A multi-dimensional fusion prediction method and system for battery shell SPR riveting quality

Abstract

The embodiment of the application discloses a kind of battery shell SPR riveting quality multi-dimensional fusion prediction method and system, it is related to battery shell connection quality detection technical field.The method comprises: obtaining force displacement data, acoustic vibration response data and equipment running data in the process of battery shell SPR riveting, and obtaining rivet point visual topography data after riveting is completed;According to the force displacement data, determine the multiple forming stages in the process of SPR riveting, and the stage matching of acoustic vibration response data and equipment running data is carried out;From the acoustic vibration response data and equipment running data corresponding to each forming stage, riveting micro-response feature is extracted;From the rivet point visual topography data, rivet topography feature is extracted;The riveting micro-response feature and rivet topography feature are fused, the risk of battery shell SPR riveting defect is determined, and corresponding riveting quality prediction result is generated.The application improves the accuracy, timeliness and traceability of battery shell SPR riveting quality prediction.

Description

Technical Field

[0001] This application relates to the field of battery casing connection quality inspection technology, and in particular to a multi-dimensional fusion prediction method and system for battery casing SPR riveting quality. Background Technology

[0002] Self-Piercing Riveting (SPR) is a common mechanical connection method used in thin-plate structures such as battery casings, battery trays, and battery pack frames. This process typically involves a rivet piercing the upper sheet material under pressure and expanding within the lower sheet material to form a mechanically locked structure. Current SPR quality inspection relies heavily on the visual morphology of the rivet points, riveting force-displacement curves, or post-construction cross-section inspection. While visual inspection can identify surface defects such as rivet misalignment, head deformation, and abnormal surface indentations, its ability to identify defects that are not visually apparent or obvious, such as internal cracks in the lower sheet, early stages of localized tearing, and incomplete bonding between sheets, is limited. During SPR riveting, the contact of the sheets, rivet piercing, rivet expansion, and bottom forming all cause subtle changes in acoustic emission, vibration impact, and equipment load. When cracks appear in the lower sheet, localized tears occur, or the rivet deviates, these micro-response signals often appear before obvious surface defects.

[0003] Therefore, how to effectively integrate the acoustic and vibration micro-response during the riveting process, the equipment operating status, and the visual morphology after riveting in order to predict the risk of defects in SPR riveting of battery casings in advance is a technical problem that needs to be solved. Summary of the Invention

[0004] This application provides a multi-dimensional fusion prediction method and system for battery casing SPR riveting quality, which improves the accuracy, timeliness and traceability of battery casing SPR riveting quality prediction.

[0005] This application provides the following solution: According to the first aspect, a multi-dimensional fusion prediction method for battery casing SPR riveting quality is provided, comprising: acquiring force-displacement data, acoustic-vibration response data, and equipment operation data during the battery casing SPR riveting process, and acquiring visual morphology data of the rivet points after riveting is completed; determining multiple forming stages in the SPR riveting process based on the force-displacement data, and performing stage matching on the acoustic-vibration response data and equipment operation data; extracting riveting micro-response features from the acoustic-vibration response data and equipment operation data corresponding to each forming stage; extracting rivet point morphology features from the rivet point visual morphology data; fusing the riveting micro-response features and rivet point morphology features to determine the risk of SPR riveting defects in the battery casing, and generating corresponding riveting quality prediction results.

[0006] According to the second aspect, a multi-dimensional fusion prediction system for battery casing SPR riveting quality is provided, comprising: a data acquisition module for acquiring force-displacement data, acoustic-vibration response data, and equipment operation data during the battery casing SPR riveting process, and acquiring visual morphology data of the rivet points after riveting is completed; a stage matching module for determining multiple forming stages in the SPR riveting process based on the force-displacement data, and performing stage matching on the acoustic-vibration response data and equipment operation data; a micro-response feature extraction module for extracting riveting micro-response features from the acoustic-vibration response data and equipment operation data corresponding to each forming stage; a morphology feature extraction module for extracting rivet point morphology features from the rivet point visual morphology data; and a fusion prediction module for fusing the riveting micro-response features and rivet point morphology features to determine the risk of SPR riveting defects in the battery casing, and generating corresponding riveting quality prediction results.

[0007] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application achieves multi-dimensional perception of the riveting process status and rivet point formation results by simultaneously acquiring force-displacement data, acoustic-vibration response data, equipment operation data, and post-riveting visual morphology data during the SPR riveting process. The scheme first uses force-displacement data to divide the riveting formation stages, then matches the acoustic-vibration response and equipment operation status to the corresponding stages, enabling the association between minute acoustic-vibration anomalies and load disturbances and processes such as plate contact, rivet puncture, rivet expansion, and bottom forming. Simultaneously, it combines the visual morphology characteristics of the rivet points for fusion judgment, which can identify appearance anomalies such as rivet misalignment, non-adhesion between plates, and local deformation. It can also use process micro-response to compensate for the insufficient identification of latent defects such as lower plate cracks and local tears by visual inspection, thereby improving the accuracy, timeliness, and traceability of battery casing SPR riveting quality prediction.

[0008] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart of a multi-dimensional fusion prediction method for battery casing SPR riveting quality provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the division of the SPR riveting forming stages provided in an embodiment of this application; Figure 3This is a schematic diagram illustrating multidimensional evidence fusion and risk output provided in an embodiment of this application. Figure 4 The structural block diagram of the multidimensional fusion prediction system for battery casing SPR riveting quality provided in the embodiments of this application is shown. Detailed Implementation

[0011] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0012] Figure 1 A flowchart illustrating the multi-dimensional fusion prediction method for SPR riveting quality of battery casing provided in this application embodiment. Figure 1 As shown, the method may include the following steps: Step 101: Obtain force displacement data, acoustic vibration response data, and equipment operation data during the SPR riveting process of the battery casing, and obtain visual morphology data of the rivet points after the riveting is completed.

[0013] Step 102: Determine multiple forming stages in the SPR riveting process based on the force displacement data, and perform stage matching between the acoustic vibration response data and the equipment operation data.

[0014] Step 103: Extract riveting micro-response features from the acoustic and vibration response data and equipment operation data corresponding to each forming stage.

[0015] Step 104: Extract rivet shape features from the rivet visual shape data.

[0016] Step 105: Integrate the riveting micro-response features and riveting point morphology features to determine the risk of SPR riveting defects in the battery casing and generate the corresponding riveting quality prediction results.

[0017] As can be seen from the above process, this application achieves multi-dimensional perception of the riveting process status and rivet point forming results by simultaneously acquiring force displacement data, acoustic vibration response data, equipment operation data, and post-riveting visual morphology data during the SPR riveting process. This scheme first uses force displacement data to divide the riveting forming stages, then matches the acoustic vibration response and equipment operation status to the corresponding stages, enabling the association between minute acoustic vibration anomalies and load disturbances and processes such as plate contact, rivet piercing, rivet expansion, and bottom forming. Simultaneously, it combines the visual morphology features of the rivet points for fusion judgment, which can identify appearance anomalies such as rivet misalignment, non-adhesion between plates, and local deformation, and can also compensate for the insufficient identification of latent defects such as lower plate cracks and local tears by using process micro-response. This improves the accuracy, timeliness, and traceability of battery casing SPR riveting quality prediction.

[0018] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments.

[0019] Step 101 specifically involves: acquiring force displacement data, acoustic vibration response data, and equipment operation data during the SPR riveting process of the battery casing, and acquiring visual morphology data of the rivet points after the riveting is completed.

[0020] This step is used to establish the multi-source data foundation required for predicting SPR riveting quality. Battery casing SPR riveting is not a single static forming process, but rather involves continuous changes within a short period, including sheet contact, rivet piercing, rivet expansion, and bottom forming. The mechanical, acoustic, and vibration responses, as well as the equipment operating status, generated at different stages are not the same. Therefore, it is necessary to simultaneously collect force-displacement data, acoustic-vibration response data, and equipment operating data during the riveting process, and further collect visual morphology data of the rivet points after riveting is completed.

[0021] Force-displacement data reflects the relationship between the load applied by the indenter and the rivet's advance displacement during the riveting process. This data helps determine whether the rivet has properly penetrated the sheet metal, whether the riveting resistance is abnormal, whether the final pressing is sufficient, and further identifies the boundaries of critical stages in the SPR riveting process. This data is typically acquired by pressure sensors, force sensors, or force feedback units in the servo drive system of the riveting equipment. It can also be generated in conjunction with stroke data output from displacement encoders, linear displacement sensors, or the equipment controller.

[0022] Acoustic-vibration response data is used to capture the subtle dynamic responses of materials during riveting, caused by impact, puncture, plastic deformation, or localized cracking. This data includes acoustic emission and vibration acceleration data collected during the riveting process. Acoustic emission data reflects high-frequency elastic wave signals generated by the initiation of microcracks within the sheet metal, localized tearing, fiber breakage, or rapid stress release. Vibration acceleration data reflects the mechanical impact and residual vibration changes caused by rivet puncture, rivet expansion, lower die bearing, and bottom locking forming.

