A hardware forming defect prediction method and system based on stamping parameters

Abstract

The application discloses a hardware forming defect prediction method and system based on stamping parameters, and particularly relates to the technical field of hardware forming defect prediction, and comprises the following steps: acquiring stamping parameters corresponding to each stroke in a hardware forming process and forming response data corresponding to the stamping parameters, combining the stamping parameters and the forming response data to form a stroke data entry, and writing the stroke data entry into a cache queue of edge calculation; reading a plurality of continuous stroke data entries in the order of strokes from the cache queue of edge calculation, and calculating the difference value of the stamping parameters in adjacent stroke data entries to obtain parameter change data; collecting the stamping parameters and the forming response data in real time in continuous strokes, constructing a separability index reflecting the sensitivity of the material to the parameter change, and judging whether the deformation freedom of the material is contracted according to the continuous downward trend of the separability index, so as to identify the state of the material being forced to enter an irreversible failure path before the appearance of the explicit defect.

Description

Technical Field

[0001] This invention relates to the field of metal forming defect prediction technology, and more specifically, to a method and system for predicting metal forming defects based on stamping parameters. Background Technology

[0002] In existing hardware stamping forming defect prediction technology, the judgment target is generally whether the forming result has failure phenomena such as cracks, wrinkles or corner collapse. The prediction process mainly relies on abnormal changes in stamping parameters, equipment operating status and forming response signals to identify risks in advance. In terms of technical logic, it is assumed that the material always has an adjustable deformation space throughout the forming process, and it is believed that as long as the parameters are still within a reasonable range, the forming result still has the possibility of being corrected. However, in actual stamping production, the plastic flow of material in the mold will be gradually restricted with multiple forming, enhanced local constraints and continuous strain accumulation. When the deformation coordination ability inside the material is weakened to a certain extent, its subsequent deformation path will be fixed by the process structure and historical strain state. Even if no detectable defect features have appeared, the material can no longer be restored to a stable forming state by parameter fine-tuning. Under such conditions, the occurrence of subsequent defects is inevitable rather than accidental. Because existing prediction methods only focus on whether the result of a single forming is close to the failure threshold, they lack the ability to continuously characterize the evolution process of the material's deformation degrees of freedom. They cannot identify whether the material has been forced into an irreversible failure path under normal parameter conditions, resulting in the prediction results failing to provide effective early warning before the defects actually appear. It is evident that current hardware forming defect prediction technology fails to process and judge the forming process from the perspective of the selectivity of material deformation path, and cannot identify in advance that the forming process has inevitably headed towards failure. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for predicting hardware forming defects based on stamping parameters. By collecting stamping parameters and forming response data in real time during continuous stamping cycles and constructing a separability index that reflects the material's sensitivity to parameter changes, the method determines whether the material's deformation degrees of freedom have shrunk based on the continuous downward trend of the separability index, so as to identify the state in which the material is forced into an irreversible failure path before the appearance of obvious defects.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting defects in metal forming based on stamping parameters, comprising: S1. Obtain the stamping parameters and forming response data corresponding to each stroke in the metal forming process, combine the stamping parameters and forming response data to form a stroke data entry, and write it into the cache queue of edge computing. S2. Read multiple consecutive stroke data entries from the edge computing cache queue in stroke order, and calculate the difference in stamping parameters in adjacent stroke data entries to obtain parameter change data. S3. Based on the parameter change data, the stroke data entries are grouped according to the magnitude of parameter change. The stroke data entries with parameter change ranges are divided into at least two groups of stroke data subsets with different parameter change conditions. Response feature extraction is performed on the forming response data in each stroke data subset to obtain the corresponding response feature set. S4. Perform separability calculation on the response feature sets corresponding to different stroke data subsets. By calculating the degree of difference in response features between different stroke data subsets and the degree of dispersion of response features within each stroke data subset, a separability index is formed. S5. Perform trend analysis on the separability index over multiple consecutive calculation cycles. When the separability index shows a continuous decline over multiple consecutive calculation cycles and reaches the preset judgment condition, it is determined that the material's response to changes in stamping parameters has shrunk during the current metal forming process, and the corresponding stamping data entry is marked as the metal forming defect prediction result output.

[0005] In a preferred embodiment, in S1, the process of writing to the edge computing cache queue includes: S1-1. Through the signal acquisition interface between the edge computing side and the stamping equipment, the stamping parameters corresponding to each stamping stroke are acquired, and the sampling time information of each stamping parameter is recorded during acquisition. The stamping parameters and the corresponding sampling time information are used as the original sampling data of the stamping parameters. S1-2. Within the preset punch time window, continuously sample the load measurement signal and slide position measurement signal corresponding to the punching parameters. Organize the continuously sampled load measurement signal into a load curve sampling sequence in time order, and organize the continuously sampled slide position measurement signal into a slide position curve sampling sequence in time order. Record the sampling time information corresponding to each sampling point as the original sampling data of the forming response. S1-3. Based on the stroke time window corresponding to the current stroke, associate the original sampling data of stamping parameters and the original sampling data of forming response that fall within the same stroke time window range with the same stroke, and generate the stroke identifier corresponding to the current stroke. S1-4. Using the stroke identifier as an index, the original sampling data of stamping parameters and the original sampling data of forming response that are associated with the same stroke identifier within the corresponding stroke time window are collected according to the stroke identifier correspondence to form a stamping parameter data set and a forming response data set. S1-5. Using the stroke time window as the time range, based on the sampling time information of each sampling point, a unified time series covering the stroke time window is generated sequentially at preset time intervals, starting from the beginning time of the stroke time window; interpolation is performed on the sampling points on the unified time series so that the stamping parameter data and forming response data have a one-to-one corresponding sampling time index on the time axis. S1-6. Encapsulate the aligned stamping parameter data and forming response data according to a unified data structure to generate stamping data entries; write the stamping data entries into the pre-allocated cache area in the working memory of the edge computing node in the order of sampling time information, and insert the stamping data entries into the cache queue in the order of writing.

[0006] In a preferred embodiment, in S2, the process of obtaining parameter change data includes: S2-1. In the edge computing node, based on the cache queue cached in the working memory, the order in which the impulse data entries are enqueued in the cache queue is used as the impulse order. Starting from the head of the queue, multiple consecutive impulse data entries are read sequentially, and each read impulse data entry is temporarily stored in the sequential buffer. S2-2. Based on the stroke identifier field carried by each stroke data entry in the sequential buffer, perform sequential continuity check on the stroke identifier field, determine whether the stroke identifiers of two adjacent stroke data entries are arranged continuously in an increasing pattern, and remove stroke data entries with discontinuous stroke identifiers. S2-3. Based on the stamping parameter field and sampling time index field carried inside the stamping data entries in the sequential buffer, extract the aligned stamping parameter data from each stamping data entry, and write the stamping parameter data into the stamping parameter order list according to the stamping order.

