A riveting process-based quality anomaly prediction and traceability method

Abstract

The application discloses a riveting process-based quality anomaly prediction and tracing method, and particularly relates to the field of data processing and quality prediction and tracing of a riveting device, and comprises the following steps: acquiring force data, displacement data and time data collected by the riveting device in a riveting process; and constructing a riveting process curve corresponding to the riveting process based on the collected data; binding the riveting process curve with the equipment identification corresponding to the riveting process, the process parameter version, the die service life, the material batch and the time stamp, and generating a riveting event corresponding to the riveting process. The riveting process curve is systematically bound with context information such as the equipment state, the die service life, the process parameter version and the material batch, and mutation monitoring, risk prediction and source tracing analysis are performed on the fingerprint evolution of the continuous riveting process.

Description

Technical Field

[0001] This invention relates to the field of data processing and quality prediction and traceability technology for riveting equipment, and more specifically, to a method for predicting and tracing quality anomalies in the riveting process. Background Technology

[0002] Typical intelligent riveting equipment employs two main forms: C-type and four-column. The overall structure comprises a frame and guiding structure, a power actuation structure, a pressure head and die structure, a tooling fixture structure, and a control and detection structure. The C-type riveting equipment uses a C-shaped frame with a single-sided opening to create the working space, while the four-column riveting equipment uses four columns to form a closed guiding frame, ensuring the parallelism and rigidity of the slider movement. The power actuation structure typically uses hydraulic cylinders or servo electric cylinders, which, driven by the control system, drive the slider and pressure head to reciprocate axially along the guiding structure. The pressure head and die directly act on the rivet and the connected component. The fixture structure is used for workpiece positioning, support, and limiting. The control and detection structure uses a programmable logic controller (PLC) or industrial computer as its core, in conjunction with pressure or force sensors, displacement sensors, and a time sampling unit to collect and control the force, displacement, and cycle time during the riveting process.

[0003] During operation, the equipment typically goes through a series of stages, including workpiece clamping and positioning, rapid downward movement of the pressure head, contact loading, material plastic forming, final pressing and shaping, and return unloading. In this process, a force-displacement-time response curve with a specific shape is formed. Existing technologies mostly rely on result indicators such as peak force, end displacement, or whether the set stroke has been reached to determine whether a single riveting process is qualified.

[0004] As equipment operates continuously for extended periods, the mold and pressure head gradually wear down. The friction state of the guide pair and the temperature rise of the hydraulic system constantly change. Different batches of rivets and the yield characteristics of the connected materials exhibit objective differences. Simultaneously, process parameters may undergo multiple version changes during machine adjustment or model changeover. Under the combined effect of these factors, even if the endpoint indicators of a single riveting process still meet the set threshold, the shape of the middle and early sections of the force-displacement curve, the position of the yield inflection point, and the slope of the forming stage often undergo systematic shifts. This can lead to hidden quality problems such as insufficient connection strength and reduced fatigue life in subsequent assembly stress or service conditions. However, under current technological conditions, riveting process data is usually only used for single-time judgments or simple information retention. It does not establish a calculable and correlated binding relationship with key contexts such as equipment operating status, mold life stage, process parameter version, and material batch. When batch anomalies or failures occur, it is often only possible to trace back to a certain time interval or workstation range. It is difficult to locate the root cause of the initial shift among multiple sources of changing factors, leading to repeated troubleshooting processes that rely on trial and error based on experience, downtime for mold replacement, or parameter rollback.

[0005] Therefore, it can be concluded that the core problem of the existing technology is that in the continuous riveting process of intelligent riveting equipment, there is a lack of a traceability mechanism that can systematically bind the single riveting process curve with multi-source information such as equipment status, mold life, process parameter version and material batch and support calculation and analysis, making it difficult to locate the source of quality abnormalities. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for predicting and tracing quality anomalies in the riveting process. This method systematically binds the riveting process curve with contextual information such as equipment status, mold life, process parameter version, and material batch, and performs mutation monitoring, risk prediction, and source tracing analysis on the fingerprint evolution of the continuous riveting process, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting and tracing quality anomalies in the riveting process, comprising:

[0008] S1. Obtain force data, displacement data and time data collected by the riveting equipment during a riveting process, and construct the riveting process curve corresponding to the riveting based on the collected data; bind the riveting process curve with the equipment identifier, process parameter version, mold life, material batch and timestamp corresponding to the riveting, and generate the riveting event corresponding to the riveting.

