Control system and method of intelligent servo press

Abstract

The invention discloses a control system and method of an intelligent servo press, relates to the technical field of press control, and constructs a uniform stamping state representation through deep excavation and fusion of force and displacement process data streams generated in the servo press. On the basis, a double-time-scale implicit state estimation algorithm is introduced, time sequence modeling and deep analysis are carried out on weak and gradual change characteristics contained in a full-stroke force displacement curve, and therefore accurate quantitative evaluation of the implicit state of mold abrasion is achieved. Then, the quantized wear value serves as a core decision factor, a self-adaptive control mechanism based on a compensation knowledge base is driven, and an optimal process compensation strategy and a sliding block target track are dynamically generated. Therefore, the product quality consistency in the precise stamping process can be improved, the service life of the die is effectively prolonged, the production cost is reduced, and the intelligent level of stamping production is comprehensively improved.

Description

Technical Field

[0001] This application relates to the field of press control technology, and more specifically, to a control system and method for an intelligent servo press. Background Technology

[0002] With the rapid development of high-end manufacturing, especially in the automotive, aerospace, and precision electronics industries, the requirements for precision forming of metal sheets and components are becoming increasingly stringent. Servo presses, with their significant advantages such as programmable slider motion trajectory, high control precision, and energy efficiency, have become key equipment for achieving high-precision, high-flexibility stamping production, profoundly driving the advancement of stamping technology. In precision stamping, the stability of product quality and the consistency of processes are crucial. However, under repeated contact, friction, and high pressure with the sheet metal, the working surface of the stamping die inevitably undergoes progressive wear. Once this wear accumulates to a certain extent, it leads to quality problems such as dimensional deviations, surface scratches, and shape defects in the product. In severe cases, it can even cause batch scrapping and increase the time and cost of unplanned downtime for die replacement, directly impacting production efficiency.

[0003] In existing technologies, the main strategies for addressing the challenges posed by mold wear are offline periodic inspection or planned pre-replacement. Offline inspection requires production interruption and relies on manual or specialized instruments for mold measurement and evaluation, resulting in slow response and inability to intervene in the early stages of quality deterioration. Pre-replacement strategies based on production cycle time or experience are often too conservative or delayed, making it difficult to accurately match the actual health status of the mold, potentially leading to wasted mold lifespan or product quality issues. Although some research has attempted to introduce external sensor systems such as acoustic emission, vibration, or machine vision for online condition monitoring, these solutions not only significantly increase equipment complexity and cost but are also susceptible to interference from noise signals such as oil stains and vibrations in the field environment, making data processing and integration difficult. Most existing technologies remain at the level of simple threshold alarms for isolated characteristics such as peak force in a single stroke, lacking the ability to identify weak, gradual characteristic information caused by wear that is reflected in the entire force and displacement process curve, and lacking precise quantitative assessment methods for wear status.

[0004] Therefore, there is an urgent need for an optimized control system and method for intelligent servo presses. Summary of the Invention

[0005] This application is made in order to solve the above-mentioned technical problems.

[0006] According to one aspect of this application, a control method for an intelligent servo press is provided, comprising: Receive raw pressure data and raw slider position data from the servo motor encoder and force sensor; The raw pressure data and raw slider position data are preprocessed to obtain the force-displacement curve of the kth stroke. Wear-sensitive features are extracted from the force-displacement curve of the k-th stroke to obtain the feature vector of the k-th stroke; The eigenvector of the k-th stroke is subjected to implicit state estimation on a dual time scale to obtain a quantized wear estimate. Based on the compensation knowledge base, an adaptive process compensation strategy is generated for the quantified wear estimate to obtain the compensated target trajectory. Based on the compensated target trajectory and the actual position of the slider, a servo motor torque command is generated.

[0007] According to another aspect of this application, a control system for an intelligent servo press is provided, comprising: The raw data acquisition module is used to receive raw pressure data and raw slider position data from the servo motor encoder and force sensor; The data preprocessing module is used to preprocess the raw pressure data and raw slider position data to obtain the force-displacement curve of the k-th stroke. The wear-sensitive feature extraction module is used to extract wear-sensitive features from the force-displacement curve of the k-th stroke to obtain the feature vector of the k-th stroke. The implicit state estimation module is used to perform dual-time-scale implicit state estimation on the feature vector of the k-th stroke to obtain a quantized wear estimate. The compensation strategy generation module is used to generate an adaptive process compensation strategy based on the quantitative wear estimate using a compensation knowledge base to obtain the compensated target trajectory. The control command generation module is used to generate servo motor torque commands based on the compensated target trajectory and the actual position of the slider.

[0008] Compared with existing technologies, this application provides a control system and method for an intelligent servo press. It constructs a unified stamping state representation by deeply mining and integrating the inherent force and displacement process data stream of the servo press. Based on this, a dual-timescale implicit state estimation algorithm is introduced to perform time-series modeling and in-depth analysis of the weak and gradual features contained in the full-stroke force-displacement curve, thereby achieving accurate quantitative assessment of the implicit state of die wear. Then, this quantified wear value serves as the core decision factor, driving an adaptive control mechanism based on a compensation knowledge base to dynamically generate the optimal process compensation strategy and slide target trajectory. This improves product quality consistency in the precision stamping process, effectively extends die life, reduces production costs, and comprehensively enhances the intelligence level of stamping production. Attached Figure Description

[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flowchart of a control method for an intelligent servo press according to an embodiment of this application.

[0011] Figure 2 This is a data flow diagram of the control method for an intelligent servo press according to an embodiment of this application.

[0012] Figure 3 This is a flowchart of sub-step S3 of the control method for an intelligent servo press according to an embodiment of this application.

[0013] Figure 4 This is a flowchart of sub-step S32 of the control method for an intelligent servo press according to an embodiment of this application.

[0014] Figure 5 This is a flowchart of sub-step S4 of the control method for an intelligent servo press according to an embodiment of this application.

[0015] Figure 6 This is a flowchart of sub-step S5 of the control method for an intelligent servo press according to an embodiment of this application.

[0016] Figure 7 This is a flowchart of sub-step S6 of the control method for an intelligent servo press according to an embodiment of this application.

[0017] Figure 8 This is a block diagram of the control system of an intelligent servo press according to an embodiment of this application. Detailed Implementation

[0019] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0020] To address the problems mentioned above in the background technology, this application proposes a control method for an intelligent servo press. Figure 1 This is a flowchart of a control method for an intelligent servo press according to an embodiment of this application. Figure 2 This is a data flow diagram of the control method for an intelligent servo press according to an embodiment of this application. Figure 1 and Figure 2 As shown, the control method of the intelligent servo press includes the following steps: S1, receiving raw pressure data and raw slider position data from the servo motor encoder and force sensor; S2, preprocessing the raw pressure data and raw slider position data to obtain the force-displacement curve of the k-th stroke; S3, extracting wear-sensitive features from the force-displacement curve of the k-th stroke to obtain the feature vector of the k-th stroke; S4, performing dual-time-scale implicit state estimation on the feature vector of the k-th stroke to obtain a quantized wear estimate; S5, generating an adaptive process compensation strategy based on the compensation knowledge base to obtain a compensated target trajectory; S6, generating a servo motor torque command based on the compensated target trajectory and the actual slider position.

