state through knowledge base retrieval. In this embodiment, the semantic prototype library is updated to include two types of perturbation prototypes: cold material entry and springback-dominated. After template construction, it can be directly used for subsequent online matching calculation of two-dimensional perturbation spectra, significantly improving the accuracy and robustness of multi-strategy compensation control.

[0052] S2.4: Based on the standard perturbation semantic prototypes in the perturbation semantic prototype library, construct corresponding two-dimensional perturbation spectrum semantic label templates, assign a unique semantic identifier to each template and specify its geometric compatibility constraints, so as to generate a perturbation spectrum semantic label system for subsequent online matching calculations.

[0053] Based on the standard perturbation semantic prototypes in the perturbation semantic prototype library generated in the previous steps, a template transformation process is performed on them. The multidimensional morphological feature vectors are mapped and geometrically normalized in the two-dimensional perturbation spectrum coordinate space according to the time-frequency axis, thereby forming a semantic label template with a fixed reference scale and energy density benchmark.

[0054] The mapping results of each standard perturbation semantic prototype are subjected to grid resolution adjustment and color coding processing. By using unique grayscale or pseudo-color coding for different frequency band energy densities and marking the stamping stroke stage with the time axis, the template has a clear physical meaning and process stage identification.

[0055] Using the geometric compatibility constraints required in the template matching method, an acceptable set of compensation strategy direction vectors and magnitude limits are defined for each semantic label template. The vector envelope curve is described in a unified kinematic reference frame, and the constraint interval parameters are stored in mathematical form.

[0056] Each semantic tag template is assigned a unique semantic identifier, which consists of three parts: perturbation type encoding, process stage encoding, and frequency band combination encoding, to ensure that the corresponding template instance can be quickly located and matched in online matching operations.

[0057] The templates, identifiers, and geometric compatibility constraint parameters mentioned above are stored in the perturbation spectrum semantic label system data structure to form a standardized label library that can be directly called by step S3.

[0058] Through template processing, identifier assignment, and constraint definition, the standard perturbation semantic prototype of the previous step is transformed into a two-dimensional perturbation spectrum semantic label system that is identifiable, matchable, and verifiable, thereby providing structured basic data support for the process of matching perturbation modes with compensation strategies.

[0059] like Figure 2 As shown, step S3 involves: using the perturbation spectrum semantic tagging system to perform interpretable template matching on the current two-dimensional perturbation spectrum, and combining the stamping part model and mold number context metadata to generate a multi-dimensional perturbation confidence vector reflecting the confidence of each perturbation source. Specifically, this includes: S3.1: Obtain the two-dimensional perturbation spectrum of the current stamping cycle and the pre-stored perturbation spectrum semantic tag system. Perform normalization alignment processing on the two-dimensional perturbation spectrum and the typical morphology template in the perturbation spectrum semantic tag system to eliminate the phase shift caused by the difference in sampling time reference and generate aligned perturbation spectrum data with a unified spatiotemporal reference.

[0060] The two-dimensional perturbation spectrum file of the current stamping cycle is obtained as the input object for this step, and the time-frequency matrix data of each typical morphology template in the pre-stored perturbation spectrum semantic label system are called to establish a matching reference benchmark.

[0061] The time axis resolution of the current two-dimensional perturbation spectrum is adjusted to match the sampling period of the semantic label template. The time axis difference is corrected by linear interpolation to ensure that each time slice corresponds to the same stamping stroke stage in the matching operation.

[0062] Amplitude normalization is performed on the two-dimensional perturbation spectrum after time axis correction. The maximum energy value of each frequency band is used as the normalization coefficient to map the energy amplitude of different frequency bands to a unified [0,1] standard domain, thereby eliminating the influence of sensor gain differences.

[0063] The normalized two-dimensional perturbation spectrum and the semantic tag template are aligned with the phase reference one by one. The optimal phase offset is calculated by the cross-correlation function and the offset is compensated by shift operation so that the position of the high energy peak on the time axis is consistent with the template.

[0064] The aligned spectral data is recorded as a matching input matrix with a unified spatiotemporal reference, providing consistent data conditions for subsequent structural similarity calculations.

[0065] Through the above normalization and phase alignment processing, the two-dimensional perturbation spectrum from the previous step is transformed into aligned perturbation spectrum data with a unified matching benchmark, thereby improving the comparability and accuracy of template matching under different sampling benchmarks.

[0066] S3.2: Based on the aligned perturbation spectrum data, the weighted Euclidean distance method is used to calculate the structural similarity value between it and the cold material entry template, insufficient mold preheating template and springback dominant template in the semantic labeling system of the perturbation spectrum, so as to quantify the matching degree between the current spectrum shape and each known perturbation mode, and generate an initial matching result vector containing multiple basic similarity components.

[0067] S3.3: Obtain the stamping part model identifier and mold number identifier associated with the current production task as context metadata. Based on the context metadata, retrieve the corresponding material thickness fluctuation coefficient and mold wear correction factor from the historical process knowledge base to perform working condition adaptive weighted correction on the basic similarity component in the initial matching result vector, and generate a working condition calibrated weighted matching score vector.

[0068] Based on the initial matching result vector obtained as input, the stamping part model identifier and mold number identifier associated with the current production task are read, and the identifier data is structured and stored as a context metadata set. The historical process knowledge base interface is accessed through the context metadata set to perform a material thickness fluctuation coefficient retrieval operation based on the model identifier, and the retrieval result is cached as a first working condition correction factor matrix. For the mold number identifier, a mold wear correction factor retrieval operation is performed to obtain a second working condition correction factor matrix representing the mold's usage stage and wear level. Element-level fusion operations are performed on the first and second working condition correction factor matrices to combine them into a single comprehensive working condition correction coefficient vector. The comprehensive working condition correction coefficient vector is used to perform adaptive weighting operations on each basic similarity component in the initial matching result vector, specifically using element-wise multiplication to dynamically amplify or suppress the matching degree of the working condition state and disturbance mode. For the adaptive weighting operation, the weighted matching score is calculated using the following formula:

[0069] Where S is the weighted matching score vector, R is the initial matching result vector, C is the comprehensive working condition correction coefficient vector, and \times indicates element-wise multiplication. The calculated S is written into the data processing module as the final weighted matching score vector after working condition calibration. Through the above processing method, the initial matching result vector of the previous step is transformed into a weighted matching score that considers material thickness fluctuations and mold wear factors, realizing adaptive adjustment of the disturbance mode matching degree under different working conditions.