[0023] Equipment operation data is used to characterize the operating load and control status of the riveting equipment when performing SPR riveting operations. Equipment operation data includes at least two of the following: motor current, spindle speed, indenter displacement speed, and riveting contact activation moment. Motor current reflects the output load changes of the servo system during riveting; when there is abnormal sheet resistance, rivet expansion is hindered, or there is poor adhesion between sheets, the motor current may experience a momentary increase or fluctuation. Spindle speed and indenter displacement speed reflect the propulsion stability of the riveting actuator; when the rivet encounters abnormal resistance or the forming path deviates, the speed may experience a short-term decrease or unstable changes. The riveting contact activation moment can serve as a time reference for multi-source data synchronization, enabling the alignment of force displacement data, acoustic vibration response data, and equipment operation data within the same riveting cycle.

[0024] Visual morphology data of the rivet point is used to reflect the forming state of the rivet point surface and surrounding area after riveting. This data can be acquired through industrial cameras, structured light inspection devices, or 3D contour inspection devices to identify appearance features such as rivet point center offset, rivet point outer contour asymmetry, sheet metal bonding gap, abnormal local indentation, and uneven circumferential deformation. Visual morphology data can intuitively reflect the appearance quality after riveting, but its ability to identify internal cracks or early localized tears in the lower plate is limited. Therefore, it needs to be combined with acoustic and vibration response data and equipment operation data.

[0025] By employing the aforementioned data acquisition methods, this application can integrate the internal micro-damage response during the riveting process, equipment load disturbances, and the post-riveting appearance into the quality prediction process. Force-displacement data provides a basis for stage division, acoustic emission data and vibration acceleration data provide evidence of the micro-damage process, equipment operation data provides evidence of the riveting execution status, and visual morphology data of the riveting points provides evidence of the final formed appearance.

[0026] Step 102 specifically involves: determining multiple forming stages in the SPR riveting process based on the force displacement data, and performing stage matching between the acoustic vibration response data and the equipment operation data.

[0027] Based on force-displacement data, multiple forming stages in the SPR riveting process can be determined. First, the force-displacement data of a single rivet point within one SPR riveting cycle can be obtained. This data can be represented as a curve of the indenter force changing with time, or as a curve of the indenter force changing with displacement. Since the plate contact, rivet penetration, rivet expansion, and bottom forming in the SPR riveting process correspond to different force variation patterns, the complete riveting process can be divided into multiple physically meaningful forming stages by analyzing key change points in the force-displacement data.

[0028] Figure 2This diagram illustrates the SPR riveting formation stage division provided in this embodiment. Specifically, the force-displacement data can first be filtered and smoothed to remove acquisition noise and short-term jitter, and then the slope of the force value changing with time or displacement can be calculated. When the pressure head begins to contact the sheet metal and apply a load, the force value will start to rise continuously from a near-zero state. The moment when the force value exceeds a preset starting threshold and maintains an upward trend can be determined as the initial rise point of the force value. This point usually corresponds to the starting position where the riveting action truly enters the stress state and can be used as the starting point of the sheet metal contact stage.

[0029] After the plates make contact, the rivet continues to press down and gradually penetrates the upper plate. When the rivet tip pierces the plate or the plate undergoes significant plastic deformation, the slope of the force curve will change abruptly, for example, the slope will increase rapidly, decrease rapidly, or a clear inflection point will appear. Therefore, the point of abrupt change in the force slope sequence can be determined based on the location of the abrupt change. This point reflects the transition of the riveting process from simple contact and pressing to rivet piercing or strong plastic deformation, and can serve as the boundary between the plate contact stage and the rivet piercing stage.

[0030] After the rivet completes its initial piercing, it continues to advance downwards and expand within the sheet metal. At this point, the displacement process is relatively continuous, and the force change no longer exhibits the abrupt change seen during initial piercing, but rather enters a relatively stable advancing state. The stable point of the rivet's advance can be determined based on the rate of displacement change, the magnitude of force increase, and the stability of the force slope. For example, when the displacement continues to increase and the force slope remains stable within a preset time range, this position can be identified as the stable point of displacement advance. This point can be used to characterize the transition of the rivet from the piercing process to the expansion and forming process.

[0031] As the pressure head continues to press down, the rivet legs expand within the underlying sheet metal, forming a bottom-locking structure. The force typically increases again, approaching the final pressure state. When the force reaches the final pressure setting range, the displacement approaches the end of the stroke, and the force or displacement changes stabilize, the final pressure forming point can be determined. This point corresponds to the final forming stage of SPR riveting, indicating that the rivet bottom expansion and locking are essentially complete.

[0032] After identifying the initial force rise point, the point of abrupt change in force slope, the point of stable displacement propagation, and the final compression forming point, these key points can be used as stage boundaries to segment the SPR riveting process. The period from the initial force rise point to the point of abrupt change in force slope can be defined as the sheet metal contact stage, which mainly reflects the contact between the indenter and the sheet metal, the initial clamping of the sheet metal, and the clamping state. The period from the point of abrupt change in force slope to the point of stable displacement propagation can be defined as the rivet penetration stage, which mainly reflects the process of the rivet penetrating the upper sheet metal and causing localized cracking or plastic deformation. The interval from the point of stable displacement propagation to the final compression forming point can be defined as the rivet expansion stage, which mainly reflects the process of the rivet advancing within the sheet metal and expanding radially. The area near or corresponding to the final compression forming point can be defined as the bottom forming stage, which mainly reflects the process of the rivet legs forming a mechanical lock within the lower sheet metal.

[0033] In this way, the SPR riveting process is no longer analyzed as a whole in a rough manner, but is broken down into multiple stages corresponding to the actual forming mechanism. This allows for clearer identification of whether a sudden acoustic emission, vibration shock, or equipment load disturbance occurs during the sheet contact, rivet puncture, rivet expansion, or bottom forming stage when matching acoustic and vibration response data with equipment operating data. This improves the specificity of defect risk assessment. For example, bottom plate cracks are more likely to exhibit continuous acoustic and vibration abnormalities during the rivet expansion or bottom forming stage, while poor adhesion between sheets may manifest as equipment load disturbances in both the sheet contact stage and the subsequent advancement stage.

[0034] Next, the acoustic vibration response data and equipment operation data are matched in stages. First, the riveting trigger moment corresponding to the riveting equipment issuing the riveting execution command or the pressure head starting to descend is determined as the initial time reference for the current riveting cycle, and this moment is recorded as time zero. Force displacement data, acoustic vibration response data, and equipment operation data all carry corresponding sampling times or timestamps during acquisition. By subtracting the riveting trigger moment, data from different acquisition channels can be converted to the same relative time axis. This eliminates the influence caused by different actual start times in different riveting cycles and allows various types of data to be correlated according to the same riveting point and the same riveting cycle.

[0035] After initial time alignment, the boundary times of each stage, identified based on the force-displacement data, are read. Each stage boundary can include the time corresponding to the initial force rise point, the time corresponding to the sudden change in force slope, the time corresponding to the displacement stabilization point, and the time corresponding to the final compression forming point. Based on the time range between adjacent stage boundaries, time windows are generated for the sheet metal contact stage, rivet piercing stage, rivet expansion stage, and bottom forming stage, respectively. Each time window includes at least the stage start time and the stage end time. If necessary, a short overlap interval can be retained on both sides of the stage boundary to avoid missing transient responses near the stage boundary due to data truncation.

[0036] Although force-displacement data, acoustic-vibration response data, and equipment operation data all use the riveting contact moment as the initial time reference, there may still be a millisecond-level delay in acoustic-vibration response data compared to force-displacement data due to differences in sensor installation location, mechanical vibration propagation path, signal conditioning circuit, and data acquisition channel response speed. For example, a force sensor can directly sense changes in the indenter load, while a vibration sensor installed on the lower die or fixture can only detect the corresponding response after the mechanical impact has propagated through the sheet metal, die, and fixture. If the acoustic-vibration response data is segmented directly according to the initial time axis, vibration impacts that should belong to the rivet piercing stage may be classified as part of the rivet expansion stage, thus requiring further time delay compensation.

[0037] For force-displacement data, the difference between adjacent force values ​​can be calculated based on continuous sampling times. The ratio of this difference to the corresponding sampling time interval is determined as the force change rate, thus forming a force change rate sequence. To reduce the impact of random noise on the change rate calculation, the force-displacement data can be smoothed first, or the formed force change rate sequence can be averaged using a short window. Subsequently, within a preset search interval, the sampling position where the absolute value of the force change rate is the largest or exceeds a preset change rate threshold is found, and the time corresponding to this sampling position is determined as the peak time of the force change rate. This time usually corresponds to a significant mechanical abrupt change when a rivet pierces the plate, the rivet begins to expand, or the bottom forming resistance changes rapidly.