[0007] In a preferred embodiment, in S2, the process of obtaining parameter change data further includes: S2-4. Take any two adjacent stamping parameter data in the stamping parameter sequence list as a group. Based on the sampling time index field, check each sampling time index position of the two adjacent stamping parameter data in the stamping parameter sequence list in the unified time series one by one to determine whether the two stamping parameter data have corresponding values ​​at each sampling time index position. The alignment is deemed successful only if all sampling time index positions within the unified time series satisfy the condition that two stamping parameter data have corresponding values ​​at the same time; otherwise, the alignment is deemed unsuccessful, and the adjacent stamping times corresponding to the two stamping parameter data are removed from the current difference calculation object. S2-5. Under the premise that the alignment judgment is passed, take the two stamping parameter data of adjacent strokes as objects, perform difference calculation for each sampling time index position, subtract the stamping parameter value of the sampling time index position of the previous stroke from the stamping parameter value of the sampling time index position of the next stroke, and form a list of stamping parameter difference entries that correspond one-to-one with the sampling time index. S2-6. Organize all stamping parameter difference entries corresponding to the same adjacent strokes into a parameter difference sequence according to the sampling time index, and write each parameter difference sequence into the parameter difference sequence list according to the stroke order as parameter change data.

[0008] In a preferred embodiment, in S3, the process of obtaining the corresponding set of response features includes: S3-1. Based on the parameter change data and the corresponding stroke identifier, associate the parameter difference sequence of each stroke with the corresponding stroke identifier to form a list of parameter change records arranged in stroke order. S3-2. For each record in the parameter change record list, in the parameter difference sequence corresponding to the current record, perform absolute value operation on the stamping parameter difference at each sampling time index position, and select the upper limit value of the absolute value in each parameter difference sequence as the parameter change amplitude value of the corresponding stroke, and write the parameter change amplitude value and the corresponding stroke identifier into the parameter change amplitude mapping table. S3-3. Preset multiple non-overlapping parameter variation ranges. Each parameter variation range is determined by a pair of lower and upper limits of parameter variation. Based on the parameter variation range values ​​in the parameter variation range mapping table, compare the parameter variation range value corresponding to each stroke with the multiple parameter variation range ranges one by one to determine the parameter variation range range to which the current parameter variation range value falls. Group the strokes that fall into the same parameter variation range range into the same group to form strokes with different parameter variation range ranges.

[0009] In a preferred embodiment, in S3, the process of obtaining the corresponding response feature set further includes: S3-4. Using the stroke identifier group as an index, read the corresponding stroke data entries from the cache queue of the edge computing node according to the stroke identifier, and group the stroke data entries belonging to the same stroke identifier group to form the corresponding stroke data subset. S3-5. For each stroke data subset, extract the aligned forming response data from each stroke data entry in the corresponding stroke data subset, and organize the extracted forming response data into a response data matrix according to the stroke order and sampling time index order. The row index of the response data matrix corresponds to the stroke identifier, and the column index corresponds to the sampling time index position on the unified time series. S3-6. Based on the response data matrix corresponding to each stroke data subset, perform response feature extraction processing on each row of forming response data to obtain multiple response feature quantities. The multiple response feature quantities include the peak value, peak occurrence time, rise phase slope, and curve integral value within a preset time window of the forming response calculated on a unified time series. Combine the multiple response feature quantities into the response feature vector corresponding to the current stroke, and combine the response feature vectors corresponding to each stroke in the same stroke data subset in stroke order to form the response feature set of the corresponding stroke data subset. The peak value calculation process for the forming response is as follows: traverse the forming response data of the current stroke at all sampling time index positions corresponding to the unified time series, compare the forming response values ​​at each sampling time index position, select the sampling point at the upper limit value, and take the forming response value corresponding to the sampling point as the peak value of the forming response of the current stroke. The peak occurrence time is: after determining the sampling time index position corresponding to the peak of the forming response, read the time value in the unified time series corresponding to the sampling time index position, and take the corresponding time value as the peak occurrence time of the forming response. The slope of the rising phase is as follows: In the forming response data of the current stroke, the starting sampling time index position on the unified time series is taken as the starting point of the rising phase, and the sampling time index position corresponding to the peak value of the forming response is taken as the ending point of the rising phase. The forming response values ​​and corresponding time values ​​at the starting point and the ending point are read respectively, the ratio of the increment of the forming response value to the increment of time is calculated, and the corresponding ratio is taken as the slope of the rising phase of the current stroke. The integral value of the curve within the preset time window is as follows: Select each sampling time index position between the start time and the end time corresponding to the preset time window on the unified time series, obtain the forming response value of each sampling time index position according to the sampling time index order, use the time difference between adjacent sampling time index positions as the integration step size using a preset time interval, perform trapezoidal approximate area calculation on the forming response value corresponding to each adjacent sampling time index position and accumulate it segment by segment within the preset time window, and use the sum of the accumulated areas as the integral value of the forming response curve of the current stroke within the preset time window.

[0010] In a preferred embodiment, in S4, the process of forming the separability index includes: S4-1. Taking the response feature vectors arranged in order of impulse in the response feature set as objects, write each response feature vector into the response feature vector order list; S4-2. Taking any two different subsets of impulse data as objects, based on the corresponding response feature vector order list, and using the response feature quantities at the same index position as comparison objects, calculate the absolute value of the numerical difference between the elements of each response feature vector, and write the absolute values ​​of the difference at all index positions into the cross-subset difference sequence in order. S4-3. Select the absolute value of the difference that is at the upper limit position in the cross-subset difference sequence, and use the corresponding absolute value of the difference as the upper limit difference degree between the data subsets of the current two pulses. S4-4. Taking any subset of impulse data as the object, in the order list of response feature vectors corresponding to the corresponding subset of impulse data, take the response feature quantities at the same index position as the comparison object, calculate the absolute value of the numerical difference between the elements of different response feature vectors, and write the absolute value of the difference corresponding to each index position into the difference sequence inside the subset in order. S4-5. Select the absolute value of the difference that is at the upper limit in the difference sequence within the subset, and use the corresponding absolute value of the difference as the upper limit of the dispersion within the data subset of the current impulse. S4-6. Using the upper limit difference value across subsets as the minuend and the upper limit of the upper limit dispersion value within all impulse data subsets as the subtrahend, subtract the values ​​to obtain the difference, and perform an absolute value operation on the difference. Use the resulting value as the separability index.

[0011] In a preferred embodiment, in S5, the process of outputting the hardware forming defect prediction result includes: S5-1. Based on the calculation results of the separability index, the separability indices generated sequentially within the continuous calculation cycle will be written into the separability index sequence in the order of the calculation cycle. S5-2. Take any two separable indices in the separable index sequence that are at adjacent sequence index positions as objects, perform numerical difference operation, subtract the value of the previous separable index from the value of the latter separable index, obtain the numerical difference of the separable indices in adjacent calculation periods, and write the obtained numerical difference into the separable index difference sequence. S5-3. In the sequence of differences of separability index, perform a judgment on all numerical differences, judge the numerical difference less than zero as a decreasing difference, judge the numerical difference greater than zero as an increasing difference, and judge the numerical difference equal to zero as a level difference, and write each judgment result into the change judgment sequence. S5-4. In the change determination sequence, according to the sequence index order, perform continuity determination on the determination results of adjacent index positions. If the determination results of the current index position and the next index position are both decreasing differences, then the current index position to the next index position forms a continuous decreasing segment, and continue to recursively determine whether the next index position and its next index position satisfy the same decreasing state. Connect multiple continuous decreasing segments that continuously satisfy the decreasing state of adjacent index positions to form a decreasing segment. When the determination result of any index position in the change determination sequence is not a decreasing difference, terminate the construction of the current decreasing segment, and write the completed decreasing segment into the decreasing segment set. S5-5. Compare the sequence segment length of each descent segment in the descent segment set. Descending segments whose sequence segment length reaches or exceeds the preset descent judgment threshold are regarded as descent segments that meet the descent duration condition, and the corresponding separability index sequence segment is defined as the separable index segment that meets the descent duration condition. S5-6. Based on the index position of the separability index sequence corresponding to the separability index segment that meets the condition of continuous decline, locate the stroke identifier of the corresponding calculation cycle, determine that the material's response to changes in stamping parameters has shrunk during the corresponding stroke, and mark the corresponding stroke data entry as the hardware forming defect prediction result.