[0070] For example, in a home appliance stamping parts manufacturing line, the stamping part model identifier for a certain batch of tasks is A12B, and the mold number identifier is M87C. In the historical process knowledge base, the material thickness fluctuation coefficient corresponding to this model is 0.92, and the wear correction factor corresponding to this mold number is 1.05. The initial matching result vector R contains a similarity of 0.84 for cold material entering the mold, 0.65 for insufficient mold preheating, and 0.78 for springback dominance. The material thickness fluctuation coefficient and the wear correction factor are respectively constructed into a first working condition correction factor matrix [0.92, 0.92, 0.92] and a second working condition correction factor matrix [1.05, 1.05, 1.05]. Element-wise multiplication is performed to obtain the comprehensive working condition correction coefficient vector C=[0.966, 0.966, 0.966]. Element-wise multiplication is performed on R and C to obtain the weighted matching score vector S=[0.811, 0.628, 0.753]. This vector is used as input for the normalization of the perturbation probability distribution in subsequent steps, which ultimately realizes the adjustment of the matching score after considering material fluctuations and mold wear. This significantly improves the score value of cold material entering the mold perturbation, keeps the score value of springback-dominated perturbation at a high level, and slightly reduces the score value of insufficient mold preheating perturbation. This improves the accuracy and robustness of the system strategy selection in the dynamic strategy weight allocation stage.

[0071] S3.4: Perform maximum value normalization processing on the weighted matching score vector after working condition calibration, and map the weighted matching score corresponding to each disturbance mode to the probability interval of zero to one, so as to characterize the relative probability of each disturbance source in the current stamping cycle and generate a preliminary disturbance probability distribution sequence.

[0072] S3.5: Based on the preliminary disturbance probability distribution sequence, vector recombination and confidence assignment are performed according to the disturbance source classification dimensions such as material thickness fluctuation, mold elastic deformation and transient impact. The probability values ​​are converted into confidence values ​​that can be directly used for strategy weight calculation, and the final multidimensional disturbance confidence vector is generated.

[0073] A multidimensional probability value vector composed of a preliminary disturbance probability distribution sequence is obtained, and the physical category label of the disturbance source corresponding to each component in the vector is determined. Based on the three categories of disturbance sources—material thickness fluctuation, mold elastic deformation, and transient impact—the probability value vector is grouped and rearranged according to the labels to form a recombined probability matrix with a fixed classification order. A nonlinear confidence mapping operation is performed on each disturbance source probability component in the recombined probability matrix. A transformation model based on the Sigmoid function is used, and the mapping curve parameters are adjusted according to the sensitivity differences of the disturbance source to control accuracy in the stamping part manufacturing scenario, to achieve smooth compression or expansion of probability values ​​to confidence values.

[0074] By using classification and recombination and nonlinear mapping, the probability distribution results from the previous step are transformed into confidence values ​​that can be used to calculate dynamic strategy weights, thus achieving a quantitative input of the degree of influence of multiple disturbance sources.

[0075] like Figure 3 As shown, step S4 involves dynamically generating a policy spectrum response matrix that defines policy activation weights based on the energy distribution and evolution trend of the multidimensional perturbation confidence vector, following a three-layer logic of perturbation-dominant dimension, policy-sensitive dimension, and process stage constraints. Specifically, this includes: S4.1: Normalize the confidence values ​​of each disturbance source contained in the multidimensional disturbance confidence vector and calculate the energy ratio to extract the disturbance-dominant dimension identifier that represents the most important disturbance source in the current stamping cycle, and use the disturbance-dominant dimension identifier as the input basis for determining the basic priority of the compensation strategy.

[0076] The input data of the multidimensional disturbance confidence vector is centrally processed. The multidimensional disturbance confidence vector contains the confidence values ​​of each disturbance source in the current stamping cycle, which are derived from the preceding matching calculation and working condition correction steps.

[0077] The values ​​of each perturbation source in the multidimensional perturbation confidence vector are normalized by using a linear normalization method to scale the values ​​proportionally by the difference between the maximum and minimum values ​​of all confidence components, so as to ensure that the numerical ranges of different perturbation sources are consistent and comparable.

[0078] The energy percentage is calculated for the normalized confidence component set, and the ratio of each component value to the sum of the normalized sets is used as the energy percentage of the disturbance source.

[0079] The disturbance-dominant dimension identifier is bound to a fixed mapping rule to generate an input basis for determining the basic priority of the compensation strategy, ensuring that subsequent strategy sensitivity matching can directly call the identifier and adjust the priority according to the dominant disturbance type.

[0080] By normalizing and calculating the energy percentage, the results of the preceding steps are transformed into dynamic, comparable, and sortable main disturbance identifiers, thereby enabling the quantification and automatic determination of the basic priority of the compensation strategy.

[0081] S4.2: Based on the disturbance dominance dimension identifier, query the preset strategy sensitivity mapping library, perform strategy sensitivity matching operation to filter out the set of candidate compensation strategies that are most sensitive to the current dominant disturbance type, and generate a list of strategy sensitivity dimension parameters containing the sensitivity coefficients of each candidate compensation strategy.