[0021] In the control method of the aforementioned intelligent servo press, step S1 involves receiving raw pressure data and raw slider position data from the servo motor encoder and force sensor. It should be understood that since the implicit state of die wear during the stamping process cannot be directly observed, it needs to be indirectly characterized through core data reflecting the physical process of stamping. Slider position and pressure are key physical quantities reflecting the interactive state of stamping. Therefore, this application uses adapted sensors and encoders to capture raw pressure signals and slider position signals in real time to obtain unmodified raw pressure data and raw slider position data. This ensures the originality and integrity of the data, avoids deviations in subsequent analysis due to missing or distorted data, and builds a reliable data foundation for the accurate quantitative assessment of die wear.

[0022] Specifically, in one possible embodiment, step S1 is implemented as follows: First, a servo motor encoder with a resolution of no less than 1024 lines is selected and fixed to the output shaft of the servo motor to capture the position feedback signal of the slider movement in real time. The sampling frequency is set to 5kHz to ensure the continuity of the time dimension. Second, a piezoelectric force sensor with a range matching the maximum force of the stamping process is installed on the connection surface between the slider and the upper die base to directly sense the pressure changes during the stamping process. The sampling frequency is kept consistent with the encoder to achieve time synchronization. Finally, the output signals of the sensor and encoder are transmitted to a dedicated data acquisition card through the EtherCAT industrial bus to complete the synchronous reception and real-time temporary storage of the original pressure data and slider position data, ensuring that there is no packet loss in the data transmission.

[0023] In the control method of the aforementioned intelligent servo press, step S2 involves preprocessing the raw pressure data and raw slider position data to obtain the force-displacement curve for the k-th stroke. It should be understood that the raw data collected by the sensors contains noise such as electromagnetic interference and mechanical vibration, and the position and pressure data may have issues such as asynchronous sampling and starting point offset. Directly using this data for analysis can lead to feature extraction errors. Therefore, this application further performs noise reduction, synchronization calibration, and stroke alignment processing on the raw pressure data and raw slider position data to remove invalid interference information and standardize the data format. This eliminates the impact of noise and synchronization deviation on data quality, resulting in a continuous, smooth force-displacement curve that accurately corresponds to a single stroke, ensuring the accuracy of subsequent calculations and providing more reliable data support.

[0024] Specifically, in one possible embodiment, step S2 is implemented as follows: First, wavelet transform is used to denoise the original pressure and position data. A db4 wavelet basis function is selected, and after decomposition to three levels, high-frequency noise components are removed to reconstruct clean data. Second, the two clean data streams are synchronously calibrated based on timestamps. Missing sampling points are filled in using linear interpolation to ensure that each time corresponds to a unique position and pressure value. Then, the effective data segment of a single stroke is extracted, starting from the pressure abrupt change point when the slider touches the workpiece and ending at the slider returning to its initial position. Finally, the processed position data is used as the x-axis and the pressure data as the y-axis to construct a complete force-displacement curve for the k-th stroke, which is stored in a data buffer for subsequent processing.

[0025] In the control method of the aforementioned intelligent servo press, step S3 involves extracting wear-sensitive features from the force-displacement curve of the k-th stroke to obtain the feature vector of the k-th stroke. It should be understood that since the force-displacement curve is continuous high-dimensional data, it contains redundant information unrelated to wear, such as idle strokes, and the feature changes caused by wear are weak and scattered. Directly using it for state estimation would increase computational complexity and introduce interference. Therefore, this application further filters and quantifies key wear-sensitive physical features from the force-displacement curve, integrating them into a low-dimensional feature vector for the k-th stroke. This focuses on wear-related information and reduces the difficulty of subsequent algorithm processing. This provides accurate input for dual-timescale implicit state estimation, ensuring the efficiency and accuracy of wear quantification assessment and avoiding estimation bias.

[0026] In particular, in one specific embodiment, Figure 3 This is a flowchart of sub-step S3 of the control method for an intelligent servo press according to an embodiment of this application. Figure 3As shown, step S3 includes: S31, performing dynamic segmentation of the force-displacement curve of the k-th stroke to obtain the shearing stage data segment and the unloading stage data segment; S32, performing parallel calculation of the associated physical characteristics of the shearing stage data segment, the unloading stage data segment, and the force-displacement curve of the k-th stroke to obtain the original shearing slope, the original unloading peak force, and the original total punching work; S33, vectorizing the original shearing slope, the original unloading peak force, and the original total punching work to obtain the feature vector of the k-th stroke.

[0027] Specifically, in step S31, the force-displacement curve of the k-th stroke is dynamically segmented into process stages to obtain data segments for the shearing stage and the unloading stage. It should be understood that the entire stamping stroke includes multiple stages such as feeding, positioning, shearing, and unloading. Only the force-displacement changes in the shearing stage (interaction between the die edge and the material) and the unloading stage (friction between the workpiece and the die surface) are directly related to die wear; data from other stages dilutes wear information. Therefore, this application further divides the shearing stage and the unloading stage based on the characteristic abrupt change points of the force-displacement curve to focus on the core process range and eliminate data interference from irrelevant stages. This ensures that the physical features calculated subsequently all come from stages directly interacting with die wear, improving the sensitivity of features to wear changes, avoiding deviations in wear feature extraction caused by data from irrelevant stages, and laying a foundation for accurate extraction of wear-sensitive features.

[0028] Specifically, in one possible embodiment, step S31 is implemented as follows: First, the force-displacement curve is traversed to locate the peak point corresponding to the maximum punching force and its displacement coordinates. Second, during the force rise phase before reaching the peak point, two points are determined where the force value is 20% and 80% of the peak force, respectively, and all data between these two points are extracted as the shearing stage data segment. Then, during the slider return phase (displacement decreasing direction), the peak point of the unloading force is located, and a displacement window of ±0.5mm is selected with this point as the center, and the data within the window is extracted as the unloading stage data segment.

[0029] Specifically, in step S32, the associated physical characteristics of the data segments of the shearing stage, the unloading stage, and the force-displacement curve of the k-th stroke are calculated in parallel to obtain the original shearing slope, the original unloading peak force, and the original total punching work. It should be understood that the force-displacement change in the shearing stage reflects the sharpness of the die cutting edge (wear causes the cutting edge to become blunt, which in turn changes the slope), the force change in the unloading stage reflects the friction state of the die surface (wear causes increased roughness, which in turn changes the peak force), and the total punching work reflects the overall wear energy consumption impact. Furthermore, serial calculation would prolong processing time and fail to match the stamping cycle. Therefore, this application further calculates the associated physical characteristics for each of the three intervals separately and uses parallel calculation to capture the wear impact from multiple dimensions and improve processing efficiency. In this way, wear information can be covered from different dimensions through three features, ensuring that the wear state is fully characterized. Simultaneously, parallel calculation keeps the processing cycle within the stamping cycle, meeting the timeliness requirements of real-time control.