[0009] S2. Perform process decomposition calculation on the riveting process curve in the riveting event, decomposing the riveting process curve into material fingerprint and equipment fingerprint; the material fingerprint is used to characterize the material forming characteristics during the riveting process, and the equipment fingerprint is used to characterize the structural response and driving characteristics of the riveting equipment; and use the material fingerprint and equipment fingerprint as the process fingerprint of the riveting event;

[0010] S3. According to the time sequence of riveting occurrence, the currently generated riveting event is appended to the event sequence; an event summary is generated based on the summary value of the previous riveting event in the event sequence and the content of the current riveting event, and the event summary is recorded in the current riveting event, so that an event chain with an associated verification relationship is formed between adjacent riveting events;

[0011] S4. Based on the process fingerprints of each riveting event in the event chain, perform similarity calculations on the material fingerprints and equipment fingerprints respectively, and construct a mutation index with the material fingerprints and equipment fingerprints as index keys; and use the mutation index to perform continuity monitoring on the process fingerprints of adjacent riveting events in the event chain to determine whether there are structural mutations in the process fingerprints.

[0012] In a preferred embodiment, S5 is also included:

[0013] S5-1. When the mutation index detects a structural mutation of the process fingerprint in the event chain or the offset between process fingerprints is greater than the preset offset threshold, the corresponding riveting event is determined as the target riveting event.

[0014] In the event chain, a preset number of historical riveting events are selected in chronological order, with the target riveting event as the endpoint, to form a historical event set. For each historical riveting event in the historical event set, its material fingerprint and equipment fingerprint are read. Using the material fingerprint and equipment fingerprint corresponding to the target riveting event as the reference fingerprint, the material fingerprint of each historical riveting event is subjected to a term-by-term difference operation to obtain a material difference sequence arranged in chronological order. The equipment fingerprint of each historical riveting event is then subjected to a term-by-term difference operation to obtain an equipment difference sequence arranged in chronological order.

[0015] S5-2. Calculate the increment of change between adjacent terms in the material difference sequence in chronological order to form a material change increment sequence; calculate the average absolute value of the material change increment sequence as the material offset amplitude, and calculate the difference between the first and last terms as the material offset trend; calculate the increment of change between adjacent terms in the equipment difference sequence in chronological order to form an equipment change increment sequence; calculate the average absolute value of the equipment change increment sequence as the equipment offset amplitude, and calculate the difference between the first and last terms as the equipment offset trend; combine the material offset amplitude, material offset trend, equipment offset amplitude, and equipment offset trend to form an offset evolution set;

[0016] S5-3. Compare the material offset amplitude in the offset evolution set with a first amplitude threshold, and generate a material amplitude anomaly flag when it exceeds the first amplitude threshold; compare the material offset trend with a first trend threshold, and generate a material trend anomaly flag when it exceeds the first trend threshold;

[0017] The device offset amplitude is compared with a second amplitude threshold, and a device amplitude anomaly flag is generated when it exceeds the second amplitude threshold; the device offset trend is compared with a second trend threshold, and a device trend anomaly flag is generated when it exceeds the second trend threshold.

[0018] A combined judgment is performed on the material amplitude abnormality flag, material trend abnormality flag, equipment amplitude abnormality flag and equipment trend abnormality flag. When the preset combination conditions are met, the quality abnormality prediction risk of the subsequent riveting process is generated as the trigger state, and the target riveting event is determined as the prediction starting point event.

[0019] S5-4. Starting from the predicted starting event, select backtracking riveting events one by one in reverse chronological order in the event chain; for each backtracking riveting event, repeat the fingerprint differential sequence construction in S5-1 and the offset evolution calculation in S5-2, and regenerate the corresponding predicted risk according to the predicted risk generation rule in S5-3.