[0082] An index matching relationship is established between the disturbance-dominant dimension identifier and a pre-set strategy sensitivity mapping library to determine the query key value for mapping retrieval. Key-value matching is performed on each compensation strategy sensitivity vector stored in the mapping library to extract the strategy sensitivity record corresponding to the current disturbance-dominant dimension identifier. The filtered strategy sensitivity records are sorted according to the peak amplitude of each strategy in the response curve of the dominant disturbance type, generating a candidate compensation strategy set with the sensitivity peak value as the reference order. For each strategy entry in the candidate compensation strategy set, a sensitivity coefficient is calculated based on the strategy response delay time and maximum correction amplitude under the same disturbance-dominant dimension conditions in historical stamping cycles. Through the above processing method, the disturbance-dominant dimension identifier from the previous step is transformed into a quantifiable list of strategy sensitivity coefficients, achieving precision and quantification in compensation strategy selection.

[0083] S4.3: Use the timestamp information of the current stamping stroke to retrieve the semantic database of the process stage, perform process state boundary verification to determine the specific process stage constraints, and perform a logical AND operation between the process stage constraints and the strategy sensitive dimension parameter list to generate a preliminary strategy activation weight vector that has been filtered by process feasibility.

[0084] The input condition is the list of strategy sensitivity dimension parameters output by the previous step S4.2. This list contains the sensitivity coefficient of each candidate compensation strategy and its corresponding disturbance dominance dimension identifier. At the same time, the timestamp information of the stamping stroke is obtained as the basis for process stage positioning.

[0085] Input the timestamp information into the semantic database retrieval module of the process stage, and extract the semantic label and associated constraints of the specific process stage to which the current stamping stroke belongs by matching the timestamp with the process stage boundary index table.

[0086] By utilizing the stage execution boundary parameters in the semantic tags of the process stages, stage feasibility verification is performed on the list of strategy sensitive dimension parameters, and strategy entries with high execution risk or unmet physical conditions in the current stage are removed.

[0087] The list of strategy parameters that have passed the stage feasibility verification will be logically ANDed with the process stage constraints. The operation rule is: if the strategy is allowed under the stage constraints and the sensitivity coefficient is higher than the preset threshold, it will be retained and given a stage applicable flag bit; otherwise, the flag bit will be set to zero.

[0088] The initial strategy activation weight vector, filtered by process feasibility, is output using matrix-based logical operations. The vector component values ​​are calculated by multiplying the strategy sensitivity coefficient by the applicable stage flag. The product calculation formula is as follows:

[0089] in, These are the initial weighted component values. This represents the strategy sensitivity coefficient. This is a flag used for the specific stage.

[0090] Through the above logical verification and matrix operation processing, the list of strategy-sensitive dimension parameters from the previous step is transformed into a preliminary strategy activation weight vector with stage applicability, thereby realizing the process stage feasibility constraint mapping of strategy weights.

[0091] For example, at a stamping stroke timestamp of 1.25 seconds, the process stage semantic database is retrieved, indicating the current stage is the final forming stage. The stage constraint prohibits the thickness compensation strategy amplitude from exceeding 0.15 mm. The strategy sensitivity dimension parameter list includes a springback compensation strategy sensitivity coefficient of 0.82 and a thickness compensation strategy sensitivity coefficient of 0.76. During stage feasibility verification, the thickness compensation strategy's stage applicability flag is set to 0.9 because its sensitivity coefficient is higher than the threshold but its amplitude is limited, while the springback compensation strategy's stage applicability flag is set to 1.0. After a logical AND operation, the calculated weight component value of the springback compensation strategy is 0.82; the calculated weight component value of the thickness compensation strategy is 0.684. This initial strategy activation weight vector is [0.82, 0.684], which significantly improves the stage adaptability of the weight distribution in the subsequent S4.4 dynamic weighted fusion method and enhances the stability of parallel execution of multiple strategies.

[0092] S4.4: Based on the numerical magnitude and evolution slope of each component in the activation weight vector of the preliminary strategy, execute the dynamic weighted fusion method to comprehensively consider the disturbance energy intensity and the allowable range of the process stage, calculate the final strategy spectrum response matrix that defines the specific effect intensity of each compensation strategy, and complete the dynamic mapping from disturbance characteristics to control weights.

[0093] Using the initial strategy activation weight vector filtered by process feasibility as the input object, the numerical magnitude and evolution slope parameters of each weight component are extracted, and the evolution slope is defined as the rate of weight change of the same strategy in adjacent time series samples.

[0094] The numerical values ​​are weighted and normalized to form a dimensionless scaling factor matrix, and the evolution slope is normalized to make the numerical gradients between different strategies comparable.

[0095] A product factor matrix is ​​established between the disturbance energy intensity coefficient and the allowable range coefficient of the process stage. The disturbance energy intensity coefficient is derived from the dominant dimension component value of the multidimensional disturbance confidence vector, and the allowable range coefficient of the process stage is based on the stage restriction index in the process stage semantic database.

[0096] The fusion weight vector output by the above formula is smoothed and filtered. The length of the filtering window is determined based on the number of samples in the stamping cycle to reduce the interference of high-frequency oscillations on the stability of execution.

[0097] Map the smoothed fusion weight vector to the policy spectrum Figure 2 In the response matrix of the 3D coordinate system, the rows of the response matrix correspond to the policy identifier, the columns correspond to the time-frequency energy distribution feature index, and the matrix elements take the instantaneous values ​​of the fusion weights.

[0098] Through the above processing method, the results of the previous step are transformed into a strategy spectrum response matrix that characterizes the specific intensity of each compensation strategy, realizing the dynamic mapping from disturbance characteristics to control weights, and ensuring the optimal allocation of control signals in the stamping environment where multiple disturbances coexist.