[0030] In particular, in one specific embodiment, Figure 4 This is a flowchart of sub-step S32 of the control method for an intelligent servo press according to an embodiment of this application. Figure 4 As shown, step S32 includes: S321, performing linear regression based on least squares on the data segment of the shearing stage to obtain the original shear slope; S322, taking the maximum absolute value of all force values ​​in the data segment of the unloading stage as the original unloading peak force; S323, applying the trapezoidal rule to the force-displacement curve of the k-th stroke to perform numerical integration to obtain the original total punching work.

[0031] More specifically, step S321 involves performing a least-squares linear regression on the shearing stage data segment to obtain the original shearing slope. It should be understood that during the shearing stage, since the die cutting edge contacts the workpiece and completes material shearing, the force-displacement relationship in this stage exhibits an approximately linear characteristic. The slope directly represents the rate of change of force with displacement. Specifically, the sharper the cutting edge, the smaller the material shearing resistance, and the smaller the slope. Wear causes the cutting edge to become blunt, increasing the shearing resistance and significantly increasing the slope. If the slope is calculated directly from two points, it is easily affected by instantaneous noise (such as force fluctuations caused by minute impurities in the material), failing to reflect the overall trend. Therefore, this application uses the least-squares method for linear regression on the shearing stage data segment to fit the overall linear relationship between force and displacement, accurately extracting the slope value. This effectively eliminates instantaneous fluctuation noise in the data, obtaining a statistically significant original shearing slope, accurately quantifying the change in cutting edge sharpness, and providing a reliable cutting edge wear correlation index for subsequent wear state estimation.

[0032] Specifically, in one possible embodiment, step S321 is implemented as follows: First, the data segment of the shearing stage is preprocessed to remove abnormal data points caused by instantaneous impact, i.e., points where the force value exceeds the mean of this stage by ±3 times the standard deviation. Second, the preprocessed position data is used as the independent variable and the force data as the dependent variable, and the least squares linear regression formula is substituted to calculate the slope parameter of the fitted line. Finally, the coefficient of determination R² of the fitted line is verified to ensure that R² ≥ 0.98. If it does not meet the requirement, the data segment division boundary is re-checked and adjusted until the original shearing slope that meets the accuracy requirements is obtained, ensuring the correlation between the slope and the actual wear state of the cutting edge.

[0033] More specifically, in step S322, the maximum absolute value of all force values ​​in the unloading stage data segment is taken as the original unloading peak force. It should be understood that during the unloading stage, the slider drives the mold back, causing friction between the workpiece and the mold surface. The force values ​​collected by the force sensor will be positive or negative depending on the direction of the force—positive for the pressing force and negative for the unloading pulling force. If the maximum force value is directly taken, the maximum unloading force in the negative direction will be missed. Furthermore, after mold wear, the surface roughness increases, and the frictional resistance during unloading increases, causing the maximum unloading force to rise regardless of whether it is positive or negative. Therefore, this application further takes the absolute value of all force values ​​in the unloading stage data segment and then extracts the maximum value to eliminate the interference of the force direction on the peak force judgment and accurately obtain the maximum unloading force. This comprehensively covers the force value changes in different directions during the unloading stage, ensuring that the obtained original unloading peak force accurately reflects the changes in frictional resistance caused by mold surface wear, and avoiding the omission of wear characteristics due to directional deviations.

[0034] Specifically, in one possible embodiment, step S322 is implemented as follows: First, all position-force data pairs in the unloading stage data segment are traversed, and the force value of each data point is extracted. Second, each force value is converted to its absolute value to eliminate differences in positive and negative signs. Then, all converted absolute force values ​​are compared to determine the maximum value. Finally, the position coordinates corresponding to the maximum value are recorded as the original unloading peak force and its measured position, ensuring that the peak force completely corresponds to the physical action process of the unloading stage and avoiding misjudging force values ​​generated by non-unloading actions as unloading peak forces.

[0035] More specifically, in step S323, the trapezoidal rule is applied to the force-displacement curve of the k-th stroke for numerical integration to obtain the original total punching work. It should be understood that since punching work is the sum of the work done by the die on the material throughout the entire stamping stroke, it directly reflects the energy consumption of the entire stamping process. Die wear leads to decreased cutting efficiency and increased surface friction, both of which increase the total punching work. However, the force-displacement curve is a non-standard analytical curve, making analytical integration impossible. The trapezoidal rule can achieve high-precision numerical integration based on discrete data points with minimal computational cost. Therefore, this application further divides the force-displacement curve of the k-th stroke into intervals according to sampling intervals and applies the trapezoidal rule for numerical integration to accurately calculate the total punching work for the entire stroke. This adapts to the non-analytical characteristics of the force-displacement curve, ensuring the accuracy of the original total punching work calculation while controlling the computational cost to meet real-time processing requirements. This allows the total punching function, as a characteristic of the overall wear state, to complement local characteristics (slope, peak force), comprehensively depicting the impact of wear.

[0036] Specifically, in one possible embodiment, step S323 is implemented as follows: First, the integration interval is determined to be from the starting position of the slider's downward movement to its initial position upon returning upwards, covering the entire stamping stroke. Second, the integration interval is divided into several small trapezoids according to the data sampling interval. The upper and lower bases of each small trapezoid represent the force values ​​of two adjacent sampling points, and the height represents the positional difference between adjacent sampling points. Then, the area of ​​each small trapezoid is calculated (area = (upper base + lower base) × height / 2). Finally, the areas of all small trapezoids are summed to obtain the original total stamping energy, ensuring that the integration result is consistent with the energy consumption of the actual stamping process, and the integration error is controlled within 1%.

[0037] Specifically, in step S33, the original shear slope, original unloading peak force, and original total punching work are vectorized to obtain the feature vector of the k-th stroke. It should be understood that since the original shear slope, original unloading peak force, and original total punching work are independent scalars with large dimensional differences, directly inputting them into subsequent algorithms would cause the algorithm to be overly sensitive to features with large values, and the dispersed scalars cannot form a structured input. Therefore, this application further normalizes the three original features and then combines them into a vector in a fixed order to eliminate dimensional differences and construct a structured input. This ensures that the three features contribute equally in subsequent state estimation, avoiding dimensional bias, while the vector form adapts to the model input requirements, ensuring that the model can efficiently integrate multi-dimensional wear information and improve the accuracy of wear estimation.