[0012] A hardware forming defect prediction system based on stamping parameters includes a stamping count acquisition module, a change calculation module, a feature grouping module, a separable measurement module, and a shrinkage early warning module. The stamping acquisition module is used to acquire the stamping parameters and forming response data corresponding to each stamping in the metal forming process. The stamping parameters and forming response data are combined to form a stamping data entry and written into the cache queue of edge computing. The change calculation module is used to read multiple consecutive stroke data entries from the edge computing cache queue in stroke order, and calculate the difference of the stamping parameters in adjacent stroke data entries to obtain parameter change data that characterizes the change of stamping parameters between consecutive strokes. Based on parameter change data, the feature grouping module performs grouping processing on the stroke data entries according to the magnitude of parameter change. The stroke data entries with parameter change magnitude in different ranges are divided into different stroke data subsets, and response feature extraction is performed on the forming response data in each stroke data subset to obtain the corresponding response feature set. The separability measurement module is used to perform separability calculation on the response feature set corresponding to different stroke data subsets. By calculating the degree of difference of response features between different stroke data subsets and the degree of dispersion of response features within each stroke data subset, a separability index is formed. The shrinkage early warning module is used to perform trend analysis on the separability index over multiple consecutive calculation cycles. When the separability index shows a continuous decline over multiple consecutive calculation cycles and reaches the preset judgment condition, it is determined that the material's response to changes in stamping parameters has shrunk during the current metal forming process, and the corresponding stamping data entry is marked as the metal forming defect prediction result output.

[0013] The technical effects and advantages of this invention are as follows: 1. This scheme continuously calculates the cross-group differences and intra-group dispersion of forming response in the punch data subset, and characterizes the material’s sensitive shrinkage process to parameter changes as a continuous downward trend of the separability index. This enables the identification of the implicit stage of material deformation degree of freedom loss, avoids relying solely on explicit defect signals, and allows irreversible failure paths to be warned before defects appear, thus improving the predictability and reliability. 2. Perform time indexing, data alignment, and difference construction on stamping parameters and forming response data, and group them according to the parameter change range to make the response differences under different stamping conditions comparable, avoid the distortion of response characteristics due to asynchronous sampling or mixed changing conditions, and ensure the physical consistency of the data on which trend determination depends. 3. Extract multi-dimensional response features such as peak value, peak occurrence time, rise slope, and curve integral, and perform positional comparison on a unified time series. This transforms the forming response from a single-point signal into a behavioral feature trajectory, which can capture the response structure changes caused by internal hardening and constraint enhancement of the material, and provide a more physically meaningful index basis for forming. 4. By incorporating cross-subset response differences and discrete characteristics within subsets into the measurement system, the material's response capability changes in a comprehensive manner are reflected. This enables the separability index to determine whether the material can still respond effectively to parameter adjustments, fundamentally replacing the traditional failure judgment logic that relies on absolute thresholds or abnormal amplitudes, and improving the depth of pattern recognition. 5. Constructing a separability index sequence and making a persistence judgment based on the length of continuous decreasing segments can eliminate random noise interference and transient signal anomalies caused by parameter fluctuations, realize the detection of the true normal response capability decay process, ensure the stability and interpretability of prediction results, and avoid increasing the false alarm rate. Attached Figure Description

[0014] Figure 1 This is a flowchart outlining the method steps of the present invention; Figure 2 This is a schematic diagram of the system module structure of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Refer to the instruction manual appendix Figure 1-2 An embodiment of the present invention provides a method for predicting hardware forming defects based on stamping parameters, comprising: S1. Obtain the stamping parameters and forming response data corresponding to each stroke in the metal forming process, combine the stamping parameters and forming response data to form a stroke data entry, and write it into the cache queue of edge computing. S2. Read multiple consecutive stroke data entries from the edge computing cache queue in stroke order, and calculate the difference of the stamping parameters in adjacent stroke data entries to obtain parameter change data characterizing the change of stamping parameters between consecutive strokes. S3. Based on the parameter change data, the stroke data entries are grouped according to the magnitude of parameter change. The stroke data entries with parameter change ranges are divided into at least two groups of stroke data subsets with different parameter change conditions. Response feature extraction is performed on the forming response data in each stroke data subset to obtain the corresponding response feature set. S4. Perform separability calculation on the response feature set corresponding to different stroke data subsets. By calculating the degree of difference in response features between different stroke data subsets and the degree of dispersion of response features within each stroke data subset, a separability index is formed to characterize the degree of distinguishability of forming response under different parameter changes. S5. Perform trend analysis on the separability index over multiple consecutive calculation cycles. When the separability index shows a continuous decline over multiple consecutive calculation cycles and reaches the preset judgment condition, it is determined that the material's response to changes in stamping parameters has shrunk during the current metal forming process. The corresponding stamping data entry is marked as the metal forming defect prediction result output, which is used to indicate the risk of forming failure in the stamping process before obvious defects appear.