[0099] Step S5: Based on the strategy spectrum response matrix, autonomously activate the most suitable compensation strategy combination, and calculate the initial displacement correction based on the compensation strategy combination. Specifically, this includes: S5.1: Based on the strategy spectrum response matrix generated in the previous steps, threshold judgment processing is performed on the multidimensional disturbance confidence vector to identify the springback compensation strategy identifier and material thickness fluctuation compensation strategy identifier that are active in the current stamping cycle, forming a set of compensation strategies to be executed.

[0100] Based on the strategy spectrum response matrix calculated in the previous steps, the dynamic weight coefficients corresponding to each compensation strategy in the matrix are associated with the multidimensional perturbation confidence vector, and the perturbation source identifiers with confidence values ​​greater than a set threshold are extracted from all mapping relationships.

[0101] The extracted disturbance source identifiers are classified and judged. The identifiers of disturbance types belonging to the mold elastic recovery category are mapped to the springback compensation strategy identifier, and the identifiers of disturbance types belonging to the material size deviation category are mapped to the material thickness fluctuation compensation strategy identifier.

[0102] During the mapping process, a pre-defined perturbation type to policy identifier mapping table is invoked to ensure consistency and uniqueness between different perturbation sources and compensation policies.

[0103] The set of policy identifiers is deduplicated to avoid the same policy being calculated repeatedly due to multiple disturbance sources.

[0104] The deduplicated set of strategy identifiers is used as the basic data for the set of compensation strategies to be executed, and the dynamic weight coefficients of each strategy identifier in the set in the strategy spectrum response matrix are recorded for subsequent displacement correction calculation.

[0105] Through the above-mentioned judgment and mapping process, the strategy weight matrix result of the previous step is transformed into a set of combinations of springback compensation strategy identifiers and material thickness fluctuation compensation strategy identifiers that are active in the current stamping cycle, thus realizing a set of compensation strategies to be executed that can be called later.

[0106] For example, in a home appliance stamping parts manufacturing line, the slider displacement closed-loop feedback signal, after disturbance spectrum analysis, yields a multi-dimensional disturbance confidence vector of {low-frequency disturbance = 0.82, mid-frequency disturbance = 0.77, high-frequency disturbance = 0.34}. The weight coefficients for the corresponding springback compensation strategy and material thickness fluctuation compensation strategy in the strategy spectrum response matrix are 0.88 and 0.75, respectively, with a confidence threshold set to 0.70. During processing, the confidence of each disturbance is first compared with the threshold. Low-frequency and mid-frequency disturbances both exceed the threshold and are mapped to springback compensation strategy identifiers and material thickness fluctuation compensation strategy identifiers, respectively. High-frequency disturbances are not activated. In the mapping table, the springback compensation strategy identifier is defined as RB_STR, and the material thickness fluctuation compensation strategy identifier is defined as MT_STR. After deduplication, the set {RB_STR, MT_STR} is obtained. This set records the dynamic weight coefficients {RB_STR = 0.88, MT_STR = 0.75}, which are used for subsequent calculations. In application, this set is passed to subsequent model call steps to calculate the basic displacement of springback compensation and the basic displacement of material thickness compensation, respectively. Combined with weighting coefficients, the initial displacement correction is generated, which ultimately significantly improves the real-time adaptability and stability of displacement compensation in actual stamping cycles.

[0107] S5.2: Using the springback compensation strategy identifier in the set of compensation strategies to be executed, call the preset springback compensation model and input the semantic information of the current process stage and the low-frequency component of the slider displacement deviation sequence to calculate the springback compensation basic displacement that characterizes the influence of mold elastic recovery.

[0108] The input conditions include the set of compensation strategies to be executed generated by step S5.1, which contains springback compensation strategy identifiers, semantic information of the current process stage from step S4, and low-frequency component signal data of the slider displacement deviation sequence obtained by multidimensional perturbation confidence vector analysis.

[0109] Establish a call relationship between the springback compensation strategy identifier and the preset springback compensation model. Use the identifier index to match the model parameter file to load the springback compensation calculation module corresponding to the current stamping part model and die number, and initialize the internal state of the model.

[0110] The semantic information of the process stage is input into the working condition adaptation layer of the springback compensation model. The adaptation layer adjusts the springback influence coefficient matrix according to the physical state corresponding to the process stages such as forming, intermediate pressure holding, and unloading, so that the springback amount calculated by the model conforms to the mold elasticity characteristics of the current stage.

[0111] Low-frequency component extraction is performed on the slider displacement deviation sequence. Multi-level discrete wavelet decomposition is used to retain the energy component in the 0~10Hz frequency band, and this component is used as the input signal of the springback compensation model to ensure that the calculation focuses on the slow displacement change caused by the elastic recovery of the mold.

[0112] Based on the parameter mapping table inside the model, the K value after process stage adaptation, the E value corresponding to the actual mold material properties, and the extracted LF amplitude are multiplied to obtain the quantitative springback compensation basic displacement.

[0113] The calculation results are superimposed with the historical springback error correction factor of the model to eliminate the cumulative deviation caused by long-term mold wear, and output the springback compensation base displacement with dynamic adaptability to working conditions.

[0114] By calling and adapting the springback compensation model, the strategy identifier, process semantic information and low-frequency displacement components from the previous step are transformed into quantitative displacement compensation data that reflects the influence of mold elastic recovery, thereby realizing the accurate calculation of basic compensation for springback disturbances.

[0115] S5.3: Using the material thickness fluctuation compensation strategy identifier in the set of compensation strategies to be executed, call the preset material thickness compensation model and input the mid-frequency component of the slider displacement deviation sequence and the material batch parameters to calculate the basic displacement of material thickness compensation that characterizes the influence of raw material size deviation.

[0116] Obtain the material thickness fluctuation compensation strategy identifier from the set of compensation strategies to be executed, perform bandpass filtering on the mid-frequency component of the slider displacement deviation sequence to isolate the frequency domain component corresponding to the material size fluctuation, and suppress irrelevant disturbance signals below the noise baseline.