[0038] Specifically, in one possible embodiment, step S33 is implemented as follows: First, 100 stroke feature data collected during the initial stage of the new mold are retrieved, and the maximum and minimum values ​​of the shear slope, the peak unloading force, and the total punching work are statistically obtained as normalization benchmarks. Second, min-max normalization is performed on each original feature to map the original values ​​to the [0,1] interval, eliminating the influence of dimensions. Finally, according to the preset order of normalized shear slope, normalized peak unloading force, and normalized total punching work, the three normalized feature values ​​are combined into a 3-dimensional feature vector. Each dimension of the vector corresponds to a wear-sensitive dimension and is stored in the feature vector buffer to ensure that the vector structure is completely consistent with the input layer dimension of the dual-timescale implicit state estimation model.

[0039] In the control method of the aforementioned intelligent servo press, step S4 involves performing dual-time-scale implicit state estimation on the feature vector of the k-th stroke to obtain a quantified wear estimate. It should be understood that, since the feature vector simultaneously contains the mixed effects of material parameter fluctuations (rapidly time-varying, such as batch-to-batch hardness differences) and die wear (slowly time-varying, such as gradual dulling of the cutting edge), a single-time-scale estimation model cannot decouple these two factors, leading to a distorted wear estimate due to material fluctuation interference, and thus failing to accurately quantify the degree of wear. Therefore, this application further employs dual-time-scale implicit state estimation to separately process the fast-changing and slow-changing components in the feature vector, thereby decoupling material interference and focusing on the wear state. This eliminates the influence of batch-to-batch material fluctuations on the wear estimate, obtaining a quantified value that only reflects the actual die wear, providing a reliable decision-making basis for subsequent adaptive compensation strategy generation, avoiding wear misjudgment or overcompensation due to insufficient decoupling, and ensuring product quality stability.

[0040] In particular, in one specific embodiment, Figure 5 This is a flowchart of sub-step S4 of the control method for an intelligent servo press according to an embodiment of this application. Figure 5 As shown, step S4 includes: S41, inputting the feature vector of the k-th stroke, the fast state estimate of the previous stroke, and the stable wear estimate of the previous batch into a fast filter to obtain the fast state estimate of the current stroke and the optimal estimate of the material parameters of the current stroke; S42, inputting the time series of the optimal estimate of the material parameters, the time series of the feature vector of the stroke, and the slow state estimate of the previous batch into a slow filter to obtain the stable wear estimate of the current batch as the quantized wear estimate.

[0041] Specifically, in step S41, the feature vector of the k-th stroke, the fast state estimate of the previous stroke, and the stable wear estimate of the previous batch are input into the fast filter to obtain the fast state estimate of the current stroke and the optimal estimate of the material parameters for the current stroke. It should be understood that material parameters (such as tensile strength and thickness) may change rapidly between adjacent strokes due to batch fluctuations or local inhomogeneities. This change causes fluctuations in the feature vector. If the material parameters are not estimated in real time, these fluctuations will be misjudged as wear. Furthermore, the fast filter needs to iterate based on the previous state, and fixed wear is required to focus on material changes. Therefore, this application further inputs the feature vector, the previous fast state estimate, and the previous batch wear estimate into the fast filter to update the material parameters in real time, decoupling them from wear interference. This allows for accurate capture of material parameter changes in each stroke, eliminating their influence on the feature vector, laying the foundation for the subsequent slow filter to focus on wear estimation, and avoiding compensation bias caused by misjudging wear due to material fluctuations.

[0042] Specifically, in one possible embodiment, step S41 is implemented as follows: First, the fast state estimate (including the mean and covariance of material parameters) of the previous stroke and the stable wear estimate of the previous batch are retrieved and input into an unscented Kalman filter (UKF) along with the current feature vector. Second, the UKF performs a prediction step (predicting the current value based on the previous material parameters) and an update step (correcting the predicted value with the feature vector) based on the premise that the wear estimate is fixed. Finally, the fast state estimate of the current stroke (updated mean and covariance of material parameters) and the best estimate of material parameters are output. This value reflects the material characteristics of the current stroke in real time, ensuring that the subsequent slow wear estimate is not affected by material fluctuations.

[0043] Specifically, in step S42, the time series of the best estimated material parameters, the time series of the stroke feature vector, and the slow state estimate of the previous batch are input into a slow filter to obtain a stable wear estimate for the current batch as a quantified wear estimate. It should be understood that since die wear is a slow, time-varying process at the level of 100 strokes, the single-stroke feature vector cannot reflect the cumulative wear effect, and the material parameter time series still exhibits random fluctuations, requiring smoothing through batch data. The slow filter needs to be updated based on the wear of the previous batch, combining batch material and feature data. Therefore, this application further inputs the batch material parameter sequence, feature vector sequence, and the slow state estimate of the previous batch into the slow filter to smooth fluctuations and accumulate wear information. In this way, random material fluctuations can be eliminated through batch data, accurately capturing the wear accumulation trend and obtaining a stable quantified wear estimate, providing an accurate wear basis for the compensation strategy and avoiding wear estimation deviations caused by single-stroke data fluctuations.

[0044] Specifically, in one possible embodiment, step S42 is implemented as follows: First, it is determined that both the input time series of the best estimates of material parameters and the time series of feature vectors contain data from 100 strokes. Second, the material parameter series is aggregated using the arithmetic mean method. The sum of the 100 best estimates of material parameters is divided by 100 to obtain the batch average material parameters. Similarly, the arithmetic mean is performed on each dimension of the feature vector series to obtain the batch average feature vector. Then, the two average results, along with the slow state estimate of the previous batch (including the posterior mean and covariance of wear), are input into an unscented Kalman filter. The filter first corrects the observation model based on the average data, then performs wear state prediction based on the previous batch state, and finally corrects the predicted value using the batch average feature vector, ultimately outputting the stable wear estimate of the current batch.

[0045] Here, the fundamental technical weakness of the above embodiment in processing batch data lies in its use of a simple arithmetic average method to aggregate multiple sets of results from the fast filter. The inherent defect of this processing method is that it completely ignores the uncertainty information, i.e., the corresponding covariance matrix, of the material parameter estimates output by the fast filter in each stroke.

[0046] Specifically, this method assumes that the material parameter estimates for all strokes within a batch have equal importance and reliability. However, in real-world industrial scenarios, due to factors such as instantaneous sensor noise, minor variations in lubrication conditions, or transient filter responses, the confidence levels of the estimates for different strokes vary significantly. Therefore, the original mechanism fails to fully utilize the complete statistical information contained in each estimate, particularly the close correlation between the neglected material parameter estimates and their inherent uncertainties, leading it to treat high-reliability estimates and low-reliability estimates equally. This information loss directly reduces the accuracy and robustness of batch-level aggregation results, thus providing a suboptimal, potentially biased input to the subsequent slow filter, ultimately affecting the ability to accurately estimate the core implicit state of die wear.