[0017] In S1, the process of writing to the edge computing cache queue includes: S1-1. Through the signal acquisition interface between the edge computing side and the stamping equipment, the stamping parameters corresponding to each stroke are acquired, and the sampling time information of each stamping parameter is recorded during acquisition. The stamping parameters and the corresponding sampling time information are used as the raw sampling data of the stamping parameters. The stamping parameters include the slider position, slider speed, forming load, feeding step distance, and feeding speed. The signal acquisition interface between the edge computing side and the stamping equipment refers to the communication interface deployed between the edge computing node and the stamping equipment for receiving the output signals of the stamping equipment and transmitting them to the edge computing node. This communication interface can receive the electrical or digital signals output by the sensors of the stamping equipment during the stroke, and trigger sampling and sampling time recording inside the edge computing node to write the acquired stamping parameters and the corresponding sampling time information into the working memory of the edge computing node. S1-2. Within the preset stroke time window, continuously sample the load measurement signal and slide position measurement signal corresponding to the stamping parameters. Organize the continuously sampled load measurement signal into a load curve sampling sequence in chronological order, and organize the continuously sampled slide position measurement signal into a slide position curve sampling sequence in chronological order. Record the sampling time information corresponding to each sampling point as the original sampling data of the forming response. The load measurement signal and slide position measurement signal refer to the continuously measured electrical signals output by the load sensor and slide position sensor respectively during the current stroke and acquired through the edge computing side signal acquisition interface. S1-3. Based on the time window of the current stroke, perform a judgment on the sampling time information in the original sampling data of the stamping parameters and the original sampling data of the forming response. Associate the original sampling data of the stamping parameters and the original sampling data of the forming response whose sampling time falls within the same time window of the stroke with the same stroke, and generate the stroke identifier corresponding to the current stroke. S1-4. Using the stroke identifier as an index, the original sampling data of stamping parameters and the original sampling data of forming response that are associated with the same stroke identifier within the corresponding stroke time window are collected according to the stroke identifier correspondence, forming a set of stamping parameter data and a set of forming response data to represent the content collected in the same stroke. S1-5. Using the stroke time window as the time range, based on the sampling time information of each sampling point, a unified time series covering the stroke time window is generated sequentially at preset time intervals, starting from the beginning time of the stroke time window. Interpolation is performed on the sampling points in the unified time series so that the stamping parameter data and forming response data have a one-to-one corresponding sampling time index on the time axis. Interpolation refers to finding sampling points in the corresponding data set with the same sampling time or within the preset time deviation range for each time point in the unified time series, and using the value of the sampling point as the alignment value of that time point. If no sampling point meets the conditions, two sampling points with sampling times before and after the time point and adjacent to each other are selected, and the interpolation value of that time point is calculated according to the linear interpolation rule, which is used as the alignment value of that time point. S1-6. Encapsulate the aligned stamping parameter data and the aligned forming response data according to a unified data structure to generate stamping data entries; write the stamping data entries into the pre-allocated cache area in the working memory of the edge computing node in the order of sampling time information, use the cache area as the data storage space of the cache queue, and insert the stamping data entries into the cache queue in the order of writing. It should be noted that in this solution, stamping parameters refer to the parameter data output by the stamping equipment during the metal forming process and acquired through the edge computing side signal acquisition interface to characterize the execution state of the stamping process. These parameters include slider position, slider speed, forming load, feeding step distance, and feeding speed. The corresponding data structure carries sampling time information recorded by the local clock on the edge side, which is used to arrange and position the sampling time sequence within the stamping time window, serving as the basis for subsequent parameter change analysis and time dimension alignment processing. Forming response data refers to the continuous measurement electrical signals output by the load sensor and slider position sensor configured in the stamping equipment during the current stroke, synchronized with the stamping process and acquired through the edge computing side signal acquisition interface. After sampling, these continuous measurement electrical signals are organized into a load curve sampling sequence or a slider position curve sampling sequence according to the sampling time sequence. During the sampling process, the corresponding sampling time information is generated by the edge side local clock to record the sampling time and establish the time index relationship of the curve sampling points, so that the interpolation position can be determined based on the sampling time index when performing unified time axis generation and interpolation completion in the future. A cache queue is a cache structure deployed in the working memory of edge computing nodes and managed by a first-in, first-out (FIFO) rule. It uses the stroke identifier as the key to index the writing order of stroke data entries and maintains the consistency of stroke execution order. It is used to support the real-time requirements of subsequent continuous stroke difference calculation and shaping response analysis, and to ensure that memory capacity is controllable and computation latency is predictable during continuous data writing.

[0018] In S2, the process of obtaining parameter change data includes: S2-1. In the edge computing node, based on the cache queue cached in the working memory, the order in which the impulse data entries are enqueued in the cache queue is used as the impulse order. Starting from the head of the queue, multiple consecutive impulse data entries are read sequentially, and each read impulse data entry is temporarily stored in the sequence buffer. The sequence buffer is the edge computing working memory area used to temporarily store multiple impulse data entries read from the cache queue in impulse order. This area logically maintains the reading order of the impulse data entries and provides an order guarantee for subsequent impulse identifier verification and data extraction. S2-2. Based on the stroke identifier field carried by each stroke data entry in the sequential buffer, perform a sequential continuity check on the stroke identifier field to determine whether the stroke identifiers of two adjacent stroke data entries are arranged continuously in an increasing order. Stroke data entries with discontinuous stroke identifiers are removed to ensure that the stroke data entries retained in the sequential buffer maintain continuity in stroke order. The stroke identifier field is a field embedded in the stroke data entry to uniquely identify the current stroke. It is generated by the edge computing side based on the stroke time window when constructing the stroke data entry. It is used to indicate the stamping process number to which the stroke data entry belongs and to perform the stroke order continuity check. The continuous increasing order means that the stroke identifier values ​​of two adjacent stroke data entries in the sequential buffer satisfy the condition that the value of the latter stroke identifier is greater than the value of the former stroke identifier, and the difference does not exceed the preset allowable range of stroke interval. S2-3. After removing stroke data entries with discontinuous stroke identifiers, based on the stamping parameter field and sampling time index field carried within the stroke data entries in the sequential buffer, the aligned stamping parameter data is extracted from each stroke data entry, and the stamping parameter data is written into the stamping parameter sequence list according to the stroke order, so that the stamping parameter sequence list is consistent with the stroke order in terms of entry order; where the stamping parameter field refers to the field encapsulated during the construction of the stroke data entry to store the aligned stamping parameter data corresponding to that stroke, and this field carries a sequence of stamping parameter values ​​arranged by the sampling time index; the sampling time index field refers to the sampling time index value that corresponds one-to-one with the aligned stamping parameter data, which is used to reflect the position of the stamping parameter sampling point on a unified time series, so that the corresponding sampling positions of adjacent strokes can be aligned point by point according to the index.

[0019] In S2, the process of obtaining parameter change data also includes: S2-4. Take any two adjacent stamping parameter data in the stamping parameter sequence list as a group. Based on the sampling time index field, check each sampling time index position of the two adjacent stamping parameter data in the stamping parameter sequence list in the unified time series one by one to determine whether the two stamping parameter data have corresponding values ​​at each sampling time index position. The alignment is considered successful only when all sampling time index positions within the unified time series simultaneously satisfy the condition that two stamping parameter data points have corresponding values. The two corresponding stamping parameter data points are then used as the objects for subsequent difference calculations. Otherwise, the alignment is considered unsuccessful, and the adjacent stamping times corresponding to the two stamping parameter data points are removed from the objects of this difference calculation. This avoids performing difference calculations under conditions where the sampling time correspondence is incomplete. The sampling time index position refers to the position number used in the unified time series to identify the order of each sampling time point on the time axis. Each sampling time index position corresponds one-to-one with a time point in the unified time series. It is used to determine the alignment position of the stamping parameter data on the unified time axis during difference calculation and alignment determination, enabling point-by-point comparison between the same sampling time index positions corresponding to different stamping times. S2-5. Under the premise that the alignment judgment is passed, take the two stamping parameter data of adjacent strokes as objects, perform difference calculation for each sampling time index position, subtract the stamping parameter value of the sampling time index position of the previous stroke from the stamping parameter value of the sampling time index position of the next stroke, and form a list of stamping parameter difference entries that correspond one-to-one with the sampling time index. S2-6. Organize all stamping parameter difference entries corresponding to the same adjacent strokes into a parameter difference sequence according to the sampling time index, and write each parameter difference sequence into the parameter difference sequence list according to the stroke order, as parameter change data, which is used to reflect the change of stamping parameters in the time dimension of consecutive strokes and support the subsequent execution of stroke data entry grouping processing based on parameter change magnitude.