[0117] The filtered intermediate frequency component is input into the preset material thickness compensation model. During the model initialization stage, the batch parameters of the current stamping part material are loaded, including the nominal thickness value and the batch thickness deviation tolerance. The compensation coefficient matrix of the model is set according to the batch parameters so that it can be adjusted in a targeted manner according to the characteristics of the current raw material.

[0118] By using the above method, the intermediate frequency component extracted by filtering and the material batch parameters are converted into an executable material thickness compensation base displacement, thereby realizing the generation of accurate compensation data for raw material size deviations.

[0119] S5.4: Based on the dynamic weight coefficients corresponding to the rebound compensation strategy identifier in the strategy spectrum response matrix, the basic displacement of rebound compensation is weighted and scaled to generate the initial displacement correction of rebound compensation carrying energy distribution characteristics.

[0120] The input conditions are the basic displacement of springback compensation and the dynamic weight coefficient of the corresponding springback compensation strategy identifier in the strategy spectrum response matrix. The dynamic weight coefficient is numerically extracted and used as a scaling factor input to the displacement weighting processing module. Amplitude normalization is performed on the basic displacement of springback compensation to eliminate dimensional differences between different stamping cycles and ensure that the weight coefficients maintain a linear proportional relationship. A multiplicative scaling method is used to perform a dot product operation between the normalized basic displacement of springback compensation and the dynamic weight coefficient in mathematical space to generate a scaled displacement matrix with disturbance energy distribution characteristics. The scaled displacement matrix undergoes directional consistency verification, eliminating vector components that do not conform to the geometric compatibility constraints of the process stage and retaining valid components that conform to the constraints. The scaled displacement matrix after directional consistency verification is converted back to the physical units of the original displacement and reassembled into the initial displacement correction output for springback compensation. By using a dynamic weighted scaling method, the basic displacement of the rebound compensation in the previous step is transformed into the initial displacement correction of the rebound compensation carrying the current disturbance energy distribution characteristics, so as to realize the real-time adaptive adjustment of the amplitude of the rebound compensation strategy as the disturbance spectrum changes.

[0121] S5.5: Based on the dynamic weight coefficients corresponding to the material thickness fluctuation compensation strategy identifier in the strategy spectrum response matrix, the basic displacement of material thickness compensation is weighted and scaled to generate the initial displacement correction amount of material thickness fluctuation compensation carrying energy distribution characteristics.

[0122] Step S6: Project the initial displacement correction amount onto a unified reference coordinate system to synthesize a compensation command vector, and determine whether the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system. Based on the determination result, dynamic compensation for the stamping depth is achieved. Specifically, this includes: S6.1: Obtain the initial displacement correction values ​​output by the springback compensation strategy and the material thickness fluctuation compensation strategy respectively. Based on the unified reference coordinate system defined by the press slider kinematic model, perform vector projection transformation on the initial displacement correction values ​​to generate a synthetic compensation command vector characterizing the superposition effect of multiple strategies.

[0123] The initial displacement corrections output by the springback compensation strategy and the material thickness fluctuation compensation strategy are used as input objects. The unified reference coordinate system parameter set defined in the kinematic model of the press slider is loaded, including the origin position, main motion axes and direction vector weights.

[0124] The components of the initial displacement correction amount for springback compensation in each axis of the unified reference coordinate system are analyzed, and coordinate transformation is performed according to the direction cosine matrix fixed by the model to obtain the sequence of its component values ​​in the unified reference coordinate system.

[0125] Perform the same direction cosine matrix transformation on the initial displacement correction amount for material thickness fluctuation compensation to generate a sequence of mapping values ​​of material thickness compensation components in a unified reference coordinate system, and ensure that the two sets of data are aligned on the time sampling reference.

[0126] By employing a vector superposition operation, the springback compensation component value sequence and the material thickness compensation component value sequence are summed one by one in a unified reference coordinate system to form a synthetic compensation command vector that includes all strategy contributions, where each vector element represents the total displacement correction in the corresponding axis.

[0127] By processing the unified kinematic projection of the multi-strategy displacement correction, the springback compensation and material thickness fluctuation compensation results of the previous step are transformed into a synthetic compensation command vector that can be directly used for mechanical limit and dynamic capability verification, thereby realizing the spatial interpretability of multi-strategy superposition.

[0128] S6.2: Read the preset slide mechanical limit safety envelope data in the current stamping process parameters, and based on the endpoint coordinate position of the synthetic compensation command vector, perform the geometric space inclusion detection method to generate a mechanical limit compliance flag that characterizes the mechanical interference risk.

[0129] S6.3: Collect the real-time dynamic capability boundary parameters of the servo drive system, including the maximum permissible speed and the maximum permissible acceleration. Based on the time differential characteristics of the synthetic compensation command vector, perform dynamic response matching degree evaluation calculation to generate a servo dynamic compliance flag that characterizes the overload risk of the actuator.

[0130] The system acquires real-time dynamic capability boundary parameters of the servo drive system, including maximum permissible speed and maximum permissible acceleration. It then calls a real-time monitoring interface matched to the servo motor model and driver specifications to obtain the current speed and acceleration limits as dynamic capability baseline data. Based on the timestamp sequence and displacement component sequence of the synthetic compensation command vector, it performs discrete-time difference operations to calculate the instantaneous velocity component vector and performs vector modulo operations on the velocity component vector to obtain the global instantaneous velocity value. The same sequence is then subjected to second-order difference operations to generate the instantaneous acceleration component vector, and vector modulo operations are performed on the acceleration component vector to obtain the global instantaneous acceleration value.

[0131] S6.4: Receive the mechanical limit compliance flag and the servo dynamic compliance flag, and perform a fusion judgment on the state values ​​of the two flags based on the logical AND operation rules to generate an out-of-bounds judgment result that represents whether the synthesized compensation command vector exceeds the safety boundary.