[0047] Preferably, to overcome the above-mentioned defects, a batch aggregation and weighted measurement generation mechanism based on uncertainty is proposed. The core idea of ​​this mechanism is that instead of performing an equal-weighted average of all punching data within a batch, weights are dynamically generated based on the reliability of the material parameter estimates for each stroke, and these weights are used for weighted aggregation, thereby generating more statistically meaningful and robust inputs for the slow-speed filter. Specifically, step S42 includes: performing uncertainty-based batch aggregation and weighted measurement generation on the time series of the best material parameter estimates and the time series of the stroke feature vectors to obtain batch weighted average material parameters, batch weighted average feature vectors, batch material parameter estimation covariance, and batch weighted measurement covariance; predicting the wear state of the slow-speed state estimate of the previous batch to obtain the predicted wear mean and predicted wear covariance; updating the wear state of the predicted wear mean and predicted wear covariance based on the batch weighted average material parameters, batch weighted average feature vectors, and batch weighted measurement covariance to obtain the stable wear estimate of the current batch as the quantized wear estimate.

[0048] The execution of this technique begins with quantifying the reliability of the data for each stroke within a batch. It should be understood that, in this preferred embodiment, an objective and quantifiable confidence level must first be assigned to the estimation result of each stroke within the batch in order to enable subsequent differentiated processing.

[0049] Specifically, the covariance matrix of the material parameters for each stroke k within a batch is estimated. Calculate the reciprocal of its determinant as the basic weight factor. The weights are then normalized to obtain the final weights. .

[0050]

[0051] in, , which is the weighting factor, and is the basic weight of the k-th stroke; Let be the material parameter covariance matrix, which is the covariance matrix of the material parameter estimate after the k-th stroke is updated by the fast filter; The determinant operator represents the determinant of a matrix; The normalized weights are the final weights for the k-th stroke and are used in subsequent weighted calculations. It refers to the batch size, which is the total number of strokes contained within a batch. This is another weighting factor, which is the basic weight of the j-th stroke.

[0052] In technical scenarios, the determinant of the covariance matrix can be viewed as a measure of the uncertainty volume of the estimation result; the larger its reciprocal, the higher the confidence level of the estimation. Thus, this calculation transforms the abstract statistical information (covariance) from the fast filter into an intuitive scalar weight that can be used for weighted processing, providing a solid mathematical foundation for subsequent steps to prioritize the adoption of high-confidence data.

[0053] Based on this, the time series of the best estimates of material parameters and the time series of the stroke eigenvectors are subjected to uncertainty-based batch aggregation and weighted measurement to generate the batch weighted average material parameters, the batch weighted average eigenvectors, the batch material parameter estimation covariance, and the batch weighted measurement covariance. That is, using the generated normalized weights, the mean of the material parameters within the batch is estimated. Instead of simply calculating an average, a weighted average is used to estimate the covariance of the corresponding batch material parameters. It can be estimated using the error propagation rule, and the specific calculation process is as follows:

[0054] in, This is the batch-weighted average material parameter, which is an estimate of the batch-level material parameters after polymerization; Let be the mean of the material parameter estimates, which is the material parameter estimate for the k-th stroke; Estimate the covariance of the material parameters for the k-th stroke; This represents the estimated covariance of batch material parameters.

[0055] This solves the problem of neglecting the original mechanism. The information deficit. It ensures that individual stroke material parameters, which are already more accurately estimated, can play a greater role when synthesizing batch-level material properties. This helps to obtain a more realistic and robust representative value for the entire batch of material properties, thus providing a more solid foundation for the state prediction of slow filters. This provides information on the accuracy of the aggregate estimate itself.

[0056] Similarly, the same normalized weights are used for each stroke. A weighted average is calculated and used as the measurement input for the slow filter. The corresponding batch-weighted measurement covariance. , used to represent The uncertainty can be approximated by the statistical properties of the weighted residuals. The specific calculation process is as follows:

[0057] in, The batch-weighted average feature vector is the aggregated batch-level feature vector, which will be used as the measurement input for the slow filter. Let be the feature vector, which is the feature vector extracted in the k-th stroke; It is the noise covariance of the k-th measurement. This represents the batch-weighted measurement covariance.

[0058] Specifically, The slow filter is used to update the measurements of wear condition. By assigning different weights, it ensures that characteristic data from strokes with more reliable material parameter estimates have a greater influence when constituting the total batch measurement. This allows the slow filter to extract wear information more intelligently, effectively mitigating the negative impact of high-uncertainty stroke data. The estimate provides the confidence interval for this weighted measurement itself, providing the necessary statistical information for updating the slow filter, enabling it to more reasonably balance the predicted values ​​with the new measurements.

[0059] Through the above calculations, when reaching a consensus on the material characteristics of the entire batch, data from strokes with clearer signals and more reliable estimates will be given greater weight. This ensures that the input information used by the slow filter for state prediction and updates has undergone internal quality screening and optimization, effectively suppressing the interference of occasional, low-quality data points within the batch on the overall estimation, and significantly improving the robustness and accuracy of subsequent wear state estimation.

[0060] Finally, the weighted aggregated data is input into a slow, unscented Kalman filter to predict and update the wear state. This is the closed-loop step that ultimately estimates the slow variable (mold wear). Specifically, the wear state is first predicted based on the previous batch of slow state estimates to obtain the predicted wear mean and predicted wear covariance, i.e., based on the previous batch of posterior wear estimates. and Perform condition prediction to obtain the predicted average wear value for the current batch. and its predicted wear covariance The specific calculation process is as follows:

[0061] in, and Let these represent the predicted wear mean and predicted wear covariance for the current batch, respectively. and This represents the predicted wear mean and predicted wear covariance of the previous batch. This represents process noise. In the above formula, it is assumed that wear varies slowly between batches, and its state prediction model is an identity transformation, introducing process noise. To simulate the small and random cumulative effects of wear.

[0062] Then, based on the batch-weighted average material parameters, the batch-weighted average eigenvector, and the batch-weighted measurement covariance, the predicted wear mean and predicted wear covariance are updated to obtain the current batch's stable wear estimate as the quantified wear estimate, i.e., obtained using weighted aggregation. and The Kalman gain is calculated using the UKF standard update procedure. And finally, the post-test mean wear value of the current batch is calculated. and wear-related post-test covariance Among them, the post-test mean of wear is a stable wear estimate for the current batch and can be used as a quantitative wear estimate.

[0063] Specifically, firstly based on the predicted wear average and predicted wear covariance Generate a set of Sigma points Subsequently, each wear Sigma point will be compared with the batch-weighted average material parameters of this batch. Combining phases and using a nonlinear system observation model Propagation is performed to obtain a set of propagated predicted observation Sigma points. The specific calculation process is as follows:

[0064] in, The Sigma point represents the wear condition. For nonlinear system observation models, The Sigma point represents the predicted observation after propagation.