[0020] When a retrospective riveting event with the predicted risk triggered for the first time occurs, the retrospective riveting event is identified as the source riveting event; and the equipment identifier, process parameter version, mold life and material batch bound in the source riveting event are read as the traceability result corresponding to the quality anomaly prediction risk output.

[0021] In a preferred embodiment, S1 includes:

[0022] S1-1. During a riveting process, force data, displacement data and time data are synchronously sampled according to a unified sampling clock. The collected force data and displacement data are respectively appended with corresponding time indices to form a force sampling sequence and a displacement sampling sequence that correspond one-to-one according to the time index.

[0023] S1-2. Perform threshold start determination on the force sampling sequence to determine the riveting contact start point, and perform slope change determination on the displacement sampling sequence to determine the forming termination point; using the riveting contact start point and the forming termination point as boundaries, extract the corresponding effective sampling segments from the force sampling sequence and the displacement sampling sequence;

[0024] S1-3. The force sampling sequence and displacement sampling sequence within the effective sampling segment are paired one-to-one according to the time index to form a force-displacement data pair sequence driven by the time index, and the force-displacement data pair sequence is used as the riveting process curve corresponding to this riveting.

[0025] S1-4. Bind the riveting process curve with the corresponding equipment identifier, process parameter version, mold life, material batch and timestamp to generate the riveting event corresponding to the riveting.

[0026] In a preferred embodiment, S2 includes:

[0027] S2-1. Obtain the corresponding riveting process curve in the riveting event. The curve segment between the riveting contact start point and the forming termination point determined in S1-2 is defined as the loading segment, and the curve segment after the forming termination point until the force value drops to the preset unloading threshold is defined as the return segment.

[0028] S2-2. The force-displacement data pairs in the return section are rearranged in reverse order according to the time index, and the displacement values ​​after reversal are symbolically mapped in a manner consistent with the direction of displacement change in the loading section, so as to obtain the return mapping sequence aligned with the displacement direction of the loading section.

[0029] S2-3. Align the force-displacement data pair sequence within the loading section with the return mapping sequence point by point according to the time index, and take the average of the corresponding force values ​​under the same time index to form a device projection curve, and determine the device projection curve as the device fingerprint of the riveting event.

[0030] S2-4. Using the device fingerprint as a reference, perform force value difference calculation on the riveting process curve in the loading section point by point according to the time index to form a material residual curve, and determine the material residual curve as the material fingerprint of the riveting event.

[0031] S2-5. Use the material fingerprint and the equipment fingerprint as the process fingerprint of the riveting event.

[0032] In a preferred embodiment, S3 includes:

[0033] S3-1. According to the time sequence of riveting, write the currently generated riveting event to the end of the event sequence, and record the event number of the current riveting event in the event sequence;

[0034] S3-2. Extract the summary of the equipment identifier, process parameter version, mold life, material batch, timestamp and riveting process curve bound in the current riveting event in sequence, and splice them in the preset order to generate the current content summary;

[0035] S3-3. If the current riveting event is the first event, then use the current content summary as the basic input string to perform a summary operation to generate a basic event summary; otherwise, read the event summary recorded in the previous riveting event in the event sequence as the preceding summary, concatenate the preceding summary with the current content summary to form a basic input string, and perform a summary operation on the basic input string to generate a basic event summary.

[0036] In a preferred embodiment, S3 further includes:

[0037] S3-4. Based on the event sequence number of the current riveting event, determine whether the preset hierarchical conditions are met. If met, backtrack in the event sequence to select a historical riveting event with an interval of a preset step size from the event sequence number, and read its event summary as the hierarchical pre-sequence summary. Concatenate the hierarchical pre-sequence summary with the basic event summary to form a hierarchical input string, and perform a summary operation on the hierarchical input string to generate a hierarchical event summary. Otherwise, use the basic event summary directly as the hierarchical event summary.