[0038] For acoustic-vibration response data, the vibration impact amplitude at each sampling time can be extracted from the vibration acceleration data. Specifically, the absolute value of vibration acceleration can be directly used as the vibration impact amplitude, or the amplitude can be obtained by synthesizing accelerations in multiple vibration directions. Envelope processing can also be used to highlight short-time impact responses. The vibration impact amplitudes are arranged according to sampling time to form a vibration impact amplitude sequence. Then, within the search time range corresponding to the peak force change rate, the sampling position with the largest vibration impact amplitude is determined, and the time corresponding to this position is determined as the peak vibration impact time. By limiting the search range, irrelevant peak values ​​generated by equipment return vibration, fixture collisions, or environmental interference can be avoided from being mistaken for riveting impact peak values.

[0039] Subtracting the peak time of the force change rate from the peak time of the vibration impact yields the time deviation of the acoustic vibration response data relative to the force displacement data. When the peak time of the vibration impact is later than the peak time of the force change rate, the acoustic vibration response data is relatively lagging, and the data can be shifted forward as a whole according to the time deviation. When the peak time of the vibration impact is earlier than the peak time of the force change rate, it indicates a possible reverse deviation in the time references of the two acquisition channels, and the acoustic vibration response data can be shifted backward as a whole. During this shift, sampling point translation or time coordinate correction can be used. When the time deviation is not an integer multiple of the sampling period, interpolation can be used to determine the compensated acoustic vibration response value.

[0040] Time delay compensation can be performed on the entire acoustic and vibration response data of the current rivet point, maintaining the original time correlation between acoustic emission data and vibration acceleration data. In one implementation, a uniform time deviation can be calculated using vibration and impact data, and the vibration acceleration data and acoustic emission data corresponding to the current rivet point can be synchronously corrected according to this time deviation. In another implementation, when the acoustic emission acquisition channel and the vibration acquisition channel have different fixed acquisition delays, a preset calibration delay for each channel can be superimposed on the uniform time deviation to complete the time correction of acoustic emission data and vibration acceleration data respectively.

[0041] Equipment operation data is typically output directly from the riveting equipment controller, and its time base is consistent with the time base of the riveting trigger moment and force displacement data. Therefore, motor current, spindle speed, and pressure head displacement speed within the corresponding time range can be directly extracted based on the time window corresponding to each forming stage. When the sampling frequency of the equipment operation data is not synchronized with the force displacement data, it can be matched according to timestamps, or the equipment operation data can be converted to a unified time axis through resampling or interpolation, thereby ensuring that the acoustic vibration response data and equipment operation data within each forming stage have corresponding sampling intervals.

[0042] After completing time delay compensation and time axis unification, acoustic emission data, vibration acceleration data, and equipment operation data are extracted according to the time windows corresponding to the plate contact stage, rivet piercing stage, rivet expansion stage, and bottom forming stage. Each forming stage forms a corresponding data segment, which is stored in association with the current rivet point number and stage identifier. Subsequently, the acoustic emission burst increment, vibration energy increment, vibration residual vibration holding amount, and equipment load disturbance amount can be extracted from each data segment, ensuring that each micro-response characteristic accurately corresponds to its actual forming stage.

[0043] By using the above-mentioned stage matching method, not only can the initial synchronization of multi-source data be completed at the moment of riveting contact, but the time delay caused by mechanical transmission and acquisition links can also be corrected by using the correspondence between the peak value of force change rate and the peak value of vibration impact.

[0044] Step 103 specifically involves extracting riveting micro-response features from the acoustic and vibration response data and equipment operation data corresponding to each forming stage.

[0045] After completing the stage matching of acoustic vibration response data and equipment operation data, data segments corresponding to the plate contact stage, rivet piercing stage, rivet expansion stage, and bottom forming stage can be obtained respectively. Since the plate thickness, clamping degree, sensor coupling state, and initial load of the equipment may differ at different rivet points, directly comparing the original signal amplitudes of each stage could easily lead to misinterpreting fundamental differences between rivet points as riveting abnormalities. Therefore, this application first establishes a baseline response using the plate contact stage data of the current rivet point itself, and then uses this baseline response as a comparison benchmark for subsequent forming stages.

[0046] Specifically, the number of acoustic emission events, acoustic emission amplitude, duration, and signal energy can be extracted from the acoustic emission data corresponding to the contact stage of the sheet metal. Vibration energy, vibration peak value, and vibration decay state can be extracted from the vibration acceleration data. Furthermore, the basic operating levels of motor current, spindle speed, and indenter displacement speed can be extracted from the equipment operation data. After normalization or statistical summarization of the above data, the background response characteristics of the current riveting point are formed. These background response characteristics are used to characterize the basic acoustic and vibration responses and equipment load states generated during the initial contact between the indenter and the sheet metal, the sheet metal clamping, and normal equipment loading, and are not directly regarded as riveting micro-damage responses.

[0047] For the rivet piercing stage, rivet expansion stage, and bottom forming stage, the acoustic and vibration response data and equipment operation data for each stage can be compared with the background response characteristics. During the comparison, the difference or change ratio between the stage statistics and the corresponding background statistics can be used to reduce the influence of differences in the original signal magnitude at different rivet points. Thus, the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance corresponding to each forming stage can be obtained.

[0048] Acoustic emission burst increment reflects the sudden increase in acoustic emission response in the current forming stage relative to the sheet contact stage. Specifically, it identifies acoustic emission events with amplitudes exceeding the background acoustic emission level within the current stage, and determines the acoustic emission burst increment based on the number, amplitude, or duration of these new events. Transient elastic waves are typically released when rivets pierce the sheet, microcracks form within the sheet, or rapid tearing occurs in localized areas, thus increasing the acoustic emission burst increment. Therefore, this increment can be used to characterize the likelihood of localized material damage in the current stage.

[0049] The vibration energy increment reflects the increase in mechanical impact intensity relative to the baseline vibration level during the current forming stage. It can be calculated by accumulating the absolute value of the vibration acceleration within the current stage, using the sum of squares, or by statistical analysis of frequency band energy, and then compared with the vibration energy corresponding to the sheet contact stage. When rivet penetration resistance is abnormal, rivet expansion is hindered, or localized material fracture occurs during the bottom forming process, the impact response may be significantly enhanced, thus increasing the vibration energy increment of the corresponding stage.

[0050] Vibration persistence is used to characterize the extent to which the vibration response persists within the current forming stage after a major impact. It can be achieved by first determining the moment of the major vibration impact in the current stage, and then statistically analyzing the duration, residual energy, or decay rate of the vibration amplitude above the background vibration level after the impact. When the vibration response rapidly decreases to the background level after the impact, the vibration persistence is relatively small; when the sheet metal continues to crack, rivet expansion is unstable, or the structural contact state continues to change, the vibration response may persist for a long time, thus increasing the vibration persistence. This feature can distinguish between short-term normal impacts and abnormal micro-damage responses with lasting effects.

[0051] Equipment load disturbance reflects the degree of deviation of the riveting equipment's operating state from the baseline state during the current forming stage relative to the contact state of the sheet metal. It can be calculated separately by the increase in motor current relative to the baseline current level, the decrease in spindle speed relative to the baseline speed, and the fluctuation in indenter displacement speed relative to the expected feed speed, and then combining one or more of these changes. If the rivet expansion encounters abnormal resistance, there is an area of ​​non-fitting between the sheets, or the rivet is misaligned, the equipment needs to change its output load or feed state, which may result in an increase in motor current, fluctuations in spindle speed, or a short-term decrease in indenter displacement speed.

[0052] For each forming stage, the corresponding acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance are combined in a preset order to form the stage increment characteristics of that forming stage. The stage increment characteristics do not simply represent the magnitude of an anomaly in a single signal, but simultaneously record changes in acoustic emission within the material, changes in mechanical vibration, the duration of residual vibration, and the state of equipment load disturbance, thus forming a multi-dimensional description of the current forming stage. For example, a significant increase in the acoustic emission burst increment and a high vibration residual vibration retention may indicate that there may be persistent material damage in the current stage; a large equipment load disturbance and a small acoustic emission burst increment may reflect more changes in equipment propulsion resistance or sheet bonding state.

[0053] For each forming stage, the data can be arranged in a fixed order: acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance, forming the corresponding stage increment feature. Specifically, the acoustic emission burst increment is taken as the first feature item, the vibration energy increment as the second feature item, the vibration residual vibration retention as the third feature item, and the equipment load disturbance as the fourth feature item. Since the dimensions and numerical ranges of different feature items differ, each feature item can be normalized using the normal variation range of the corresponding feature item in normal riveting point samples, converting each feature item into a unified data scale. After normalization, the stage increment features are arranged in the above fixed order. Thus, each forming stage has the same data structure, ensuring that the same position in different forming stages always corresponds to the same micro-response feature, facilitating subsequent inter-stage comparisons based on feature items.