[0065] Based on this propagated Sigma point set, the predicted observation mean can be obtained through weighted calculation. and its covariance and the cross-covariance between state and observation. Next, these statistics are used to calculate the Kalman gain. This gain determines the degree of confidence in the measurement residuals during the update process, and the specific calculation process is as follows:

[0066] in, To predict the observed covariance, The cross-variance between the state and the observation. This represents the Kalman gain for the current batch.

[0067] Finally, using weighted measurements Compared with the predicted observation mean The residuals between the two values ​​are used to correct the prior wear condition, and finally the posterior wear mean of the current batch is calculated. and wear-related post-test covariance This series of steps effectively transforms abstract measurement information into precise quantitative updates of implicit wear conditions.

[0068]

[0069] in, This is the wear posterior mean, which is the wear state estimate of the current batch B after the final update; To predict the average wear, it is a prediction of the wear status of the current batch based on the results of the previous batch; is the Kalman gain, which is a gain matrix used to balance prediction and measurement information; This is the transpose of the gain matrix; To predict the observed mean, it is based on predicted wear. and weighted material parameters The predicted eigenvectors.

[0070] In a technological context, the innovation of this update step lies in all its inputs (especially measurements). and used for observation model prediction All of these already contain prior knowledge about data quality. Thus, by performing Bayesian filtering on data that is more statistically significant and has a higher signal-to-noise ratio, the slow filter can more accurately extract the weaker long-term trend changes caused by mold wear from the observation signal contaminated by material fluctuations, ultimately producing a more accurate and reliable quantitative wear estimate.

[0071] Through the aforementioned technical means, this improved mechanism significantly optimizes the original dual-timescale filter algorithm. Its ultimate technical objective and effect lies in introducing and fully utilizing the uncertainty information from the estimation of material parameters in each stroke, enabling the entire implicit state estimation algorithm to adaptively perceive and process the quality of input data. This not only effectively solves the problems of information loss and poor robustness caused by simple arithmetic averaging, but more importantly, it allows the slow filter to more intelligently distinguish and suppress noise introduced by internal fluctuations in material batches when estimating die wear, thereby significantly improving the estimation accuracy, stability, and convergence speed of die wear—a core slowly varying parameter. Ultimately, this mechanism provides a more reliable and quantifiable wear index for the compensation control system of intelligent servo presses, providing strong data and algorithmic support for achieving more accurate, model-based proactive process compensation, extending die life, and ensuring high consistency of product quality throughout the entire product lifecycle.

[0072] In the control method of the aforementioned intelligent servo press, step S5 involves generating an adaptive process compensation strategy based on the quantified wear estimate using a compensation knowledge base to obtain the compensated target trajectory. It should be understood that because die wear alters the force distribution and material flow characteristics during the stamping process, the original reference slide trajectory and process parameters (based on the new die design) are no longer suitable for the worn die state, easily leading to defects such as wrinkling and dimensional deviations in the product. The compensation knowledge base stores the optimal process adjustment schemes corresponding to different wear levels, serving as the core bridge connecting wear quantification and process optimization. Therefore, this application further combines the quantified wear estimate with the compensation knowledge base to generate an adaptive process compensation strategy and transform it into a compensated target trajectory, thereby ensuring that the slide movement and process parameters accurately adapt to the current die wear state. This proactively offsets the negative impact of wear on the stamping process, ensuring consistent product quality throughout the die's entire lifecycle, while extending the die replacement cycle and reducing production losses caused by unplanned downtime.

[0073] In particular, in one specific embodiment, Figure 6 This is a flowchart of sub-step S5 of the control method for an intelligent servo press according to an embodiment of this application. Figure 6 As shown, step S5 includes: S51, inputting the quantified wear estimate into the compensation knowledge base for model query to obtain the process parameter adjustment vector; S52, compensating the target motion trajectory of the reference slider based on the process parameter adjustment vector to obtain the compensated target trajectory.

[0074] Specifically, in step S51, the quantified wear estimate is input into the compensation knowledge base for model query to obtain a vector of process parameter adjustments. It should be understood that since the quantified wear estimate is an abstract quantitative description of the mold's wear state, such as the increase in fillet radius or surface roughness, it cannot be directly applied to the stamping equipment. It needs to be converted into specific process parameter adjustments, such as changes in speed, blank holder force, and holding time. The compensation knowledge base, through simulation and experimental data, establishes a precise mapping relationship between wear state and process adjustment, avoiding subjective errors from manual experience adjustments. Therefore, this application further inputs the quantified wear estimate into the compensation knowledge base to perform a model query, thereby obtaining a vector of process parameter adjustments that perfectly matches the current wear state. This enables a scientific transformation from wear state to process adjustment, providing specific and executable parameter basis for subsequent target trajectory compensation, ensuring that the compensation strategy conforms to physical laws and meets actual production accuracy requirements.

[0075] Specifically, in one possible embodiment, step S51 is implemented as follows: First, the model type of the compensation knowledge base is determined to be a gradient boosting regression model (the input layer is 3D wear parameters, and the output layer is 4D process adjustment amount). Second, the quantified wear estimate is read and standardized (conforming to the input range during model training, mapped to the [0,1] interval). Then, the standardized wear value is input into the model, and the model calculates the original adjustment amount through the trained decision tree ensemble. Next, the original adjustment amount is destandardized and converted into actual physical units, such as the blank holder force adjustment amount in kN and the speed adjustment amount in mm / s. Finally, according to the preset order of the stretching speed adjustment amount, blank holder force adjustment amount, holding time adjustment amount, and unloading force adjustment amount, the four adjustment amounts are arranged into a 4D process parameter adjustment amount vector and output to the trajectory compensation module. The calculation error of the adjustment amount is strictly controlled within ±0.5% to ensure the accuracy of subsequent trajectory correction.

[0076] Specifically, in step S52, the reference slider target motion trajectory is compensated based on the process parameter adjustment vector to obtain the compensated target trajectory. It should be understood that since the reference slider target motion trajectory is designed for a new mold, its speed, pressure, and other parameters are perfectly matched to the force distribution characteristics of the new mold. After mold wear, the resistance to material shearing or stretching changes (e.g., the cutting edge becomes blunt, leading to increased shearing force). If the reference trajectory continues to be used, problems such as excessive force fluctuations or poor material flow will occur. The process parameter adjustment vector clearly defines the direction and magnitude of the parameters to be corrected and needs to be integrated into the time-position-force relationship of the reference trajectory. Therefore, this application further uses the process parameter adjustment vector to correct the reference trajectory, thereby generating a compensated target trajectory adapted to the current wear state. This allows the slider motion to accurately match the characteristics of the worn mold, ensuring the synergy of force and displacement during stamping, effectively suppressing product defects caused by wear, and avoiding additional loads on the equipment.