[0020] In S3, the process of obtaining the corresponding response feature set includes: S3-1. Based on the parameter change data and the corresponding stroke identifier, associate the parameter difference sequence of each stroke with the corresponding stroke identifier to form a list of parameter change records arranged in stroke order. S3-2. For each record in the parameter change record list, in the parameter difference sequence corresponding to the current record, perform absolute value operation on the stamping parameter difference at each sampling time index position, and select the upper limit value of the absolute value in each parameter difference sequence as the parameter change amplitude value of the corresponding stroke. Write the parameter change amplitude value and the corresponding stroke identifier into the parameter change amplitude mapping table to represent the stamping parameter change intensity between consecutive strokes. The absolute value operation means that for the stamping parameter difference corresponding to each sampling time index position in the parameter difference sequence, if the difference is positive, its original value is retained; if the difference is negative, its opposite is taken as the absolute value value of the current sampling position; if the difference is zero, it is kept as zero, so as to obtain a stamping parameter difference absolute value sequence that only represents the magnitude of the difference and does not distinguish the direction of change. S3-3. Pre-define multiple non-overlapping parameter variation range intervals. Each parameter variation range interval is determined by a pair of lower and upper limits of parameter variation range. Based on the parameter variation range values ​​in the parameter variation range mapping table, compare the parameter variation range value corresponding to each stroke with the multiple parameter variation range intervals one by one to determine the parameter variation range interval into which the current parameter variation range value falls. Group the strokes that fall into the same parameter variation range interval into the same group, forming at least two stroke identifier groups with different parameter variation range intervals. The lower and upper limits of parameter variation range are pre-set numerical boundaries based on the parameter adjustment range allowed by the stamping process and the parameter variation range distribution characteristics statistically obtained from historical production data. The lower and upper limits of parameter variation range for each parameter variation range interval are arranged in ascending order of numerical value. There is no overlap between the upper and lower limits of parameter variation range for two adjacent parameter variation range intervals, so that the parameter variation range value achieves segmented coverage across multiple parameter variation range intervals.

[0021] In S3, the process of obtaining the corresponding response feature set also includes: S3-4. Using the stroke identifier group as an index, read the corresponding stroke data entries from the cache queue of the edge computing node according to the stroke identifier. Group the stroke data entries belonging to the same stroke identifier group to form corresponding stroke data subsets, so that each stroke data subset corresponds to a different parameter change range, and serves as a stroke data subset with different parameter change conditions. S3-5. For each stroke data subset, extract the aligned forming response data from each stroke data entry within the corresponding stroke data subset. Organize the extracted forming response data into a response data matrix according to the stroke order and sampling time index order. The row index of the response data matrix corresponds to the stroke identifier, and the column index corresponds to the sampling time index position on the unified time series. The forming response field is the field used to store the aligned forming response data corresponding to the stroke when constructing the stroke data entry. This field stores the values ​​of the forming response curve at each sampling time index position in the order of the sampling time index of the unified time series. It is used to read the forming response data according to the stroke identifier and sampling time index position when constructing the response data matrix. S3-6. Based on the response data matrix corresponding to each stroke data subset, perform response feature extraction processing on each row of forming response data to obtain multiple response feature quantities. The multiple response feature quantities include the peak value, peak occurrence time, rising phase slope, and curve integral value within a preset time window of the forming response calculated on a unified time series. Combine the multiple response feature quantities into the response feature vector corresponding to the current stroke, and combine the response feature vectors corresponding to each stroke in the same stroke data subset in stroke order to form the response feature set of the corresponding stroke data subset, which serves as the input data for subsequent separability calculation. The peak value calculation process for the forming response is as follows: traverse the forming response data of the current stroke at all sampling time index positions corresponding to the unified time series, compare the forming response values ​​at each sampling time index position, select the sampling point at the upper limit value, and take the forming response value corresponding to the sampling point as the peak value of the forming response of the current stroke; where the forming response value refers to a single scalar value corresponding to a certain sampling time index position of the forming response data under the unified time series. This scalar value is the aligned forming response data read from the forming response field at that sampling time, used to represent the load magnitude and slider position state in the metal forming process at that sampling time, and is the basic data for participating in peak value calculation, rising phase slope calculation, and curve integral calculation on the unified time series; The peak occurrence time is: after determining the sampling time index position corresponding to the peak of the forming response, the time value in the unified time series corresponding to the sampling time index position is read, and the corresponding time value is used as the peak occurrence time of the forming response, which is used to characterize the moment when the peak occurs in the forming process; The rising phase slope is as follows: In the forming response data of the current stroke, the starting sampling time index position on the unified time series is taken as the starting point of the rising phase, and the sampling time index position corresponding to the peak value of the forming response is taken as the ending point of the rising phase. The forming response values ​​and corresponding time values ​​at the starting point and the ending point are read respectively, the ratio of the increment of the forming response value to the increment of time is calculated, and the corresponding ratio is taken as the rising phase slope of the current stroke, which is used to characterize the average rate of change of the forming response from the initial state to the peak value. The integral value of the curve within the preset time window is as follows: Select each sampling time index position between the start and end times corresponding to the preset time window on a unified time series. Obtain the forming response value of each sampling time index position according to the sampling time index order. Use a preset time interval to take the time difference between adjacent sampling time index positions as the integration step size. Perform trapezoidal approximate area calculation on the forming response value corresponding to each adjacent sampling time index position and accumulate it segment by segment within the preset time window. Use the accumulated area sum as the integral value of the forming response curve of the current stroke within the preset time window, which is used to characterize the overall energy or cumulative change of the forming response within the time window. The trapezoidal approximate area calculation refers to reading the forming response value and the corresponding time value of two adjacent sampling time index positions within the preset time window, respectively. Multiply the arithmetic mean of the two forming response values ​​by the time difference between the two sampling time index positions. Use the product as the approximate value of the area of ​​the forming response curve between the adjacent sampling time index positions. Accumulate the approximate area values ​​corresponding to all adjacent sampling time index positions within the preset time window segment by segment to obtain the integral value of the forming response curve within the preset time window.

[0022] In S4, the process of forming the separability index includes: S4-1. Taking the response feature vectors arranged in the order of impulses in the response feature set as objects, write each response feature vector into the response feature vector order list, so that the order of the response feature vector order list corresponds one-to-one with the order of impulses in the order of the entries. S4-2. Taking any two different subsets of stroke data as objects, based on the corresponding response feature vector order list, and comparing response feature quantities at the same index position, calculate the absolute value of the numerical difference between each element of the response feature vector. Write the absolute values ​​of the differences at all index positions into a cross-subset difference sequence in order, as the difference degree data characterizing the degree of difference in response features between different stroke data subsets. Here, each element of the response feature vector refers to the numerical value of a specific response feature quantity extracted from the forming response data of the same stroke corresponding to a unified time series. This value represents the forming response behavior characteristics of that stroke at a specific sampling time index position or time interval. Each response feature vector element is arranged in a uniform time series order. The absolute value of the numerical difference between each response feature vector element is calculated as follows: for response feature vector elements at the same index position in any two different impulse data subsets, the numerical values ​​are subtracted to obtain the difference. If the difference is positive, the original value is retained; if the difference is negative, its opposite is taken as the absolute value of the difference; if the difference is zero, it is taken as zero. The obtained absolute value of the difference is used as the response feature difference value at the corresponding index position. The same index position refers to the position number arranged in a uniform time series order within the response feature vector, so that the response feature vectors of different impulses are consistent in index number, thereby realizing one-to-one comparison between corresponding sampling positions. S4-3. Select the absolute value of the difference at the upper limit position in the cross-subset difference sequence, and use the corresponding absolute value of the difference as the upper limit difference degree between the current two impulse data subsets to characterize the distinguishability of response features between different impulse data subsets. S4-4. Taking any subset of impulse data as the object, in the response feature vector order list corresponding to the corresponding impulse data subset, using the response feature quantities at the same index position as the comparison object, calculate the absolute value of the numerical difference between different response feature vector elements, and write the absolute value of the difference corresponding to each index position into the difference sequence within the subset in order; wherein calculating the absolute value of the numerical difference between different response feature vector elements is specifically as follows: within the same impulse data subset, for the response feature vector elements at the same index position in the response feature vector order list, perform numerical subtraction to obtain the difference, and perform absolute value operation on the difference, and use the obtained absolute value of the difference as the response feature dispersion value at that index position; S4-5. Select the absolute value of the difference that is at the upper limit in the difference sequence within the subset, and use the corresponding absolute value of the difference as the upper limit of the dispersion within the current impulse data subset to characterize the consistency of the response feature changes under the same parameter change conditions. S4-6. Using the upper limit difference value across subsets as the minuend and the upper limit of the upper limit dispersion value within all stroke data subsets as the subtrahend, subtract the values ​​to obtain the difference. Perform an absolute value operation on the difference and use the resulting value as a separability index to characterize whether the forming response still maintains distinguishability under different parameter changes.