[0038] S3-5. Select a preset number of historical riveting events in the event sequence, with the current riveting event as the endpoint, read their current content summaries in sequence and concatenate them in chronological order to form a segment input string; concatenate the segment input string with the hierarchical event summary to form a fusion input string, and perform a summary operation on the fusion input string to generate the current event summary;

[0039] S3-6. Write the current event summary and the previous summary into the current riveting event; then select a preset number of riveting events, including the current riveting event, from the event sequence, and repeat the calculations of S3-3 to S3-5 for each selected riveting event to generate a recalculated event summary; if any recalculated event summary is inconsistent with the event summary recorded in the corresponding riveting event, the event chain is determined to be invalid at the riveting event and marked as a chain abnormal event; otherwise, the event chain is maintained as valid.

[0040] In a preferred embodiment, S4 includes:

[0041] S4-1. In the event chain, select two adjacent riveting events sequentially according to time order. Read the material fingerprint and equipment fingerprint of the previous riveting event and the material fingerprint and equipment fingerprint of the subsequent riveting event respectively. Subtract the corresponding items of the material fingerprint of the previous riveting event and the material fingerprint of the subsequent riveting event to obtain a material difference sequence. Calculate the average of the absolute values ​​of the material difference sequence as the material similarity value. Subtract the corresponding items of the equipment fingerprint of the previous riveting event and the equipment fingerprint of the subsequent riveting event to obtain a device difference sequence. Calculate the average of the absolute values ​​of the device difference sequence as the device similarity value. Register the material similarity value and the device similarity value as similarity records for corresponding adjacent riveting event pairs.

[0042] S4-2. Using the similarity record as an index entry, write the event number, material similarity value and equipment similarity value of adjacent riveting event pairs into the mutation index, so that the mutation index can be searched according to the time sequence corresponding to the event number, and located and searched according to the material similarity value and equipment similarity value.

[0043] In a preferred embodiment, S4 further includes:

[0044] S4-3. Traverse the similarity records in the mutation index in sequence along the event number, compare the material similarity value with the first preset mutation threshold, and compare the equipment similarity value with the second preset mutation threshold; if the material similarity value is greater than the first preset mutation threshold or the equipment similarity value is greater than the second preset mutation threshold, it is determined that there is a structural mutation in the process fingerprint of the corresponding adjacent riveting event, and the subsequent riveting event is determined as a mutation event; otherwise, it is determined that the process fingerprint of the corresponding adjacent riveting event remains continuous and the mutation index remains unchanged.

[0045] The technical effects and advantages of this invention are as follows:

[0046] This invention systematically binds the riveting process curve with equipment status, mold life, process parameter version, and material batch, and constructs an event chain and mutation index to locate the source of systematic deviation of curve shape during continuous riveting. This solves the problem that existing technologies cannot accurately trace the root cause of quality abnormalities under the superposition of multiple factors.

[0047] This invention decomposes and calculates the riveting process curve to form material fingerprints and equipment fingerprints, decoupling the material forming characteristics and equipment structural response. This allows process changes to be quantitatively analyzed from both the material and equipment sides, providing a physically meaningful feature basis for subsequent anomaly detection.

[0048] By calculating the similarity of process fingerprints of adjacent riveting events and constructing a mutation index, the subtle evolution in the continuous riveting process is transformed into a sequentially searchable change record. This enables the identification of structural mutations in the process fingerprint while the final indicators are still qualified, thus identifying potential quality risks in advance.

[0049] This invention expands a single mutation into a measurable evolutionary process by constructing a historical fingerprint differential sequence and calculating the offset amplitude and offset trend, thereby generating a quality anomaly prediction risk for the subsequent riveting process and realizing the transformation from result judgment to process prediction.

[0050] By recalculating in reverse order according to the prediction rules on the event chain, the earliest riveting event that triggers the predicted risk is identified, and the corresponding equipment, process, mold and material information is output, so that the quality anomaly can be directly attributed to the specific operating status and resource source, thereby improving the accuracy of investigation and location.