[0077] Specifically, in one possible embodiment, step S52 is implemented as follows: First, the reference slider target motion trajectory (including three core curves: position-time, speed-time, and pressure-time) and process parameter adjustment vectors (such as stretching speed -5%, blank holder force +1.5%) are read. Second, the speed-time curve is processed: the reference speed value at each time point of the curve is traversed and multiplied by (1 + speed adjustment) to obtain the corrected speed, ensuring that the speed is reduced by 5% during the stretching stage, while the speed remains unchanged during the idle stroke stage (to avoid affecting the production cycle). Then, the corrected speed curve is integrated to obtain a new position-time curve, ensuring that the slider's final bottom dead center position error is ≤0.01mm. Next, the pressure-time curve is processed: during the blank holder force application stage, the reference pressure value is multiplied by (1 + blank holder force adjustment) to obtain the corrected pressure setting. Finally, the three corrected curves are integrated to form the compensated target trajectory. The trajectory data is processed by a 5-point moving average to ensure smoothness (to avoid shocks caused by sudden speed changes), and output to the servo control module. The trajectory format meets the real-time transmission requirements of the EtherCAT bus.

[0078] In the control method of the aforementioned intelligent servo press, step S6 generates a servo motor torque command based on the compensated target trajectory and the actual position of the slider. It should be understood that the compensated target trajectory is merely an ideal motion reference (including the changes in position and speed over time) that the slider must follow. The actual movement of the slider is affected by factors such as mechanical transmission clearance, stamping load fluctuations (e.g., changes in material resistance), and motor dynamic response delay, making it impossible to spontaneously match the target trajectory. Furthermore, the servo motor needs to drive the slider movement through torque output, requiring the trajectory's motion requirements to be converted into specific torque signals; otherwise, the compensation strategy remains merely theoretical. Therefore, this application further combines the compensated target trajectory with the actual position of the slider, generating a servo motor torque command through closed-loop control logic to drive the motor to adjust its output in real time, ensuring the slider tracks the target trajectory. This transforms the abstract compensation trajectory into a physical execution action, allowing the wear compensation strategy to truly apply to the stamping process, effectively offsetting deviations caused by mold wear, and ensuring the stability of product dimensional accuracy and forming quality.

[0079] In particular, in one specific embodiment, Figure 7 This is a flowchart of sub-step S6 of the control method for an intelligent servo press according to an embodiment of this application. For example... Figure 7 As shown, step S6 includes: S61, based on the compensated target trajectory, performing real-time trajectory tracking error calculation on the actual position and actual speed of the slider to obtain position error and speed error; S62, inputting the position error and speed error into the PID controller to obtain the servo motor torque command.

[0080] Specifically, in step S61, based on the compensated target trajectory, the actual position and actual speed of the slider are calculated in real time to obtain the position error and speed error. It should be understood that during the actual movement of the slider, factors such as elastic deformation of mechanical transmission, instantaneous fluctuations in material stamping resistance, and minor sensor noise can cause its actual position and speed to deviate from the compensated target trajectory. If this deviation (error) is not quantified in real time, the subsequent controller will be unable to determine the direction and magnitude of adjustment, and the closed-loop control will lose its basis. Therefore, this application further calculates the tracking error of the slider's actual position and speed simultaneously based on the compensated target trajectory to accurately quantify the degree of deviation between the actual movement and the ideal trajectory. This provides a clear adjustment basis for the subsequent PID controller, ensuring that the controller can specifically correct the deviation, avoiding trajectory tracking inaccuracies caused by unknown errors, thereby ensuring the effective implementation of the wear compensation strategy and maintaining product forming quality.

[0081] Specifically, in one possible embodiment, step S61 is implemented as follows: First, the servo control system extracts the desired position at the current moment from the compensated target trajectory according to a fixed control cycle, and obtains the desired speed at the corresponding moment by differentiating the position-time curve of the target trajectory. Second, the actual position of the slider (from the servo motor encoder) is read, the actual position data is smoothed and then differentiated to obtain a noise-free actual speed, avoiding high-frequency noise introduced by direct differentiation. Then, the position error (the difference between the desired position and the actual position) is calculated, and the speed error (the difference between the desired speed and the actual speed) is calculated simultaneously. Finally, outlier removal processing is performed on both types of errors (removing extreme errors caused by abnormal sensor pulses), and the results are stored in the error buffer as real-time input to the PID controller, ensuring that the entire error calculation process meets the control cycle requirements.

[0082] Specifically, in step S62, the position error and speed error are input into the PID controller to obtain the servo motor torque command. It should be understood that since the position error and speed error are only quantified deviation values, they cannot directly drive the servo motor—the error needs to be converted into a torque command that conforms to the dynamic characteristics of the motor through a control algorithm. The PID controller has the coordinated control capabilities of proportional (instantaneous correction of deviation), integral (elimination of steady-state error), and derivative (suppression of oscillation overshoot), which can adapt to the dynamic changes in the slider movement, such as error fluctuations caused by sudden changes in stamping load. A single error signal or simple proportional control cannot meet the high-precision tracking requirements. Therefore, this application further inputs the position error and speed error into the PID controller to convert the deviation signal into a precise servo motor torque command. This allows for rapid and stable correction of trajectory tracking errors, ensuring that the actual movement of the slider continuously approaches the compensated target trajectory, ensuring the effective execution of the wear compensation strategy, guaranteeing the force and displacement coordination in the stamping process, and improving product quality consistency.

[0083] Specifically, in one possible embodiment, step S62 is implemented as follows: First, feedback control calculations are performed. The parameters of the PID controller need to be pre-tuned based on the slider inertia and transmission stiffness. For example, the proportional gain can be set to 500, the integral gain to 10, and the derivative gain to 20, with the integral saturation upper limit set to 20% of the total torque. The filtered position error and velocity error are input to the PID controller in real time, and a feedback torque command is calculated based on the above gains to dynamically correct any errors deviating from the target trajectory. Second, a load-slider position mapping model is established based on historical stamping load data collected under the same process. Based on the current position of the compensated target trajectory, a feedforward torque command is queried or calculated from this model to compensate for the main stamping load in advance, reducing the burden on feedback control. Finally, the feedback torque command and the feedforward torque command are added to obtain a composite torque command. This command is limited to ensure it is within the rated torque range of the servo motor. The final torque command after processing is sent to the servo driver via the industrial bus. The driver converts it into motor stator current, thereby driving the slider to move precisely along the compensated trajectory.

[0084] In summary, the control method for the intelligent servo press based on the embodiments of this application is elucidated. It constructs a unified stamping state representation by deeply mining and fusing the inherent force and displacement process data stream of the servo press. Based on this, a dual-timescale implicit state estimation algorithm is introduced to perform time-series modeling and in-depth analysis of the weak and gradual features contained in the full-stroke force-displacement curve, thereby achieving accurate quantitative assessment of the implicit state of die wear. Then, this quantified wear value serves as the core decision factor, driving an adaptive control mechanism based on a compensation knowledge base to dynamically generate the optimal process compensation strategy and slider target trajectory. This improves product quality consistency in the precision stamping process, effectively extends die life, reduces production costs, and comprehensively enhances the intelligence level of stamping production.