[0023] In S5, the process of outputting the prediction results for metal forming defects includes: S5-1. Based on the results of the separability index calculation, the separable indices generated sequentially within a continuous calculation cycle are written into the separable index sequence according to the calculation cycle order. Each separable index corresponds to a calculation cycle and is associated with the corresponding impulse identifier, so that the separable index sequence maintains consistency with the calculation cycle order in terms of sequence index values. The calculation cycle order refers to the sequence order determined by the order in which the separable indices are generated on the time axis. Based on the trigger time of each separable index calculation or the increasing relationship of the corresponding impulse identifier, the separable indices generated earlier are assigned to earlier sequence index positions, and the separable indices generated later are assigned to later sequence index positions, so that the separable index sequence maintains consistency with the time sequence of the actual calculation cycle in terms of sequence index values. S5-2. Take any two separable indices in the separable index sequence that are at adjacent sequence index positions as objects, perform numerical difference operation, subtract the value of the previous separable index from the value of the latter separable index, obtain the numerical difference of the separable indices in adjacent calculation periods, and write the obtained numerical difference into the separable index difference sequence. S5-3. In the separability index difference sequence, perform a judgment on all numerical differences. Determine that the numerical difference is less than zero as a decreasing difference, that the numerical difference is greater than zero as an increasing difference, and that the numerical difference is equal to zero as a level difference. Write each judgment result into the change judgment sequence so that the change judgment sequence can be used to reflect the change direction attribute of the separability index during adjacent weeks on the sequence index. S5-4. In the change determination sequence, according to the sequence index order, perform continuity determination on the determination results at adjacent index positions. If the determination results corresponding to the current index position and the next index position are both decreasing differences, then the current index position to the next index position forms a continuous decreasing segment, and continue to recursively determine whether the next index position and its subsequent index position satisfy the same decreasing state. Connect multiple continuous decreasing segments that continuously satisfy the decreasing state at adjacent index positions to form a decreasing segment. When the determination result corresponding to any index position in the change determination sequence is not a decreasing difference, terminate the construction of the current decreasing segment, and write the completed decreasing segment into the decreasing segment set for subsequent determination based on decreasing persistence. S5-5. Compare the sequence segment length of each descent segment in the descent segment set. Descending segments whose sequence segment length reaches or exceeds the preset descent judgment threshold are regarded as descent segments that meet the descent duration condition, and the corresponding separability index sequence segment is defined as the separable index segment that meets the descent duration condition. S5-6. Based on the index position of the separability index sequence corresponding to the separability index segment that meets the condition of continuous decline, locate the stroke identifier of the corresponding calculation cycle, determine that the material's response to changes in stamping parameters has shrunk during the corresponding stroke, mark the corresponding stroke data entry as the hardware forming defect prediction result, and output it to indicate the risk of forming failure in the stamping process.

[0024] A hardware forming defect prediction system based on stamping parameters includes a stamping count acquisition module, a change calculation module, a feature grouping module, a separable measurement module, and a shrinkage early warning module. The stamping acquisition module is used to acquire the stamping parameters and forming response data corresponding to each stamping in the metal forming process. The stamping parameters and forming response data are combined to form a stamping data entry and written into the cache queue of edge computing. The change calculation module is used to read multiple consecutive stroke data entries from the edge computing cache queue in stroke order, and calculate the difference of the stamping parameters in adjacent stroke data entries to obtain parameter change data that characterizes the change of stamping parameters between consecutive strokes. Based on parameter change data, the feature grouping module performs grouping processing on the stroke data entries according to the magnitude of parameter change. The stroke data entries with parameter change magnitude in different ranges are divided into different stroke data subsets, and response feature extraction is performed on the forming response data in each stroke data subset to obtain the corresponding response feature set. The separability measurement module is used to perform separability calculation on the response feature set corresponding to different stroke data subsets. By calculating the degree of difference of response features between different stroke data subsets and the degree of dispersion of response features within each stroke data subset, a separability index is formed. The shrinkage early warning module is used to perform trend analysis on the separability index over multiple consecutive calculation cycles. When the separability index shows a continuous decline over multiple consecutive calculation cycles and reaches the preset judgment condition, it is determined that the material's response to changes in stamping parameters has shrunk during the current metal forming process, and the corresponding stamping data entry is marked as the metal forming defect prediction result output.

[0025] It should be noted that, including but not limited to: This solution is derived layer by layer from the problems existing in the metal stamping field: During continuous stamping, the internal work hardening, strain accumulation, and die constraints of the material gradually weaken its response to changes in stamping parameters. However, this decline usually does not immediately manifest as obvious defects such as cracks or wrinkles. Existing methods only focus on whether defects eventually occur, ignoring the process by which the material loses its adjustment space in the latent stage. To solve the problem of early identification of this hidden failure, this solution chooses to regard continuous stamping as a trajectory of material response capability evolution. By capturing the real changes in stamping parameters and the real-time response characteristics of the material, and continuously analyzing whether the response can still be distinguished under different changing conditions, it can determine whether the material has entered a shrinkage state in the latent stage and output defect prediction results in advance. To support this goal, the solution sets up processes S1 to S5: The core function of S1 is to strictly bind the actual changes in stamping parameters that occur in each stroke to the material response behavior as a calculable item. The reason for recording the sampling time, aligning the time, and encapsulating the data is not for the sake of formal completeness, but to avoid alignment errors caused by asynchronous sampling when comparing cross-stroke differences, cross-feature comparisons, and cross-interval groupings, which would distort the material response capability. The purpose of S2 is not simply to perform a difference, but to construct the parameter change between consecutive strokes. The difference between strokes only has physical meaning when the consecutive strokes are strictly aligned in time order and the sampling time indices correspond one-to-one. Therefore, the continuity of the stroke identification order must be checked, and the sampling time index must be confirmed point by point before calculating the difference. The final result is parameter change data that can truly represent the intensity of the input change in consecutive strokes. S3 groups the parameter change data according to the magnitude of change in order to establish the comparability of the material's response capability under different parameter change conditions. If the data is not grouped, the material's response under slight and large parameter fluctuations will be mixed together, making it impossible to determine whether the material is still sensitive to the change. After grouping, response features such as peak value, slope, and integral are extracted for each group to make the differences in the material's response under different conditions explicit. The separability index in S4 is the core criterion of this scheme. It does not simply compare whether the characteristics are different, but examines the relationship between the difference in response under different parameter changes and the fluctuation of response under the same change. If the difference between groups gradually decreases, it means that the material is no longer sensitive to the input change. At the same time, if the dispersion within the group increases, it means that the random fluctuation of the material performance increases. The separability index combines the two into one quantity, reflecting whether the material's adjustment elasticity is shrinking. S5 constructs a trend of the separability index according to the time series. A continuous decrease indicates that the material is losing its distinguishable response to changes in stamping parameters. When the decrease reaches the preset duration condition, it is determined that the material has entered the shrinkage state and the defect prediction result is output. The reason for emphasizing duration is to eliminate short-term fluctuations caused by random noise and ensure that predictions are made only for the truly irreversible decay of response capability.