[0051] This invention constructs an event chain with associated verification relationships and introduces a skip-level and segment fusion summary mechanism to make the riveting process data form a recalcible and verifiable continuous structure, thereby ensuring that the prediction and traceability results are based on a consistent data evolution relationship and improving the reliability in engineering applications. Attached Figure Description

[0052] Figure 1This is a flowchart of the method steps of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Refer to the instruction manual appendix Figure 1 An embodiment of the present invention provides a method for predicting and tracing quality anomalies in the riveting process, comprising:

[0055] S1. Riveting event generation steps: Obtain force data, displacement data, and time data collected by the riveting equipment during a riveting process; construct the riveting process curve corresponding to the riveting process based on the collected data; bind the riveting process curve with the equipment identifier, process parameter version, mold life, material batch, and timestamp corresponding to the riveting process to generate the riveting event corresponding to the riveting process.

[0056] S2. Process fingerprint decomposition step: Perform process decomposition calculation on the riveting process curve in the riveting event, decompose the riveting process curve into material fingerprint and equipment fingerprint; the material fingerprint is used to characterize the material forming characteristics during the riveting process, and the equipment fingerprint is used to characterize the structural response and driving characteristics of the riveting equipment; and use the material fingerprint and equipment fingerprint as the process fingerprint of the riveting event;

[0057] S3. Event chain construction steps: According to the time sequence of riveting occurrence, the currently generated riveting event is appended to the event sequence; an event summary is generated based on the summary value of the previous riveting event in the event sequence and the content of the current riveting event, and the event summary is recorded in the current riveting event, so that an event chain with an associated verification relationship is formed between adjacent riveting events;

[0058] S4. Mutation Index Construction Steps: Based on the process fingerprints of each riveting event in the event chain, perform similarity calculations on the material fingerprints and equipment fingerprints respectively, and construct a mutation index with the material fingerprints and equipment fingerprints as index keys; and use the mutation index to perform continuous monitoring on the process fingerprints of adjacent riveting events in the event chain to determine whether there are structural mutations in the process fingerprints.

[0059] Also includes S5:

[0060] S5-1, Fingerprint Differential Sequence Construction Steps: When the mutation index detects a structural mutation of the process fingerprint in the event chain or when the offset between process fingerprints is greater than a preset offset threshold, the corresponding riveting event is determined as the target riveting event.

[0061] In the event chain, a preset number of historical riveting events are selected in chronological order, with the target riveting event as the endpoint, to form a historical event set. For each historical riveting event in the historical event set, its material fingerprint and equipment fingerprint are read. Using the material fingerprint and equipment fingerprint corresponding to the target riveting event as the reference fingerprint, the material fingerprint of each historical riveting event is subjected to a term-by-term difference operation to obtain a material difference sequence arranged in chronological order. The equipment fingerprint of each historical riveting event is then subjected to a term-by-term difference operation to obtain an equipment difference sequence arranged in chronological order.

[0062] S5-2, Calculation steps for offset evolution: For the material difference sequence, calculate the increment of change between adjacent terms in chronological order to form a material change increment sequence; calculate the average absolute value of the material change increment sequence as the material offset amplitude, and calculate the difference between the first and last terms as the material offset trend; For the equipment difference sequence, calculate the increment of change between adjacent terms in chronological order to form an equipment change increment sequence; calculate the average absolute value of the equipment change increment sequence as the equipment offset amplitude, and calculate the difference between the first and last terms as the equipment offset trend; Combine the material offset amplitude, material offset trend, equipment offset amplitude, and equipment offset trend to form an offset evolution set;

[0063] S5-3, Risk prediction generation steps: Compare the material offset amplitude in the offset evolution set with a first amplitude threshold, and generate a material amplitude anomaly flag when it exceeds the first amplitude threshold; compare the material offset trend with a first trend threshold, and generate a material trend anomaly flag when it exceeds the first trend threshold;

[0064] The device offset amplitude is compared with a second amplitude threshold, and a device amplitude anomaly flag is generated when it exceeds the second amplitude threshold; the device offset trend is compared with a second trend threshold, and a device trend anomaly flag is generated when it exceeds the second trend threshold.