[0132] Receive the mechanical limit compliance flag generated by the preceding sub-step S6.2 and the servo dynamic compliance flag generated by S6.3 as input data, and ensure that the status values ​​of the two have been represented as Boolean values ​​according to the unified encoding rules (true value indicates compliance, false value indicates non-compliance).

[0133] The mechanical limit compliance flag and the servo dynamic compliance flag are sent to the dual-input logic AND operation unit in the order of the input channels of the logic operation module, and the Boolean operation rule is set to output the true value only when both input terminal status values ​​are true.

[0134] The fusion result R is mapped to the preset boundary judgment semantic mapping table. If R is true, it is judged as a non-boundary state; if R is false, it is judged as a boundary state, and a boundary judgment result data object containing a state identifier is generated.

[0135] Through the above-mentioned logical AND operation and semantic mapping processing, the detection result of the previous step is transformed into an out-of-bounds judgment result that can be used for security closed-loop decision-making, thereby realizing the integrated judgment of cross-domain security conditions of synthetic compensation instruction vector.

[0136] S6.5: Based on the Boolean state value of the boundary crossing determination result, perform control flow branch routing operation. If it is determined that there is no boundary crossing, output a pass-through permission signal. If it is determined that there is a boundary crossing, trigger the downgrade coordination process start instruction to complete the closed-loop decision output for the security verification of the compensation instruction vector.

[0137] The Boolean state value of the boundary judgment result generated by the previous step S6.4 is received as the input condition for this step, and the Boolean state value is bound to the safety assessment task of the synthetic compensation command vector of the current stamping cycle.

[0138] Parse the binary flag of the out-of-bounds judgment result, establish a state mapping table, map a true flag value to a trigger signal to start the downgrade coordination process, and map a false flag value to an output instruction for a pass-through permission signal.

[0139] After the flag value is parsed, the synthetic compensation command vector of the current stamping cycle is matched with the state mapping table, and the dual-branch routing selection is performed through the logic branch operation module to ensure that the processing paths of commands with different state values ​​are completely isolated.

[0140] A hard real-time execution framework that binds state values ​​to branch paths is adopted. The pass-through permission signal is directly forwarded to the servo drive execution unit via a high-speed communication bus to achieve the issuance of compensation instructions without delay. The trigger signal for starting the degraded coordination process is encapsulated into a degraded process start command by the task scheduler and sent to the safety correction subsystem of the compensation control module.

[0141] A signal integrity verification mechanism is established to perform CRC checks and timing consistency checks on the output of the direct path and the output of the degraded path, preventing branch path execution errors caused by misinterpretation of state values. Through this processing method, the out-of-bounds judgment result of the previous step is transformed into a clear control flow branch routing decision, realizing the final output of the instruction security closed loop.

[0142] For example, in a home appliance stamping parts manufacturing line, the obtained boundary judgment result Boolean value is false. The flag parsing module reads the value and maps it to a pass-through permission signal. The synthesized compensation command vector is [0.12mm, 0.08mm]. The logic branch operation module selects the pass-through path and transmits the vector to the servo drive unit via the high-speed bus. The servo drive system executes the slider displacement correction action within 10ms. Finally, the measured slider position meets the preset compensation target at the end of the forming process, and the mechanical limit envelope and dynamic capability parameters are fully compliant. In another scenario, the out-of-bounds determination result is true (Boolean value), and the synthesized compensation command vector is [0.35mm, 0.21mm]. After being parsed and mapped by the flag bit, it is mapped to a reduced-order start command. The safety correction subsystem of the compensation control module receives the command and freezes the low-confidence strategy components. After removing the out-of-bounds contribution, it recalculates the reduced-order compensation command and obtains a correction amplitude of 0.28mm through the formula. After applying a smooth transition filter, it is sent to the servo drive system. After the slider performs the correction action, it remains within the safety envelope. Verification data shows that this process significantly improves the stability and robustness of the system under multiple disturbance concurrent conditions.

[0143] Step S7: If the compensation command vector exceeds the safety envelope of the slider's mechanical limit or the dynamic capability boundary of the servo system, a reduced-order coordination process is initiated. This involves freezing the principal components of the low-confidence strategy and superimposing a smooth transition filter to generate the final compensation control command that satisfies geometric compatibility constraints. Specifically, this includes: S7.1: Obtain the initial displacement correction set and the corresponding multidimensional disturbance confidence vector generated by the previous steps, perform kinematic space projection processing on the initial displacement correction set to generate a synthetic compensation command vector containing the direction and amplitude of each strategy component, and compare the synthetic compensation command vector with the preset slider mechanical limit safety envelope and servo system dynamic capability boundary to output the safety boundary judgment result indicating the out-of-bounds state.

[0144] The initial displacement correction set and multidimensional disturbance confidence vector transmitted by the preceding step S6 are input and received to ensure the integrity of the data under a unified time base and a unified reference coordinate system.

[0145] A three-dimensional vector projection operation based on the kinematic model of the press slider is performed on the initial displacement correction set to map the displacement correction of each strategy component from the local strategy reference system to the unified reference coordinate system, so that each strategy component has the direction and amplitude characteristics that can be directly superimposed.

[0146] The mechanical limit compliance flag and the servo dynamic compliance flag are input into the logic fusion calculation module. The safety boundary judgment result is generated through Boolean logic AND rules to indicate whether the synthesized compensation command vector has exceeded the boundary.

[0147] Through the aforementioned kinematic space projection, vector synthesis, geometric inclusion detection, and dynamic capability matching evaluation, the initial displacement correction and confidence data from the previous step are transformed into safety boundary determination results that can be used in the subsequent order reduction coordination process, thereby achieving safety pre-verification of the parallel output of multiple strategies.