[0026] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting defects in metal forming based on stamping parameters, characterized in that, include: S1. Obtain the stamping parameters and forming response data corresponding to each stroke in the metal forming process, combine the stamping parameters and forming response data to form a stroke data entry, and write it into the cache queue of edge computing. S2. Read multiple consecutive stroke data entries from the edge computing cache queue in stroke order, and calculate the difference in stamping parameters in adjacent stroke data entries to obtain parameter change data. S3. Based on the parameter change data, the stroke data entries are grouped according to the magnitude of parameter change. The stroke data entries with parameter change ranges are divided into at least two groups of stroke data subsets with different parameter change conditions. Response feature extraction is performed on the forming response data in each stroke data subset to obtain the corresponding response feature set. S4. Perform separability calculation on the response feature sets corresponding to different stroke data subsets. By calculating the degree of difference in response features between different stroke data subsets and the degree of dispersion of response features within each stroke data subset, a separability index is formed. S5. Perform trend analysis on the separability index in multiple consecutive calculation cycles. When the separability index shows a continuous decline in multiple consecutive calculation cycles and reaches the preset judgment condition, it is determined that the material's response to the change of stamping parameters in the current metal forming process has shrunk, and the corresponding stamping data entry is marked as the metal forming defect prediction result output. In S4, the process of forming the separability index includes: S4-1. Taking the response feature vectors arranged in order of impulse in the response feature set as objects, write each response feature vector into the response feature vector order list; S4-2. Taking any two different subsets of impulse data as objects, based on the corresponding response feature vector order list, and using the response feature quantities at the same index position as comparison objects, calculate the absolute value of the numerical difference between the elements of each response feature vector, and write the absolute values ​​of the difference at all index positions into the cross-subset difference sequence in order. S4-3. Select the absolute value of the difference that is at the upper limit position in the cross-subset difference sequence, and use the corresponding absolute value of the difference as the upper limit difference degree between the data subsets of the current two pulses. S4-4. Taking any subset of impulse data as the object, in the order list of response feature vectors corresponding to the corresponding subset of impulse data, take the response feature quantities at the same index position as the comparison object, calculate the absolute value of the numerical difference between the elements of different response feature vectors, and write the absolute value of the difference corresponding to each index position into the difference sequence inside the subset in order. S4-5. Select the absolute value of the difference that is at the upper limit in the difference sequence within the subset, and use the corresponding absolute value of the difference as the upper limit of the dispersion within the data subset of the current impulse. S4-6. Using the upper limit difference value across subsets as the minuend and the upper limit of the upper limit dispersion value within all impulse data subsets as the subtrahend, subtract the values ​​to obtain the difference, and perform an absolute value operation on the difference. Use the resulting value as the separability index.

2. The method for predicting hardware forming defects based on stamping parameters according to claim 1, characterized in that: In S1, the process of writing to the edge computing cache queue includes: S1-1. Through the signal acquisition interface between the edge computing side and the stamping equipment, the stamping parameters corresponding to each stamping stroke are acquired, and the sampling time information of each stamping parameter is recorded during acquisition. The stamping parameters and the corresponding sampling time information are used as the original sampling data of the stamping parameters. S1-2. Within the preset punch time window, continuously sample the load measurement signal and slide position measurement signal corresponding to the punching parameters. Organize the continuously sampled load measurement signal into a load curve sampling sequence in time order, and organize the continuously sampled slide position measurement signal into a slide position curve sampling sequence in time order. Record the sampling time information corresponding to each sampling point as the original sampling data of the forming response. S1-3. Based on the stroke time window corresponding to the current stroke, associate the original sampling data of stamping parameters and the original sampling data of forming response that fall within the same stroke time window range with the same stroke, and generate the stroke identifier corresponding to the current stroke. S1-4. Using the stroke identifier as an index, the original sampling data of stamping parameters and the original sampling data of forming response that are associated with the same stroke identifier within the corresponding stroke time window are collected according to the stroke identifier correspondence to form a stamping parameter data set and a forming response data set. S1-5. Using the stroke time window as the time range, based on the sampling time information of each sampling point, a unified time series covering the stroke time window is generated sequentially at preset time intervals, starting from the beginning time of the stroke time window; interpolation is performed on the sampling points on the unified time series so that the stamping parameter data and forming response data have a one-to-one corresponding sampling time index on the time axis. S1-6. Encapsulate the aligned stamping parameter data and forming response data according to a unified data structure to generate stamping data entries; write the stamping data entries into the pre-allocated cache area in the working memory of the edge computing node in the order of sampling time information, and insert the stamping data entries into the cache queue in the order of writing.

3. The method for predicting hardware forming defects based on stamping parameters according to claim 2, characterized in that: In S2, the process of obtaining parameter change data includes: S2-1. In the edge computing node, based on the cache queue cached in the working memory, the order in which the impulse data entries are enqueued in the cache queue is used as the impulse order. Starting from the head of the queue, multiple consecutive impulse data entries are read sequentially, and each read impulse data entry is temporarily stored in the sequential buffer. S2-2. Based on the stroke identifier field carried by each stroke data entry in the sequential buffer, perform sequential continuity check on the stroke identifier field, determine whether the stroke identifiers of two adjacent stroke data entries are arranged continuously in an increasing pattern, and remove stroke data entries with discontinuous stroke identifiers. S2-3. Based on the stamping parameter field and sampling time index field carried inside the stamping data entries in the sequential buffer, extract the aligned stamping parameter data from each stamping data entry, and write the stamping parameter data into the stamping parameter order list according to the stamping order.

4. The method for predicting hardware forming defects based on stamping parameters according to claim 3, characterized in that: In S2, the process of obtaining parameter change data also includes: S2-4. Take any two adjacent stamping parameter data in the stamping parameter sequence list as a group. Based on the sampling time index field, check each sampling time index position of the two adjacent stamping parameter data in the stamping parameter sequence list in the unified time series one by one to determine whether the two stamping parameter data have corresponding values ​​at each sampling time index position. The alignment is deemed successful only if all sampling time index positions within the unified time series satisfy the condition that two stamping parameter data have corresponding values ​​at the same time; otherwise, the alignment is deemed unsuccessful, and the adjacent stamping times corresponding to the two stamping parameter data are removed from the current difference calculation object. S2-5. Under the premise that the alignment judgment is passed, take the two stamping parameter data of adjacent strokes as objects, perform difference calculation for each sampling time index position, subtract the stamping parameter value of the sampling time index position of the previous stroke from the stamping parameter value of the sampling time index position of the next stroke, and form a list of stamping parameter difference entries that correspond one-to-one with the sampling time index. S2-6. Organize all stamping parameter difference entries corresponding to the same adjacent strokes into a parameter difference sequence according to the sampling time index, and write each parameter difference sequence into the parameter difference sequence list according to the stroke order as parameter change data.