[0054] After obtaining the incremental characteristics of the rivet piercing stage, rivet expansion stage, and bottom forming stage, the changes between adjacent forming stages can be compared according to their chronological order. When determining the changes in each comparative feature between adjacent forming stages, the direction of change of various increments between the rivet piercing stage and the rivet expansion stage is compared according to the chronological order of the SPR riveting process, and the direction of change of various increments between the rivet expansion stage and the bottom forming stage is also compared. Specifically, for any comparative feature among acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance, the difference between the increment corresponding to the later forming stage and the increment corresponding to the previous forming stage is calculated. When the difference is greater than the corresponding preset change threshold, the direction of change of the comparative feature between two adjacent forming stages is determined to be increasing; when the difference is less than the negative value corresponding to the preset change threshold, the direction of change is determined to be decreasing; when the difference is between the two, the direction of change is determined to be maintaining.

[0055] Furthermore, the direction of change between the first stage (rivet piercing stage to rivet expansion stage) is first obtained, and then the direction of change between the second stage (rivet expansion stage to bottom forming stage) is obtained. For the same feature to be compared, when both the direction of change between the first and second stages increases or both decreases, it is determined that the feature to be compared has a unidirectional relationship of change in consecutive adjacent forming stages; when the two directions are different, or when the direction of change between any two stages remains unchanged, it is determined that the feature to be compared does not have a unidirectional relationship of change. Thus, based on the continuous change trend of the same type of micro-response feature in the rivet piercing, rivet expansion, and bottom forming processes, it can be determined whether the micro-response is continuously transmitted along the forming process, rather than judging solely based on the increase or decrease of a single forming stage relative to the background response. The degree of transmission can be determined based on the number of feature items with a unidirectional relationship of change between adjacent forming stages, the amplification of the corresponding increment in the subsequent forming stage, and the time interval between the two stages. The more feature terms exhibiting a unidirectional relationship, the more fully the micro-response continues across different data dimensions; the larger the increment in the subsequent forming stage, the more the micro-damage effect generated in the previous stage is amplified in the subsequent forming process; the shorter the time interval between adjacent stages, the higher the reliability of the response continuing from the previous stage to the next. Through the above processing, the degree of transmission from the rivet piercing stage to the rivet expansion stage, and the degree of transmission from the rivet expansion stage to the bottom forming stage can be obtained.

[0056] Specifically, determining the degree of transmission of the micro-damage response from the previous forming stage to the next forming stage based on the continuity relationship of the stage increment characteristics between adjacent forming stages includes: identifying the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance as comparison features in the stage increment characteristics; comparing the change direction of each comparison feature in the previous and next forming stages to determine the corresponding unidirectional change relationship; determining the response continuity intensity between adjacent forming stages based on the number of comparison features with unidirectional change relationships; determining the response amplification intensity between adjacent forming stages based on the increment amplitude of the comparison features with unidirectional change relationships in the next forming stage; determining the response attenuation intensity between adjacent forming stages based on the time interval between the end time of the previous forming stage and the start time of the next forming stage; and generating the degree of transmission of the micro-damage response from the previous forming stage to the next forming stage based on the response continuity intensity, response amplification intensity, and response attenuation intensity.

[0057] In practical implementation, after generating the incremental features corresponding to each forming stage, the incremental features of adjacent forming stages can be compared according to the sequence of the SPR riveting process. Adjacent forming stages can include the rivet piercing stage and the rivet expansion stage, as well as the rivet expansion stage and the bottom forming stage. By analyzing whether various micro-responses persist, further enhance, or decay between adjacent forming stages, it can be determined whether the micro-damage response generated in the previous forming stage has a lasting impact on the subsequent forming stage, and the corresponding degree of transmission can be determined accordingly.

[0058] Specifically, the incremental characteristics of each forming stage are broken down into acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance, and these four types of increments are identified as the characteristics to be compared. Among them, the acoustic emission burst increment mainly reflects the increase in transient energy release within the material relative to the background response; the vibration energy increment mainly reflects the enhancement of riveting impact and structural vibration relative to the background response; the vibration residual vibration retention mainly reflects the degree of persistence of the vibration response after the impact; and the equipment load disturbance mainly reflects the deviation of the riveting equipment's output load and propulsion state from the background operating state. By retaining these characteristics to be compared, the inter-stage continuity of the micro-damage response can be analyzed from different perspectives, such as material damage, mechanical impact, response persistence, and equipment load changes.

[0059] For any two adjacent forming stages, compare the direction of change of the same type of comparative feature item in the preceding and following forming stages. The direction of change can be determined based on the increase or decrease of each comparative feature item relative to the background response. When a comparative feature item increases relative to the background response in the preceding forming stage and continues to increase relative to the background response in the following forming stage, it can be determined that the comparative feature item has a trend of change in the same direction. When the corresponding increment in the following forming stage further increases, it can be determined that the comparative feature item not only has a continuity relationship but also an increasing trend.

[0060] For example, if the burst increment of acoustic emission is positive during the rivet piercing stage and also positive during the rivet expansion stage, it can be assumed that the burst acoustic emission response inside the material continues from the rivet piercing stage to the rivet expansion stage. When the residual vibration intensity increases during the rivet piercing stage and continues to increase during the rivet expansion stage, it can be assumed that the vibration influence generated during the piercing process does not completely disappear during the stage transition but persists throughout the rivet expansion process. When the equipment load disturbance increases in both the preceding and following forming stages, it can be assumed that the abnormal stress state formed in the preceding stage may continue to affect the rivet advancement or expansion process in the following stage.

[0061] To avoid small random fluctuations being mistaken for unidirectional changes, minimum effective increments can be set for different features to be compared. Only when the corresponding increments in both the previous and subsequent forming stages reach the corresponding minimum effective increments is a valid unidirectional change relationship determined for the feature to be compared. When the corresponding increment does not reach the minimum effective increment, it can be considered as background fluctuation or acquisition noise and not included in the calculation of subsequent response duration and amplification.

[0062] For comparison features that do not reach the minimum effective increment, a corresponding validity flag can be set. When the absolute value of the increment of a comparison feature in the previous or subsequent forming stage is less than the corresponding minimum effective increment, the validity flag of the comparison feature is set to invalid, and it is removed during the calculation of response duration and response amplification. Specifically, when calculating response duration, invalid features are neither included in the number of features with the same direction of change nor in the total number of valid comparison features; when calculating response amplification, the increment amplitude of invalid features in the subsequent forming stage is not read, nor is it included in the summation of the increment amplitudes of multiple features. For comparison features with a valid validity flag, the change direction judgment and increment amplitude comparison are still performed. By setting a validity flag, small fluctuations close to the background level can be prevented from being included in the stage transmission degree calculation, reducing the impact of acquisition noise and normal equipment vibration on the micro-damage response judgment.

[0063] After determining the unidirectional change relationship of each comparative feature, the number of valid comparative features and the number of comparative features with unidirectional change relationships can be counted, and the response continuity intensity can be determined based on the ratio of the two. Specifically, features that reach the minimum effective increment in both the previous and subsequent shaping stages are identified as valid comparative features, and those with unidirectional change relationships among the valid comparative features are identified as continuity features. The response continuity intensity can be determined as the ratio between the number of continuity features and the number of valid comparative features. For example, when all four comparative features are valid, and three of them have unidirectional change relationships, the corresponding response continuity intensity is 0.75. When there are no valid comparative features, the response continuity intensity is determined to be 0. Thus, the value of the response continuity intensity can be limited to between 0 and 1; the larger the value, the more data dimensions the micro-response in the previous shaping stage continues into the subsequent shaping stage. In another implementation, different continuity weights can be configured for each comparative feature based on the degree of correlation between different comparative features and micro-damage. For example, higher continuation weights can be assigned to the burst increment of acoustic emission and the residual vibration duration, while corresponding auxiliary weights can be assigned to the vibration energy increment and the equipment load disturbance. When a certain characteristic to be compared has a unidirectional relationship, its corresponding continuation weight is included in the response continuation intensity. In this way, the characterization effect of different types of micro-responses on stage transmission can be avoided by simply treating them as completely identical.

[0064] When determining the amplification intensity of the response, the incremental amplitude of each comparative feature item that exhibits a unidirectional change relationship in the subsequent forming stage can be read and compared with the corresponding incremental amplitude in the previous forming stage. When the incremental amplitude in the subsequent forming stage is basically close to that in the previous forming stage, it indicates that this type of micro-response mainly exhibits continuous maintenance; when the incremental amplitude in the subsequent forming stage is significantly greater than that in the previous forming stage, it indicates that the micro-damage effect generated in the previous forming stage is further enhanced by subsequent loads in the subsequent forming stage; when the incremental amplitude in the subsequent forming stage is less than that in the previous forming stage, but still higher than the background response level, it indicates that although this type of micro-response has undergone some attenuation, it still continues into the subsequent forming stage.

[0065] The incremental retention ratio or incremental amplification ratio corresponding to each feature to be compared can be determined separately, and the response amplification intensity can be determined based on the ratios corresponding to multiple feature items to be compared. For example, the degree of change in the incremental amplitude in the subsequent forming stage relative to the incremental amplitude in the previous forming stage can be calculated, and multiple feature items to be compared that have a unidirectional relationship can be summarized. When the incremental amplitudes of multiple feature items to be compared all increase significantly in the subsequent forming stage, the response amplification intensity is high; when the incremental amplitudes mainly remain stable, the response amplification intensity is at an intermediate level; when the incremental amplitudes decrease overall, the response amplification intensity is low.