[0148] S7.2: Based on the boundary crossing status identifier in the safety boundary determination result and the numerical distribution of the multidimensional disturbance confidence vector, perform confidence sorting and threshold filtering on each strategy component in the initial displacement correction set to identify and mark the low-confidence strategy principal components to be frozen, thereby generating a strategy classification list containing high-confidence retained components and low-confidence frozen components.

[0149] S7.3: Using the low-confidence frozen component information in the policy hierarchy list, perform component removal and vector reconstruction processing on the synthetic compensation instruction vector to remove low-confidence policy contribution values ​​that cause out-of-bounds errors, thereby generating a reduced-order compensation instruction intermediate vector consisting only of high-confidence retained components.

[0150] The input conditions include the identification information of the high-confidence retained components and low-confidence frozen components output from the policy grading list, as well as the magnitude and direction dataset of the synthetic compensation command vector generated in the preceding steps.

[0151] The synthetic compensation command vector is matched and searched by component index to locate the vector component corresponding to the low confidence frozen component to its specific coordinate position and amplitude range in the synthetic vector structure.

[0152] The amplitude of the low confidence component obtained from the localization is set to zero, and its direction vector is normalized and eliminated to obtain the intermediate vector skeleton structure after the removal operation.

[0153] The remaining components in the intermediate vector skeleton are reconstructed using the high-confidence retained component index. The amplitude is then limited by calling the dynamic capability boundary parameters of the servo system to prevent the reconstructed amplitude from exceeding the safe envelope.

[0154] By using component removal and amplitude reconstruction, the strategy classification result of the previous step is transformed into a reduced-order compensation instruction intermediate vector containing only high-confidence retained components, thereby achieving multi-strategy collaborative execution within the safety envelope.

[0155] S7.4: To address the potential step abrupt changes that may occur in the intermediate vector of the reduced-order compensation command during strategy switching, an adaptive smooth transition filtering method is applied to perform differential constraint processing on the time-domain waveform of the intermediate vector of the reduced-order compensation command, in order to eliminate high-frequency jitter and limit the rate of acceleration change, thereby generating a smooth transition compensation command sequence that satisfies the dynamic response characteristics of the servo system.

[0156] S7.5: Based on the smooth transition compensation command sequence and the geometric compatibility constraint rules of the stamping process, the final amplitude and direction of the command vector are verified and fine-tuned to output the final compensation control command that is strictly within the safety envelope of the slider mechanical limit and has geometric compatibility, thus completing the closed-loop output of the order reduction coordination process.

[0157] Based on the generated smooth transition compensation command sequence and the geometric compatibility constraint rules of the stamping process, the compensation command sequence is used as the input signal to load the press slide geometric motion constraint model to establish a joint verification framework for amplitude and direction.

[0158] The amplitude limit detection operation is performed on each vector component of the compensation command sequence, and the Euclidean distance between its endpoint coordinates and the boundary curve of the slider mechanical limit safety envelope is calculated. If the distance is less than the preset safety margin, the component is marked as an object that needs fine-tuning.

[0159] The directional compatibility verification method is applied to the components marked as objects requiring fine-tuning to calculate the angle between them and the compensation direction vector allowed in the process stage, and the rotation matrix is ​​corrected for the out-of-limit directions based on the angle threshold.

[0160] All components after direction and amplitude correction are vector recombined to form the final set of compensation command vectors in a unified kinematic reference coordinate system, and geometric compatibility is checked again to ensure that there is no inconsistency in direction caused by strategy conflict.

[0161] By using the above-mentioned direction correction and amplitude compression processing methods, the smooth transition compensation command sequence of the previous step is transformed into a final compensation control command that is strictly within the safety envelope of the slider mechanical limit and has geometric compatibility, thereby achieving the expected technical effects of conflict resolution and safety assurance.

[0162] Step S8: Based on the final compensation control command, drive the press slide to perform displacement compensation action, and feed back the actual slide position after execution to the closed-loop control system to update the slide displacement deviation sequence for the next cycle. Specifically, this includes: S8.1: Obtain the final compensation control command after the reduction coordination process, and perform digital-to-analog conversion and servo drive signal mapping processing on the final compensation control command to generate the target torque setpoint for driving the press servo motor.

[0163] S8.2: Based on the target torque setpoint, perform high-precision displacement compensation on the press slide, and eliminate the influence of mechanical transmission backlash through a servo closed-loop adjustment mechanism to output the actual physical position of the slide containing dynamic correction.

[0164] S8.3: The actual physical position of the slider is sampled and digitally encoded in real time using a high-resolution grating ruler sensor to obtain measured displacement feedback data that characterizes the end point of the current stamping stroke.

[0165] S8.4: Perform deviation reconstruction calculation on the measured displacement feedback data according to the preset ideal stamping depth trajectory curve to calculate the updated slider displacement deviation sequence.

[0166] S8.5: Write the updated slider displacement deviation sequence into the data buffer of the closed-loop control system and replace the historical data of the previous cycle to form the initial input source for generating the two-dimensional disturbance spectrum for the next stamping cycle.

[0167] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0168] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0169] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement, specifically including: S1: Obtain the high sampling rate slider displacement deviation sequence within the stamping cycle, and perform short-time Fourier transform and continuous wavelet transform on the slider displacement deviation sequence to generate a two-dimensional perturbation spectrum characterizing the energy of multi-source perturbation. S2: Based on the time-frequency energy distribution characteristics and process stage semantic information of the two-dimensional perturbation spectrum, construct a perturbation spectrum semantic tag system that includes typical forms such as cold material entering the mold, insufficient mold preheating, and springback dominance. S3: Utilize the semantic tagging system of the disturbance spectrum to perform interpretable template matching on the current two-dimensional disturbance spectrum, and combine the context metadata of the stamping part model and mold number to generate a multi-dimensional disturbance confidence vector reflecting the confidence of each disturbance source; S4: Based on the energy distribution and evolution trend of the multidimensional perturbation confidence vector, and following the three-layer logic of perturbation-dominant dimension, strategy-sensitive dimension, and process stage constraints, dynamically generate a strategy spectrum response matrix that defines the strategy activation weight; S5: Based on the strategy spectrum response matrix, automatically activate the most suitable compensation strategy combination, and calculate the initial displacement correction amount based on the compensation strategy combination; S6: Project the initial displacement correction amount onto a unified reference coordinate system to synthesize a compensation command vector, and determine whether the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system. Based on the determination result, realize dynamic compensation for the stamping depth.

2. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, According to step S6, dynamic compensation for the stamping depth is achieved based on the judgment result, including: S7: If it is determined that the compensation command vector exceeds the safety envelope of the slider mechanical limit or the dynamic capability boundary of the servo system, the order reduction coordination process is started. By freezing the principal components of the low confidence strategy and superimposing a smooth transition filter, the final compensation control command that satisfies the geometric compatibility constraint is generated. S8: Based on the final compensation control command, drive the press slide to perform displacement compensation action, and feed back the actual slide position after execution to the closed-loop control system to update the slide displacement deviation sequence for the next cycle.

3. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, The optimal combination of compensation strategies includes a springback compensation strategy and a material thickness fluctuation compensation strategy.

4. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, Step S3 specifically includes: The two-dimensional perturbation spectrum and the semantic tag system of the perturbation spectrum of the current stamping cycle are obtained. Normalization and alignment processing is performed on the two-dimensional perturbation spectrum and the typical morphological templates in the semantic tag system of the perturbation spectrum to eliminate the phase shift caused by the difference in sampling time reference and generate aligned perturbation spectrum data with unified spatiotemporal reference. Based on the aligned perturbation spectrum data, the weighted Euclidean distance method is used to calculate the structural similarity value between it and the cold material entry template, insufficient mold preheating template and springback dominant template in the semantic labeling system of the perturbation spectrum, so as to quantify the matching degree between the current spectrum shape and each known perturbation mode and generate an initial matching result vector containing multiple basic similarity components. Obtain the stamping part model identifier and mold number identifier associated with the current production task as context metadata, and generate a weighted matching score vector calibrated by working conditions based on the context metadata; The weighted matching score vector calibrated under operating conditions is subjected to maximum value normalization processing, and the weighted matching score corresponding to each disturbance mode is mapped to the probability interval of zero to one to characterize the relative probability of each disturbance source in the current stamping cycle, and a preliminary disturbance probability distribution sequence is generated. Based on the preliminary disturbance probability distribution sequence, vector recombination and confidence assignment are performed according to the disturbance source classification dimensions such as material thickness fluctuation, mold elastic deformation and transient impact. The preliminary disturbance probability distribution sequence is converted into a confidence value that can be directly used for strategy weight calculation, and the final multidimensional disturbance confidence vector is generated.

5. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 4, characterized in that, The step of generating a weighted matching score vector calibrated based on contextual metadata includes retrieving the corresponding material thickness fluctuation coefficient and mold wear correction factor from the historical process knowledge base based on the contextual metadata, so as to perform condition-adaptive weighted correction on the basic similarity component in the initial matching result vector, and generate a weighted matching score vector calibrated based on the operating conditions.

6. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, Step S4 specifically includes: The confidence values ​​of each disturbance source contained in the multidimensional disturbance confidence vector are normalized and their energy proportions are calculated to extract the disturbance-dominant dimension identifier that represents the most important disturbance source in the current stamping cycle. Based on the disturbance-dominant dimension identifier, a list of strategy sensitivity dimension parameters containing the sensitivity coefficients of each candidate compensation strategy is obtained; The current stamping stroke information is used to determine the specific process stage constraints, and the process stage constraints are logically ANDed with the strategy sensitivity dimension parameter list to generate a preliminary strategy activation weight vector after process feasibility filtering. Based on the numerical magnitude and evolution slope of each component in the initial strategy activation weight vector, a dynamic weighted fusion method is executed to comprehensively consider the disturbance energy intensity and the allowable range of the process stage, and calculate the final strategy spectrum response matrix that defines the specific effect intensity of each compensation strategy, thus completing the dynamic mapping from disturbance characteristics to control weights.

7. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 6, characterized in that, The step of obtaining a list of strategy sensitivity dimension parameters containing sensitivity coefficients of each candidate compensation strategy based on the disturbance dominance dimension identifier includes performing a strategy sensitivity matching operation based on the disturbance dominance dimension identifier to filter out the set of candidate compensation strategies that are most sensitive to the current dominant disturbance type, and generating a list of strategy sensitivity dimension parameters containing each set of candidate compensation strategies.

8. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, The slider displacement deviation sequence is segmented by a Hanning window, and each data segment is subjected to a Fast Fourier Transform after windowing to obtain a set containing low, medium and high frequency discrete spectral components.

9. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, The construction of a perturbation spectrum semantic labeling system includes typical forms such as cold material entering the mold, insufficient mold preheating, and springback dominance. This includes identifying cold material entering the mold perturbation as a composite feature of low-frequency sudden increase and mid-frequency continuous rise, identifying springback dominance perturbation as a combination feature of high-frequency burrs and low-frequency baseline drift in the second half of the process, and establishing a unique label identifier and response template. The perturbation spectrum semantic labeling system is obtained based on the unique label identifier and response template.

10. The dynamic compensation control method for stamping depth based on closed-loop feedback of press slide displacement according to claim 1, characterized in that, The step of projecting the initial displacement correction amount onto a unified reference coordinate system to synthesize a compensation command vector includes using the three-dimensional space defined by the unified kinematic model of the press slider to project and superimpose the initial displacement correction amount through a direction cosine matrix to obtain the compensation command vector.

Images