[0065] A combined judgment is performed on the material amplitude abnormality flag, material trend abnormality flag, equipment amplitude abnormality flag and equipment trend abnormality flag. When the preset combination conditions are met, the quality abnormality prediction risk of the subsequent riveting process is generated as the trigger state, and the target riveting event is determined as the prediction starting point event.

[0066] S5-4, Source Event Backtracking Step: Starting from the predicted starting event, select backtracking riveting events one by one in reverse chronological order in the event chain; for each backtracking riveting event, repeat the fingerprint differential sequence construction in S5-1 and the offset evolution calculation in S5-2, and regenerate the corresponding predicted risk according to the predicted risk generation rule in S5-3.

[0067] When a retrospective riveting event with the predicted risk in the triggered state occurs for the first time, the retrospective riveting event is identified as the source riveting event; and the equipment identifier, process parameter version, mold life and material batch bound in the source riveting event are read as the traceability result corresponding to the quality anomaly prediction risk output;

[0068] In the S5-1 fingerprint differential sequence construction steps, it is necessary to explain that: First, by comparing adjacent riveting events in the event chain one by one using the similarity records registered in the mutation index, the purpose is to find the position where the "change is first significantly amplified" in the entire event chain. When the material similarity value or equipment similarity value exceeds the corresponding threshold, it indicates that the process fingerprint difference between the two adjacent rivetings has exceeded the range of continuous evolution. Therefore, the latter riveting event in the adjacent event pair is determined as the target riveting event and used as a reference point for subsequent analysis. Second, after determining the target riveting event, a preset number of historical riveting events are selected from the event chain in chronological order, ending at the target riveting event. The purpose is to fix a historical window of controllable length so that subsequent calculations can reflect the evolution trend before the target event while avoiding the introduction of too much irrelevant disturbance from early data.

[0069] Then, for each historical riveting event in the historical event set, the material fingerprint and equipment fingerprint are read separately to ensure that subsequent comparisons are always performed between similar objects "material to material, equipment to equipment," avoiding the mixing of features from different sources. Next, the material fingerprint and equipment fingerprint corresponding to the target riveting event are used as the reference fingerprint to use the target state as a unified reference coordinate, so that the fingerprints at each historical moment can be converted into "deviations relative to the target state." In specific execution, the material fingerprint of each historical riveting event is subtracted from the reference material fingerprint at corresponding positions to obtain the material difference value of that historical moment relative to the target moment, and arranged in chronological order to form a material difference sequence. Similarly, the equipment fingerprint of each historical riveting event is subtracted from the reference equipment fingerprint and arranged in chronological order to form an equipment difference sequence. Finally, through these two difference sequences, the discrete historical fingerprint changes are transformed into a continuous deviation trajectory around the target riveting event, so that subsequent steps can uniformly calculate and judge the offset amplitude and offset trend under the same reference.

[0070] In the S5-2 offset evolution calculation step, it needs to be explained that the reason for first calculating the "change increment between adjacent terms" for the material difference sequence and the equipment difference sequence is to further convert the "deviation relative to the target riveting event" into evolutionary information on "whether the deviation is accelerating, decelerating, or reversing within the historical window." This is because simply looking at the difference sequence itself only tells us how much the deviation is, and cannot distinguish whether it is a one-time jump or a gradual accumulation. In specific execution, the kth term and the (k+1)th term of the difference sequence are subtracted one by one in chronological order to obtain the kth change increment, thus forming a change increment sequence, so that each change increment corresponds to "how much the deviation has changed between two adjacent historical riveting events." Subsequently, the... The absolute value of the change increment sequence is averaged as the offset amplitude. The purpose is to compress the overall intensity of multiple changes into a comparable index and eliminate the underestimation caused by the cancellation of positive and negative directions by taking the absolute value. The difference between the first and last terms of the change increment sequence is taken as the offset trend. The purpose is to use the start and end points of the same window to characterize the directionality of "whether the change increment is amplifying or converging as a whole", so that subsequent judgments have both intensity and direction. Finally, the material offset amplitude, material offset trend, equipment offset amplitude, and equipment offset trend are combined into an offset evolution set. The purpose is to put the evolution quantities of the material side and the equipment side into the same judgment input set to avoid object omissions or inconsistencies in the subsequent threshold judgment.