[0066] Furthermore, the response attenuation intensity between adjacent forming stages is determined based on the time interval between the end of the previous forming stage and the start of the next forming stage. This time interval can be obtained from the time position of the corresponding stage boundary in the force-displacement data. When adjacent forming stages occur consecutively or the time interval is short, the acoustic emission, residual vibration, and load disturbance generated in the previous forming stage can quickly enter the next forming stage, resulting in less impact from natural response attenuation. In this case, a lower response attenuation intensity can be determined. When the time interval between two forming stages increases, the transient acoustic and vibration response generated in the previous stage may gradually weaken over time, allowing for a corresponding increase in the response attenuation intensity.

[0067] When determining the response attenuation intensity, the normal response attenuation pattern corresponding to the current riveting equipment and sheet metal combination can also be considered. Specifically, historical riveting data of normal riveting points can be used in advance to statistically analyze the natural attenuation range of acoustic emission and vibration responses at different time intervals. When the actual time interval is long, but the sudden increase in acoustic emission, vibration energy, or residual vibration in the subsequent forming stage remains at a high level, it can be considered that the response is not solely caused by the normal impact residue of the previous stage, but may be due to continuous micro-damage or abnormal stress. In this case, the weakening effect of the time interval on the transmission degree can be reduced.

[0068] Finally, the degree of transmission of the micro-damage response from the previous forming stage to the next forming stage is generated based on the response duration intensity, response amplification intensity, and response decay intensity. Specifically, the duration intensity is first used to determine whether the micro-response persists across multiple feature dimensions. Then, the response amplification intensity is used to determine whether the micro-response is further enhanced in the next forming stage. The natural decay effect caused by the time interval between stages is then corrected using the response decay intensity. Higher response duration intensity and higher response amplification intensity result in a greater degree of transmission; under the same conditions, a higher response decay intensity results in a correspondingly lower degree of transmission.

[0069] For example, when a significant increase in acoustic emission and vibration energy occurs during the rivet piercing stage, and these increases continue or further increase during the rivet expansion stage, while the residual vibration and equipment load disturbance also show a similar trend, it can be determined that the micro-damage response during the rivet piercing stage has a high degree of transmission to the rivet expansion stage. When multiple types of increases in the rivet expansion stage continue into the bottom forming stage, it can be further determined that the micro-damage response during the rivet expansion stage has a high degree of transmission to the bottom forming stage. This indicates that the abnormal response formed during riveting is not a single random impact, but may continue with the rivet's continued expansion and bottom locking.

[0070] Conversely, when a single acoustic emission burst or short-term vibration shock occurs in a certain forming stage, and the corresponding increment in the subsequent forming stage recovers to the background response level, and there is no unidirectional change in multiple types of comparative features, it can be determined that the response duration intensity is low and a low degree of transmission is generated.

[0071] Finally, according to the order of occurrence of the rivet piercing stage, rivet expansion stage, and bottom forming stage, the incremental features corresponding to each forming stage are arranged, and the corresponding transmission degree is added between adjacent stages to form a stage micro-response sequence. This stage micro-response sequence records not only the acoustic and vibration responses and equipment load changes of each forming stage, but also the continuation and amplification of micro-damage responses between different forming stages. This stage micro-response sequence can be used directly as riveting micro-response features, or it can be encoded or normalized for subsequent internal damage evidence generation.

[0072] Step 104 specifically involves extracting rivet shape features from the rivet visual shape data.

[0073] After riveting is completed, visual shape data of the area where the rivet point is located is collected. This visual shape data can be two-dimensional image data, three-dimensional contour data, or data formed by a two-dimensional image and a height contour. The collection range can cover the rivet head, the surrounding sheet metal area, and the boundary area between the upper and lower sheets. To improve the stability of subsequent feature extraction, brightness equalization, noise removal, edge enhancement, and rivet area localization processing can be performed on the rivet visual shape data to make the main rivet area, the circumferential deformation area, and the sheet metal bonding boundary clearer in the image.

[0074] When identifying the rivet center, the rivet head region or rivet indentation region can be segmented from the rivet visual shape data first. Then, the rivet center is determined based on the geometric center of the contour, the gray-scale centroid, or the height distribution center of this region. The rivet center is used to characterize the actual riveting position. By comparing the rivet center with the preset riveting position in the design documents, tooling positioning information, or visual calibration results, the offset of the rivet center relative to the preset riveting position can be obtained. The larger the offset, the higher the possibility of positioning deviation or tilting of the rivet during the riveting process. Therefore, a rivet offset feature can be generated based on this offset.

[0075] When identifying the outer contour of a rivet point, the contour can be extracted from the edge of the rivet head, the edge of the indentation, or the edge of the protrusion, thus obtaining the closed boundary of the rivet point's outer contour. Subsequently, using the rivet point center as a reference, the distance from the outer contour to the rivet point center can be calculated along multiple circumferential directions, or the contour radius, contour roundness, and contour eccentricity in different directions can be calculated. When the difference in contour distances in each direction is small, it indicates that the rivet point formation is relatively uniform; when the contour radius in a certain direction significantly increases or decreases, or the outer contour exhibits ellipticization, irregular concavity, or unilateral expansion, it indicates that the rivet point formation is uneven. Therefore, a forming uniformity feature can be generated based on the circumferential symmetry of the rivet point's outer contour.

[0076] When identifying the bonding boundary of boards, the bonding boundary can be determined based on the contact line between the upper and lower boards, edge shadows, changes in brightness or height of the gap, or abrupt changes in height. For two-dimensional images, the presence of local gaps can be determined by grayscale changes, shadow length, and edge continuity. For three-dimensional contour data, the bonding quality can be determined based on the height difference between the two sides of the bonding boundary, the gap width, and the length of the gap continuity. When dark bands, broken bright lines, discontinuous edges, or height gaps appear locally at the bonding boundary of the boards, it can be considered that there is a risk of non-bonding between the boards in the corresponding area, and bonding features between the boards can be generated accordingly.

[0077] When identifying the circumferential deformation area of ​​a rivet point, a ring-shaped or near-ring-shaped detection area can be set outside the outer contour of the rivet point, using the center of the rivet point as a reference. Within this area, changes in brightness, texture, indentation distribution, or height fluctuations can be analyzed. During normal riveting, the deformation of the sheet metal around the rivet point usually exhibits good circumferential consistency. However, when the rivet is misaligned, the sheet metal experiences uneven local stress, or there is local tearing, the circumferential area of ​​the rivet point may show abrupt changes in brightness on one side, deepened local indentations, abnormal protrusions, abnormal depressions, or discontinuous textures. Based on these abnormal changes in brightness or morphology, local deformation features can be generated.

[0078] After obtaining the rivet offset characteristics, forming uniformity characteristics, inter-plate bonding characteristics, and local deformation characteristics, these characteristics can be combined in a preset order to form the rivet morphology characteristics of the current rivet. The rivet morphology characteristics may include rivet position offset, contour symmetry, bonding boundary gap, and circumferential local deformation, etc., used to characterize the rivet quality from an appearance forming perspective. This rivet morphology characteristic can be fused with the aforementioned riveting micro-response characteristics, thus reflecting both the post-riveting appearance forming state and providing visual evidence for judging defects such as rivet misalignment, inter-plate non-bonding, and local deformation.

[0079] Step 105 specifically involves: integrating the riveting micro-response features and riveting point morphology features to determine the risk of SPR riveting defects in the battery casing, and generating corresponding riveting quality prediction results.

[0080] Using the aforementioned micro-response sequence as input for the riveting process and the morphological features of the rivet point as input for the post-riveting appearance, a fusion judgment is performed on the same rivet point. The riveting micro-response features mainly originate from the burst increment of acoustic emission, the increment of vibration energy, the amount of residual vibration, the amount of equipment load disturbance, and the degree of transmission between adjacent forming stages. These features can reflect whether internal micro-cracks, local tearing, abnormal stress, or forming stagnation occur during the riveting process. The morphological features of the rivet point mainly originate from rivet offset characteristics, forming uniformity characteristics, inter-plate bonding characteristics, and local deformation characteristics. These features can reflect the appearance and forming state of the rivet point surface and surrounding area after riveting.

[0081] Before fusion, the micro-response characteristics of the riveting can be mapped as evidence of internal damage. Specifically, based on the stage increment characteristics and transmission degree of different forming stages, it can be determined which type of internal anomaly the process signal is closer to. For example, when there are high burst increments of acoustic emission and residual vibration in both the rivet piercing stage and the rivet expansion stage, and the rivet expansion stage and the bottom forming stage, and the transmission degree is high, the micro-response characteristics of the riveting can be mapped as evidence of internal damage corresponding to a lower plate crack or local tear. When the vibration energy increment increases significantly in the rivet expansion stage, while the equipment load disturbance shows a single-stage anomaly, it can be mapped as evidence of internal process corresponding to rivet misalignment or obstructed expansion. When the equipment load disturbance is continuously abnormal in multiple stages, while the burst increment of acoustic emission is not significant, it can be mapped as evidence of internal process corresponding to non-adhesion between plates or abnormal clamping status.