5. The method for predicting hardware forming defects based on stamping parameters according to claim 4, characterized in that: In S3, the process of obtaining the corresponding response feature set includes: S3-1. Based on the parameter change data and the corresponding stroke identifier, associate the parameter difference sequence of each stroke with the corresponding stroke identifier to form a list of parameter change records arranged in stroke order. S3-2. For each record in the parameter change record list, in the parameter difference sequence corresponding to the current record, perform absolute value operation on the stamping parameter difference at each sampling time index position, and select the upper limit value of the absolute value in each parameter difference sequence as the parameter change amplitude value of the corresponding stroke, and write the parameter change amplitude value and the corresponding stroke identifier into the parameter change amplitude mapping table. S3-3. Preset multiple non-overlapping parameter variation ranges. Each parameter variation range is determined by a pair of lower and upper limits of parameter variation. Based on the parameter variation range values ​​in the parameter variation range mapping table, compare the parameter variation range value corresponding to each stroke with the multiple parameter variation range ranges one by one to determine the parameter variation range range to which the current parameter variation range value falls. Group the strokes that fall into the same parameter variation range range into the same group to form strokes with different parameter variation range ranges.

6. The method for predicting hardware forming defects based on stamping parameters according to claim 5, characterized in that: In S3, the process of obtaining the corresponding response feature set also includes: S3-4. Using the stroke identifier group as an index, read the corresponding stroke data entries from the cache queue of the edge computing node according to the stroke identifier, and group the stroke data entries belonging to the same stroke identifier group to form the corresponding stroke data subset. S3-5. For each stroke data subset, extract the aligned forming response data from each stroke data entry in the corresponding stroke data subset, and organize the extracted forming response data into a response data matrix according to the stroke order and sampling time index order. The row index of the response data matrix corresponds to the stroke identifier, and the column index corresponds to the sampling time index position on the unified time series. S3-6. Based on the response data matrix corresponding to each stroke data subset, perform response feature extraction processing on each row of forming response data to obtain multiple response feature quantities. The multiple response feature quantities include the peak value, peak occurrence time, rise phase slope, and curve integral value within a preset time window of the forming response calculated on a unified time series. Combine the multiple response feature quantities into the response feature vector corresponding to the current stroke, and combine the response feature vectors corresponding to each stroke in the same stroke data subset in stroke order to form the response feature set of the corresponding stroke data subset. The peak value calculation process for the forming response is as follows: traverse the forming response data of the current stroke at all sampling time index positions corresponding to the unified time series, compare the forming response values ​​at each sampling time index position, select the sampling point at the upper limit value, and take the forming response value corresponding to the sampling point as the peak value of the forming response of the current stroke. The peak occurrence time is: after determining the sampling time index position corresponding to the peak of the forming response, read the time value in the unified time series corresponding to the sampling time index position, and take the corresponding time value as the peak occurrence time of the forming response. The slope of the rising phase is as follows: In the forming response data of the current stroke, the starting sampling time index position on the unified time series is taken as the starting point of the rising phase, and the sampling time index position corresponding to the peak value of the forming response is taken as the ending point of the rising phase. The forming response values ​​and corresponding time values ​​at the starting point and the ending point are read respectively, the ratio of the increment of the forming response value to the increment of time is calculated, and the corresponding ratio is taken as the slope of the rising phase of the current stroke. The integral value of the curve within the preset time window is as follows: Select each sampling time index position between the start time and the end time corresponding to the preset time window on the unified time series, obtain the forming response value of each sampling time index position according to the sampling time index order, use the time difference between adjacent sampling time index positions as the integration step size using a preset time interval, perform trapezoidal approximate area calculation on the forming response value corresponding to each adjacent sampling time index position and accumulate it segment by segment within the preset time window, and use the sum of the accumulated areas as the integral value of the forming response curve of the current stroke within the preset time window.

7. The method for predicting hardware forming defects based on stamping parameters according to claim 6, characterized in that: In S5, the process of outputting the prediction results for metal forming defects includes: S5-1. Based on the calculation results of the separability index, the separability indices generated sequentially within the continuous calculation cycle will be written into the separability index sequence in the order of the calculation cycle. S5-2. Take any two separable indices in the separable index sequence that are at adjacent sequence index positions as objects, perform numerical difference operation, subtract the value of the previous separable index from the value of the latter separable index, obtain the numerical difference of the separable indices in adjacent calculation periods, and write the obtained numerical difference into the separable index difference sequence. S5-3. In the sequence of differences of separability index, perform a judgment on all numerical differences, judge the numerical difference less than zero as a decreasing difference, judge the numerical difference greater than zero as an increasing difference, and judge the numerical difference equal to zero as a level difference, and write each judgment result into the change judgment sequence. S5-4. In the change determination sequence, according to the sequence index order, perform continuity determination on the determination results of adjacent index positions. If the determination results of the current index position and the next index position are both decreasing differences, then the current index position to the next index position forms a continuous decreasing segment, and continue to recursively determine whether the next index position and its next index position satisfy the same decreasing state. Connect multiple continuous decreasing segments that continuously satisfy the decreasing state of adjacent index positions to form a decreasing segment. When the determination result of any index position in the change determination sequence is not a decreasing difference, terminate the construction of the current decreasing segment, and write the completed decreasing segment into the decreasing segment set. S5-5. Compare the sequence segment length of each descent segment in the descent segment set. Descending segments whose sequence segment length reaches or exceeds the preset descent judgment threshold are regarded as descent segments that meet the descent duration condition, and the corresponding separability index sequence segment is defined as the separable index segment that meets the descent duration condition. S5-6. Based on the index position of the separability index sequence corresponding to the separability index segment that meets the condition of continuous decline, locate the stroke identifier of the corresponding calculation cycle, determine that the material's response to changes in stamping parameters has shrunk during the corresponding stroke, and mark the corresponding stroke data entry as the hardware forming defect prediction result.

8. A hardware forming defect prediction system based on stamping parameters, comprising the hardware forming defect prediction method based on stamping parameters as described in claim 7, including a stamping sequence acquisition module, a change calculation module, a feature grouping module, a separable measurement module, and a shrinkage early warning module, characterized in that: The stamping acquisition module is used to acquire the stamping parameters and forming response data corresponding to each stamping in the metal forming process. The stamping parameters and forming response data are combined to form a stamping data entry and written into the cache queue of edge computing. The change calculation module is used to read multiple consecutive stroke data entries from the edge computing cache queue in stroke order, and calculate the difference of the stamping parameters in adjacent stroke data entries to obtain parameter change data that characterizes the change of stamping parameters between consecutive strokes. Based on parameter change data, the feature grouping module performs grouping processing on the stroke data entries according to the magnitude of parameter change. The stroke data entries with parameter change magnitude in different ranges are divided into different stroke data subsets, and response feature extraction is performed on the forming response data in each stroke data subset to obtain the corresponding response feature set. The separability measurement module is used to perform separability calculation on the response feature set corresponding to different stroke data subsets. By calculating the degree of difference of response features between different stroke data subsets and the degree of dispersion of response features within each stroke data subset, a separability index is formed. The shrinkage early warning module is used to perform trend analysis on the separability index over multiple consecutive calculation cycles. When the separability index shows a continuous decline over multiple consecutive calculation cycles and reaches the preset judgment condition, it is determined that the material's response to changes in stamping parameters has shrunk during the current metal forming process, and the corresponding stamping data entry is marked as the metal forming defect prediction result output.

Images