[0085] Figure 8 This is a block diagram of the control system of an intelligent servo press according to an embodiment of this application. Figure 8 As shown, the control system 100 of the intelligent servo press according to an embodiment of this application includes: a raw data acquisition module 110, used to receive raw pressure data and raw slider position data from a servo motor encoder and a force sensor; a data preprocessing module 120, used to preprocess the raw pressure data and raw slider position data to obtain the force-displacement curve of the k-th stroke; a wear-sensitive feature extraction module 130, used to extract wear-sensitive features from the force-displacement curve of the k-th stroke to obtain the feature vector of the k-th stroke; an implicit state estimation module 140, used to perform dual-time-scale implicit state estimation on the feature vector of the k-th stroke to obtain a quantized wear estimate; a compensation strategy generation module 150, used to generate an adaptive process compensation strategy based on a compensation knowledge base to obtain a compensated target trajectory; and a control command generation module 160, used to generate a servo motor torque command based on the compensated target trajectory and the actual slider position.

[0086] As described above, the control system 100 of the intelligent servo press according to the embodiments of this application can be implemented in various wireless terminals, such as servers with control algorithms for the intelligent servo press. In one possible implementation, the control system 100 of the intelligent servo press according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the control system 100 of the intelligent servo press can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the control system 100 of the intelligent servo press can also be one of many hardware modules of the wireless terminal.

[0087] Alternatively, in another example, the control system 100 of the intelligent servo press and the wireless terminal can also be separate devices, and the control system 100 of the intelligent servo press can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0088] Here, those skilled in the art will understand that the specific operations of each step in the control system of the aforementioned intelligent servo press have been referenced above. Figures 1 to 7 The control method of the intelligent servo press is described in detail in the description, and therefore, its repeated description will be omitted.

Claims

1. A control method of an intelligent servo press characterized by, The method comprises the following steps: Receiving raw pressure data and raw slider position data from servo motor encoder and force sensor; Preprocessing the raw pressure data and the raw slider position data to obtain a force displacement curve of the kth stroke; Performing wear-sensitive feature extraction on the force displacement curve of the kth stroke to obtain a feature vector of the kth stroke; Performing double-time-scale implicit state estimation on the feature vector of the kth stroke to obtain a quantitative wear estimation value; Generating an adaptive process compensation strategy for the quantitative wear estimation value based on a compensation knowledge base to obtain a compensated target trajectory; Generating a servo motor torque instruction based on the compensated target trajectory and the actual position of the slider.

2. The control method of the intelligent servo press according to claim 1, characterized by, The wear-sensitive feature extraction on the force displacement curve of the kth stroke to obtain a feature vector of the kth stroke comprises the following steps: Performing process stage dynamic segmentation on the force displacement curve of the kth stroke to obtain a shearing stage data segment and a unloading stage data segment; Performing parallel calculation of associated physical features on the shearing stage data segment, the unloading stage data segment and the force displacement curve of the kth stroke to obtain an original shearing slope, an original unloading peak force and an original total blanking work; Vectorizing the original shearing slope, the original unloading peak force and the original total blanking work to obtain the feature vector of the kth stroke.

3. The control method of the intelligent servo press according to claim 2, characterized by, The parallel calculation of associated physical features on the shearing stage data segment, the unloading stage data segment and the force displacement curve of the kth stroke to obtain an original shearing slope, an original unloading peak force and an original total blanking work comprises the following steps: Performing linear regression on the shearing stage data segment based on the least square method to obtain the original shearing slope; Taking the maximum value of the absolute values of all force values in the unloading stage data segment as the original unloading peak force; Applying the trapezoidal rule to the force displacement curve of the kth stroke to perform numerical integration to obtain the original total blanking work.

4. The control method of the intelligent servo press according to claim 1, characterized by, The double-time-scale implicit state estimation on the feature vector of the kth stroke to obtain a quantitative wear estimation value comprises the following steps: Inputting the feature vector of the kth stroke, the fast state estimation of the last stroke and the stable wear estimation value of the last batch into a fast filter to obtain the fast state estimation of the current stroke and the optimal estimation value of the material parameter of the current stroke; Inputting the time sequence of the optimal estimation value of the material parameter, the time sequence of the feature vector of the stroke and the slow state estimation of the last batch into a slow filter to obtain the stable wear estimation value of the current batch as the quantitative wear estimation value.

5. The control method of the intelligent servo press according to claim 4, characterized by, The inputting the time sequence of the optimal estimation value of the material parameter, the time sequence of the feature vector of the stroke and the slow state estimation of the last batch into a slow filter to obtain the stable wear estimation value of the current batch as the quantitative wear estimation value comprises the following steps: Performing batch aggregation and weighted measurement generation based on uncertainty on the time sequence of the optimal estimation value of the material parameter and the time sequence of the feature vector of the stroke to obtain the batch weighted average material parameter, the batch weighted average feature vector, the batch material parameter estimation covariance and the batch weighted measurement covariance; Performing wear state prediction on the slow state estimation of the last batch to obtain the predicted wear mean and the predicted wear covariance; Based on the batch-weighted average material parameters, the batch-weighted average feature vector and the batch-weighted measurement covariance, the wear state is updated to obtain the current batch stable wear estimation value as the quantitative wear estimation value.

6. The control method of the intelligent servo press according to claim 1, wherein Based on the compensation knowledge base, an adaptive process compensation strategy is generated for the quantitative wear estimation value to obtain a compensated target trajectory, including: The quantitative wear estimation value is input into the compensation knowledge base for model query to obtain a process parameter adjustment amount vector; Based on the process parameter adjustment amount vector, the reference slider target motion trajectory is compensated to obtain the compensated target trajectory.

7. The control method of the intelligent servo press according to claim 1, wherein Based on the compensated target trajectory and the actual slider position, a servo motor torque instruction is generated, including: Based on the compensated target trajectory, real-time trajectory tracking error calculation is performed on the actual slider position and the actual slider speed to obtain a position error and a speed error; The position error and the speed error are input into a PID controller to obtain the servo motor torque instruction.

8. A control system for an intelligent servo press characterized by, It includes: An original data acquisition module is configured to receive original pressure data and original slider position data from a servo motor encoder and a force sensor; A data preprocessing module is configured to preprocess the original pressure data and the original slider position data to obtain a force displacement curve of the kth stroke; A wear-sensitive feature extraction module is configured to extract wear-sensitive features from the force displacement curve of the kth stroke to obtain a feature vector of the kth stroke; An implicit state estimation module is configured to perform double-time-scale implicit state estimation on the feature vector of the kth stroke to obtain a quantitative wear estimation value; A compensation strategy generation module is configured to generate an adaptive process compensation strategy for the quantitative wear estimation value based on a compensation knowledge base to obtain a compensated target trajectory; A control instruction generation module is configured to generate a servo motor torque instruction based on the compensated target trajectory and the actual slider position.

Images