[0082] Simultaneously, the morphological features of rivets can be mapped to evidence of appearance formation. Specifically, rivet offset features can be used to determine whether there is rivet skewing or abnormal positioning; uniformity of formation can be used to determine whether there is unilateral expansion or irregular deformation of the rivet outline; inter-plate bonding features can be used to determine whether there are local gaps between the upper and lower plates; and local deformation features can be used to determine whether there are abnormal indentations, abrupt changes in brightness or darkness, or local morphological abnormalities in the circumferential area of ​​the rivet. Thus, visual inspection results can be transformed from simple image features into evidence of appearance formation corresponding to the defect type.

[0083] Figure 3 This diagram illustrates the multi-dimensional evidence fusion and risk output provided in this application embodiment. In one implementation, internal anomaly judgment conditions can be set for internal damage evidence, and appearance anomaly judgment conditions can be set for appearance forming evidence. Internal anomaly judgment conditions may include conditions such as the stage increment feature reaching a preset threshold, the degree of transmission between adjacent forming stages reaching a preset threshold, and simultaneous anomalies in acoustic emission burst increment and vibration residual vibration retention. Appearance anomaly judgment conditions may include conditions such as the rivet center offset exceeding the allowable offset range, the circumferential symmetry of the rivet outer contour being lower than a preset value, the presence of local gaps at the sheet metal bonding boundary, and abnormal brightness changes in the circumferential deformation area of ​​the rivet.

[0084] When the evidence of internal damage does not meet the criteria for internal anomaly determination, and the evidence of external forming also does not meet the criteria for external anomaly determination, it indicates that the rivet did not exhibit obvious abnormal acoustic and vibration micro-responses or equipment load disturbances during the riveting process, and the appearance after riveting does not show obvious defects. Therefore, the rivet can be identified as a low-risk rivet, and a normal or low-risk riveting quality prediction result can be generated. For this type of rivet, it can be released directly on the production line, or only its rivet number, process characteristics, and morphological characteristics can be retained as quality traceability data.

[0085] When internal damage evidence points to a crack or local tear in the lower plate, but the external forming evidence does not meet the criteria for judging abnormal appearance, it indicates that the surface forming state of the rivet point may still be normal, but the sudden acoustic emission, residual vibration, or micro-damage transmission during the riveting process has already shown signs of internal damage. Since lower plate cracks and local tears may not be reflected on the rivet point surface in the early stages, this situation is easily missed by visual inspection alone. Therefore, this application increases the latent crack risk level of the corresponding rivet point in this situation and marks the riveting quality prediction result as latent crack risk or internal damage review risk, to prompt subsequent use of sectioning re-inspection, ultrasonic re-inspection, or process parameter review.

[0086] When internal damage evidence does not meet the criteria for internal anomaly determination, but external forming evidence points to rivet misalignment, misalignment between plates, or localized deformation, it indicates that the riveting process signals did not show obvious internal micro-damage, but the post-riveting visual morphology already shows an abnormal appearance. This situation may be caused by rivet positioning deviation, poor plate adhesion, abnormal clamping, or localized surface deformation. In this case, the corresponding rivet point can be identified as a risk of abnormal appearance, and the corresponding abnormal appearance type can be output in the riveting quality prediction results, enabling subsequent processing to focus on checking positioning accuracy, clamping status, or the surface forming quality of the rivet point.

[0087] When internal damage evidence and external forming evidence point to the same defect type, it indicates that the evidence from the riveting process and the post-riveting appearance corroborate each other. For example, internal damage evidence may show a significant skewed micro-response during the rivet expansion stage, while external forming evidence may show rivet center offset and unilateral contour deformation; or internal damage evidence may show equipment load disturbances corresponding to non-adhesion between plates, while external forming evidence may show local gaps at the plate bonding boundary. In such cases, the corresponding defect type can be identified as a high-confidence defect risk, and the defect type and risk level can be directly output in the riveting quality prediction results.

[0088] When internal damage evidence and external forming evidence point to different defect types, it indicates that there are inconsistent evidence sources for the same rivet point. For example, internal damage evidence may point to a crack in the lower plate, while external forming evidence may point to rivet misalignment; or internal damage evidence may point to a lack of fit between plates, while external forming evidence may show abnormal local deformation. This situation may indicate a compound defect at the rivet point, or it may indicate that the process signal is affected by sensor coupling, station vibration, or external obstruction. Therefore, the defect types pointed to by internal damage evidence and the defect types pointed to by external forming evidence can be jointly identified as defect risks to be reviewed, and both types of evidence sources should be recorded in the riveting quality prediction results for subsequent re-inspection or manual confirmation.

[0089] When generating riveting quality prediction results, the defect risk type, risk level, corresponding rivet number, forming stage that triggered the risk, main internal damage evidence, and main external forming evidence can be output together. The risk type can include low-risk rivets, latent crack risk, abnormal external forming risk, high-confidence defect risk, and defect risk requiring review. The risk level can be determined based on the strength of internal damage evidence, the strength of external forming evidence, and the degree of consistency between the two. The stronger and more consistent the internal and external evidence, the higher the risk level; when only a single piece of evidence is abnormal, a medium risk or a review risk can be determined by combining the defect type and the strength of the evidence.

[0090] As an implementable approach, after generating the corresponding riveting quality prediction result, the method further includes: associating and storing the riveting quality prediction result with the corresponding rivet number, forming stage, riveting micro-response characteristics, and rivet morphology characteristics; when the riveting quality prediction result indicates a risk of latent cracks, outputting prompts for section re-inspection, ultrasonic re-inspection, or process parameter verification; and updating the micro-response judgment threshold of the corresponding forming stage based on the crack confirmation result obtained from the re-inspection.

[0091] In practical implementation, after generating the riveting quality prediction result, this prediction result can be saved together with the basic information and characteristic information of the current riveting point to form a traceable riveting quality record. The riveting point number can be generated by the production line control system, vision positioning system, or battery casing design coordinates, and is used to uniquely identify a specific SPR riveting position on the battery casing. After storing the riveting quality prediction result in association with the riveting point number, the specific riveting point location with risk can be quickly determined during subsequent quality traceability, rework positioning, or spot check verification, avoiding obtaining only a rough result of whether the entire product is qualified or unqualified.

[0092] When storing related information, the forming stage that triggers the risk assessment can also be saved. By saving the forming stage information, it can be determined whether the anomaly occurred during sheet contact, rivet puncture, rivet expansion, or bottom forming, which is beneficial for subsequent analysis of the cause of the defect.

[0093] Simultaneously, the riveting micro-response characteristics and rivet point morphology characteristics can be associated and stored with the riveting quality prediction results. The riveting micro-response characteristics can include the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, equipment load disturbance, and the degree of transmission between adjacent forming stages at each forming stage. The rivet point morphology characteristics can include rivet offset characteristics, forming uniformity characteristics, inter-plate bonding characteristics, and local deformation characteristics.

[0094] When the riveting quality prediction result indicates a risk of latent cracks, it means that the rivet joint may not show obvious defects in its appearance, but the acoustic and vibration micro-response or equipment load disturbance during the riveting process has already shown signs of internal damage such as cracks or localized tears in the lower plate. For this type of rivet joint, prompts for section re-inspection, ultrasonic re-inspection, or process parameter verification can be output. Section re-inspection can be used to directly observe cracks, tears, locking amount, and plate deformation in the riveting section; ultrasonic re-inspection can be used to detect internal cracks or abnormal adhesion in the lower plate without damaging the product; process parameter verification can be used to check whether the riveting pressure, pressure head stroke, rivet specifications, mold status, and clamping status deviate from the set range.

[0095] When issuing re-inspection prompts, the re-inspection method can be determined based on the risk level of latent cracks. For example, when there is strong evidence of internal damage and a high degree of transfer across multiple forming stages, a sectioning re-inspection prompt can be issued first to confirm the presence of actual cracks. When there is strong evidence of internal damage but the product is not suitable for destructive testing, an ultrasonic re-inspection prompt can be issued. When the risk level of latent cracks is low and the main anomaly comes from equipment load disturbances, a process parameter verification prompt can be issued first to check for process-related causes such as loose fixtures, abnormal pressure head speed, or mold wear.

[0096] After the re-inspection is completed, the crack confirmation results obtained from the re-inspection can be returned to the system. The crack confirmation results may include confirmation of crack presence, absence of cracks, presence of localized tears, lack of bonding between plates, or uncertainty in the re-inspection results. The system can then re-associate and store the crack confirmation results with the corresponding rivet number, the original riveting quality prediction results, riveting micro-response characteristics, and rivet morphology characteristics, thereby forming a complete record from the prediction results to the re-inspection results. This record can be used not only for single-piece product quality traceability but also for quality statistical analysis under the same batch, the same equipment, or the same process parameters.

[0097] Based on the crack confirmation results obtained from the re-inspection, the micro-response judgment threshold for the corresponding forming stage can be updated. Specifically, when multiple rivet points that were judged to be at risk of latent cracks are confirmed to have cracks after re-inspection, the sudden increase in acoustic emission, the amount of residual vibration, and the degree of transmission of these rivet points in the rivet piercing stage, rivet expansion stage, or bottom forming stage can be statistically analyzed. The risk trigger threshold for the corresponding stage can be appropriately reduced or the weight of such micro-response characteristics can be increased, so that subsequent similar crack risks can be identified earlier.

[0098] During the threshold update process, adjustments can be made separately for different forming stages. For example, if the crack confirmed by re-inspection is mainly related to the residual vibration retention amount during the rivet expansion stage, then the threshold for the residual vibration retention amount corresponding to the rivet expansion stage should be updated accordingly. If the crack is mainly related to the equipment load disturbance amount and the stage transmission degree during the bottom forming stage, then the threshold for the equipment load disturbance amount and the transmission degree threshold during the bottom forming stage should be updated accordingly. By updating the threshold separately according to the forming stage, the problem of overly coarse judgment caused by using a uniform threshold for all stages can be avoided.

[0099] To further illustrate the effects of this application, a specific embodiment is tested.

[0100] The upper and lower aluminum alloy plates of new energy battery casings were selected as the objects to be riveted, and continuous riveting tests were conducted using a servo SPR riveting device. During the riveting process at each riveting point, force-displacement data, acoustic emission data, vibration acceleration data, motor current, and indenter displacement velocity were collected simultaneously. After riveting, visual morphology data of the riveting points were collected using an industrial camera. The system first identifies the plate contact stage, rivet piercing stage, rivet expansion stage, and bottom forming stage based on the force-displacement data, and then matches the acoustic and vibration response data and equipment operation data to the corresponding stages. Subsequently, the system generates the bottom response characteristics based on the plate contact stage, and calculates the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention, and equipment load disturbance for each subsequent forming stage, forming a stage micro-response sequence. At the same time, rivet offset characteristics, forming uniformity characteristics, plate bonding characteristics, and local deformation characteristics are extracted from the visual morphology data. Finally, the riveting micro-response characteristics are fused with the riveting point morphology characteristics to output the riveting quality prediction results.

[0101] To verify the effectiveness of the proposed solution, 400 rivet points were selected from four categories: normal riveting, lower plate cracks, rivet misalignment, and misalignment between plates, with 100 rivet points in each category. The results of cross-sectional inspection and ultrasonic testing were used as controls. Test results showed that the proposed solution achieved an accuracy rate of 97.0% for normal rivet points, 94.0% for lower plate crack rivet points, 95.0% for rivet misalignment rivet points, and 93.0% for misalignment between plates, with an overall accuracy rate of 94.75%. Specifically, for samples with latent lower plate cracks where no obvious visual abnormalities were observed after riveting, the detection rate using visual inspection alone was 68.0%, which increased to 92.0% after using the multi-dimensional fusion prediction method of this application.

[0102] According to another embodiment, a multidimensional fusion prediction system for battery casing SPR riveting quality is provided. Figure 4 A schematic block diagram of a multidimensional fusion prediction system for the SPR riveting quality of a battery casing, according to one embodiment, is shown. Figure 4 As shown, the system includes: The data acquisition module 401 is used to acquire force displacement data, acoustic vibration response data and equipment operation data during the SPR riveting process of the battery casing, and to acquire visual morphology data of the rivet points after the riveting is completed. The stage matching module 402 is used to determine multiple forming stages in the SPR riveting process based on the force displacement data, and to perform stage matching on the acoustic vibration response data and equipment operation data. The micro-response feature extraction module 403 is used to extract riveting micro-response features from the acoustic and vibration response data and equipment operation data corresponding to each forming stage; The morphology feature extraction module 404 is used to extract morphology features of rivets from rivet visual morphology data; The fusion prediction module 405 is used to fuse the riveting micro-response features and the riveting point morphology features to determine the risk of SPR riveting defects in the battery casing and generate corresponding riveting quality prediction results.

[0103] As an feasible solution, the acoustic and vibration response data acquired by the data acquisition module 401 includes acoustic emission data and vibration acceleration data collected during the riveting process. The equipment operation data includes at least two of the following: motor current, spindle speed, pressure head displacement speed, and riveting contact triggering moment of the riveting equipment.

[0104] As an feasible solution, the stage matching module 402, when determining multiple forming stages in the SPR riveting process based on force-displacement data, includes: identifying the initial rise point of force value, the abrupt change point of force value slope, the stable point of displacement propagation, and the final pressure forming point from the force-displacement data; and using the initial rise point of force value, the abrupt change point of force value slope, the stable point of displacement propagation, and the final pressure forming point as stage boundaries, dividing the SPR riveting process into the sheet metal contact stage, the rivet piercing stage, the rivet expansion stage, and the bottom forming stage.

[0105] As an feasible solution, the stage matching module 402, when performing stage matching on the acoustic vibration response data and equipment operation data, includes: using the riveting contact triggering moment as the initial time reference, determining the time window corresponding to each forming stage based on the stage boundaries in the force displacement data; calculating the force value change rate sequence based on the force displacement data, and determining the peak moment of the force value change rate from the force value change rate sequence; extracting the vibration impact amplitude sequence from the acoustic vibration response data, and determining the vibration impact peak moment from the vibration impact amplitude sequence; performing time delay compensation on the acoustic vibration response data based on the time deviation between the peak moment of the force value change rate and the peak moment of the vibration impact; and extracting the time delay-compensated acoustic vibration response data and equipment operation data to the corresponding forming stages.

[0106] As an feasible solution, the micro-response feature extraction module 403 extracts riveting micro-response features from the acoustic and vibration response data and equipment operation data corresponding to each forming stage, including: generating the background response features of the current riveting point using the acoustic and vibration response data and equipment operation data corresponding to the plate contact stage; determining the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention amount, and equipment load disturbance amount relative to the background response features for the rivet piercing stage, rivet expansion stage, and bottom forming stage; combining the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention amount, and equipment load disturbance amount corresponding to the same forming stage into the stage increment feature of that forming stage; determining the degree of transmission of the micro-damage response of the previous forming stage to the next forming stage based on the continuity relationship of the stage increment features between adjacent forming stages; generating a stage micro-response sequence based on the stage increment features corresponding to each forming stage and the degree of transmission between adjacent forming stages, and using the stage micro-response sequence as the riveting micro-response feature.

[0107] As an implementable solution, the micro-response feature extraction module 403, when determining the degree of transmission of the micro-damage response of the previous forming stage to the next forming stage based on the continuity relationship of the stage increment features between adjacent forming stages, includes: determining the acoustic emission burst increment, vibration energy increment, vibration residual vibration retention amount, and equipment load disturbance amount in the stage increment features as comparison features; comparing the change direction of each comparison feature in the previous forming stage and the next forming stage to determine the same-direction change relationship corresponding to each comparison feature; determining the response continuity intensity between adjacent forming stages based on the number of comparison features with the same-direction change relationship; determining the response amplification intensity between adjacent forming stages based on the increment amplitude of the comparison features with the same-direction change relationship in the next forming stage; determining the response attenuation intensity between adjacent forming stages based on the time interval between the end time of the previous forming stage and the start time of the next forming stage; and generating the degree of transmission of the micro-damage response of the previous forming stage to the next forming stage based on the response continuity intensity, response amplification intensity, and response attenuation intensity.

[0108] As an feasible solution, the fusion prediction module 405, when extracting rivet shape features from the rivet visual shape data, includes: identifying the rivet center, rivet outer contour, sheet metal bonding boundary, and rivet circumferential deformation area from the rivet visual shape data; generating rivet offset features based on the offset between the rivet center and a preset riveting position; generating forming uniformity features based on the circumferential symmetry of the rivet outer contour; generating inter-plate bonding features and local deformation features based on the local gap of the sheet metal bonding boundary and the abnormal brightness and darkness changes of the rivet circumferential deformation area; and determining the rivet shape features based on the rivet offset features, the forming uniformity features, the generated inter-plate bonding features, and the local deformation features.

[0109] As an feasible solution, the fusion prediction module 405, when determining the risk of SPR riveting defects in the battery casing by fusing the riveting micro-response features and the riveting point morphology features, includes: mapping the riveting micro-response features to internal damage evidence and mapping the riveting point morphology features to appearance forming evidence; when the internal damage evidence does not meet the internal anomaly judgment criteria and the appearance forming evidence does not meet the appearance anomaly judgment criteria, determining the corresponding riveting point as a low-risk riveting point; when the internal damage evidence points to a lower plate crack or local tear and the appearance forming evidence does not meet the appearance anomaly judgment criteria. When the internal damage evidence does not meet the internal anomaly judgment criteria, and the external forming evidence points to rivet misalignment, non-fitting between plates, or local deformation, the corresponding rivet is determined to be an external forming anomaly risk. When the internal damage evidence and the external forming evidence point to the same defect type, the corresponding defect type is determined to be a high-confidence defect risk. When the internal damage evidence and the external forming evidence point to different defect types, the defect type pointed to by the internal damage evidence and the defect type pointed to by the external forming evidence are jointly determined as a defect risk to be reviewed.

[0110] As an feasible solution, after the fusion prediction module 405 generates the corresponding riveting quality prediction result, it further includes: associating and storing the riveting quality prediction result with the corresponding rivet number, forming stage, riveting micro-response characteristics and rivet morphology characteristics; when the riveting quality prediction result indicates a risk of latent cracks, outputting prompts for section re-inspection, ultrasonic re-inspection or process parameter verification; and updating the micro-response judgment threshold of the corresponding forming stage based on the crack confirmation result obtained from the re-inspection.

[0111] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-dimensional fusion prediction method for the SPR riveting quality of battery casings, characterized in that, include: Acquire force displacement data, acoustic vibration response data, and equipment operation data during the SPR riveting process of the battery casing, and acquire visual morphology data of the rivet points after the riveting is completed; Identify the initial rise point of force value, the point of sudden change in force value slope, the point of stable displacement propulsion, and the point of final compression forming from the force-displacement data; Using the initial rise point of force value, the sudden change point of force value slope, the stable point of displacement propulsion and the final pressure forming point as stage boundaries, the SPR riveting process is divided into multiple forming stages, which include: plate contact stage, rivet piercing stage, rivet expansion stage and bottom forming stage. Using the moment of rivet contact as the initial time reference, the time window corresponding to each forming stage is determined based on the stage boundary in the force-displacement data. Calculate the force value change rate sequence based on the force displacement data, and determine the peak time of the force value change rate from the force value change rate sequence; The vibration impact amplitude sequence is extracted from the acoustic-vibration response data, and the peak vibration impact time is determined from the vibration impact amplitude sequence; The acoustic-vibration response data is time-delayed based on the time deviation between the peak time of the force change rate and the peak time of the vibration impact. The acoustic and vibration response data after time delay compensation and the equipment operation data are respectively extracted to the corresponding forming stages; Extract riveting micro-response features from acoustic and vibration response data and equipment operation data corresponding to each forming stage; Extract rivet shape features from the rivet visual shape data; By integrating the micro-response characteristics and morphological characteristics of the riveting joint, the risk of SPR riveting defects in the battery casing is determined, and the corresponding riveting quality prediction results are generated.

2. The method according to claim 1, characterized in that, The acoustic and vibration response data includes acoustic emission data and vibration acceleration data collected during the riveting process, and the equipment operation data includes at least two of the following: motor current, spindle speed, pressure head displacement speed, and riveting contact trigger moment of the riveting equipment.

3. The method according to claim 1, characterized in that, The extraction of riveting micro-response features from the acoustic and vibration response data and equipment operation data corresponding to each forming stage includes: The background response characteristics of the current rivet point are generated using the acoustic and vibration response data corresponding to the plate contact stage and the equipment operation data. The acoustic emission burst increment, vibration energy increment, vibration residual vibration retention amount, and equipment load disturbance amount relative to the background response characteristics are determined for the rivet piercing stage, rivet expansion stage, and bottom forming stage, respectively. The acoustic emission burst increment, vibration energy increment, vibration residual vibration retention amount, and equipment load disturbance amount corresponding to the same forming stage are combined to form the stage increment characteristics of that forming stage. Based on the continuity relationship of the incremental characteristics between adjacent forming stages, the degree of transmission of the micro-damage response of the previous forming stage to the next forming stage is determined. Based on the stage increment characteristics corresponding to each forming stage and the degree of transmission between adjacent forming stages, a stage micro-response sequence is generated, and the stage micro-response sequence is used as the riveting micro-response characteristic.

4. The method according to claim 3, characterized in that, The determination of the degree of transmission of the micro-damage response from the previous forming stage to the next forming stage based on the continuity relationship of the incremental characteristics between adjacent forming stages includes: The acoustic emission burst increment, vibration energy increment, vibration residual vibration retention amount, and equipment load disturbance amount in the aforementioned stage incremental characteristics are respectively determined as the feature items to be compared; By comparing the direction of change of each feature item to be compared in the previous forming stage and the next forming stage, the same direction of change relationship corresponding to each feature item to be compared is determined. The response continuity intensity between adjacent forming stages is determined based on the number of comparable feature terms that exhibit a unidirectional change relationship. Based on the incremental magnitude of the comparative feature terms that have a unidirectional change relationship in the subsequent forming stage, the response amplification intensity between adjacent forming stages is determined. The response attenuation intensity between adjacent forming stages is determined based on the time interval between the end of the previous forming stage and the start of the next forming stage. Based on the response duration intensity, response amplification intensity, and response decay intensity, the degree of transmission of the micro-damage response from the previous forming stage to the next forming stage is generated.

5. The method according to claim 1, characterized in that, The step of extracting rivet shape features from the rivet visual shape data includes: Identify the rivet center, rivet outer contour, sheet metal bonding boundary, and circumferential deformation area of ​​the rivet from the rivet visual morphology data. A rivet offset feature is generated based on the offset between the rivet point center and the preset riveting position; The forming uniformity feature is generated based on the circumferential symmetry of the outer contour of the rivet point; Based on the local gaps at the bonding boundaries of the plates and the abnormal brightness variations in the circumferential deformation areas of the rivet points, inter-plate bonding features and local deformation features are generated. The morphological features of the rivet are determined based on the rivet offset features, the forming uniformity features, the bonding features between the generated plates, and the local deformation features.

6. The method according to claim 1, characterized in that, The method of integrating the micro-response characteristics and morphological characteristics of the riveting joint to determine the SPR riveting defect risk of the battery casing includes: The riveting micro-response features are mapped as evidence of internal damage, and the rivet morphology features are mapped as evidence of external formation. When the evidence of internal damage does not meet the criteria for internal anomaly determination, and the evidence of external forming does not meet the criteria for external anomaly determination, the corresponding rivet point is determined to be a low-risk rivet point. When the internal damage evidence points to a crack or local tear in the lower plate, and the external forming evidence does not meet the criteria for judging abnormal appearance, the risk level of the hidden crack at the corresponding rivet point is increased. When the evidence of internal damage does not meet the criteria for internal anomaly determination, and the evidence of external forming points to rivet misalignment, non-fitting between plates, or local deformation, the corresponding rivet point is determined to be a risk of external forming anomaly. When the evidence of internal damage and the evidence of external formation point to the same defect type, the corresponding defect type is determined to be a high-confidence defect risk. When the evidence of internal damage and the evidence of external shaping point to different defect types, the defect type pointed to by the evidence of internal damage and the defect type pointed to by the evidence of external shaping are jointly identified as the defect risk to be reviewed.

7. The method according to claim 1, characterized in that, After generating the corresponding riveting quality prediction result, the method further includes: The riveting quality prediction results are associated and stored with the corresponding rivet number, forming stage, riveting micro-response characteristics, and rivet morphology characteristics. When the riveting quality prediction result indicates a risk of hidden cracks, a prompt for section re-inspection, ultrasonic re-inspection, or process parameter verification will be output. The micro-response judgment threshold for the corresponding forming stage is updated based on the crack confirmation results obtained from the re-inspection.

8. A multi-dimensional fusion prediction system for the SPR riveting quality of battery casings, characterized in that, include: The data acquisition module is used to acquire force displacement data, acoustic vibration response data and equipment operation data during the SPR riveting process of the battery casing, and to acquire visual morphology data of the rivet points after the riveting is completed. The stage matching module is used to identify the initial rise point of force value, the abrupt change point of force value slope, the stable point of displacement advancement, and the final pressure forming point from the force-displacement data; using the initial rise point of force value, the abrupt change point of force value slope, the stable point of displacement advancement, and the final pressure forming point as stage boundaries, the SPR riveting process is divided into multiple forming stages, including: sheet metal contact stage, rivet piercing stage, rivet expansion stage, and bottom forming stage; using the rivet contact triggering moment as the initial time reference, the time window corresponding to each forming stage is determined according to the stage boundaries in the force-displacement data; the force value change rate sequence is calculated based on the force-displacement data, and the peak moment of the force value change rate is determined from the force value change rate sequence; the vibration impact amplitude sequence is extracted from the acoustic vibration response data, and the peak moment of the vibration impact is determined from the vibration impact amplitude sequence; the acoustic vibration response data is time-delay compensated based on the time deviation between the peak moment of the force value change rate and the peak moment of the vibration impact; the time-delay compensated acoustic vibration response data and equipment operation data are respectively extracted to the corresponding forming stages. The micro-response feature extraction module is used to extract riveting micro-response features from the acoustic and vibration response data and equipment operation data corresponding to each forming stage; The morphology feature extraction module is used to extract morphology features of the rivet points from the visual morphology data of the rivet points; The fusion prediction module is used to fuse the riveting micro-response features and the riveting point morphology features to determine the risk of SPR riveting defects in the battery casing and generate corresponding riveting quality prediction results.

Images