[0071] In the S5-3 risk prediction generation step, it should be noted that: comparing the material offset amplitude with the first amplitude threshold and generating a material amplitude anomaly flag aims to determine whether the material-side offset intensity exceeds the allowable fluctuation range, thereby converting "numerical offset" into a combinable discrete flag; comparing the material offset trend with the first trend threshold and generating a material trend anomaly flag aims to determine whether the material-side offset has a tendency to continuously increase, avoiding treating "temporarily large amplitude but declining trend" and "continuously increasing amplitude" as the same; comparing the equipment offset amplitude with the second amplitude threshold and generating an equipment amplitude anomaly flag aims to determine whether the equipment-side response fluctuation has increased abnormally, in order to capture the deviation caused by changes in structural stiffness, abnormal driving, or control drift. The displacement intensity increases; the equipment offset trend is compared with the second trend threshold to generate an abnormal equipment trend flag. The purpose is to determine whether the equipment-side offset shows a continuous deterioration direction, thereby distinguishing between one-time disturbances and cumulative anomalies. Furthermore, the purpose of performing combined judgment on the four types of anomaly flags is to introduce "multi-condition closed-loop constraints" to avoid false alarms triggered by a single flag due to noise, occasional load changes, or measurement errors. The quality anomaly prediction risk is set to the trigger state only when the preset combination conditions are met. This means that "the offset intensity and offset direction on the material side or equipment side have reached an interpretable abnormal combination," and the target riveting event is determined as the prediction starting point event. The purpose is to fix the unified reference point for subsequent prediction and backtracking so that the starting point drift does not occur in subsequent tracing.

[0072] In the S5-4 source event backtracking step, it should be noted that: starting from the predicted starting event, backtracking and riveting events are selected one by one in reverse chronological order. The purpose is to find the "first event that satisfies the same risk rule" along the event chain, thereby pushing the predicted risk back from the "occurrence position" to the "cause position". For each backtracking and riveting event, the fingerprint differential sequence construction in S5-1 and the offset evolution calculation in S5-2 are repeated. The purpose is to gradually move the "reference point" from the predicted starting event to an earlier backtracking and riveting event, and after each forward movement, the offset evolution set corresponding to the reference point is obtained again, so that the backtracking process always uses the same set of recalculated input structures. Then, the pre-predicted event is regenerated according to the rules in S5-3. The purpose of risk assessment is to ensure that retrospective judgments and predictive judgments are completely consistent, avoiding the introduction of new judgment standards during the retrospective stage that would render the source location inexplicable. Finally, when a retrospective riveting event with the predicted risk triggering state occurs for the first time, it is identified as the source riveting event. The purpose is to select the "earliest trigger point" as the causal starting point, thereby ensuring the uniqueness and verifiability of the source event. The equipment identifier, process parameter version, mold life, and material batch bound to the source riveting event are read as traceability results output. The purpose is to assign the source event to an actionable object, enabling subsequent confirmation and handling directly targeting specific equipment, specific parameter version, specific mold wear stage, and specific material source.

[0073] S1 includes:

[0074] S1-1, Timing Alignment Sampling Step: During a riveting process, force data, displacement data and time data are synchronously sampled according to a unified sampling clock. The collected force data and displacement data are respectively appended with corresponding time indices to form a force sampling sequence and a displacement sampling sequence that correspond one-to-one according to the time index.

[0075] In the riveting equipment used in this solution, the force data includes the riveting force values ​​collected in real time by pressure sensors or force sensors installed on the actuator. Specifically, it includes at least: the force amplitude reflecting the instantaneous load applied by the pressure head to the rivet and the connected parts, the force change corresponding to each sampling moment, and the maximum load value when the force reaches its peak. When a hydraulic drive structure is used, the force data can also be obtained by converting the pressure inside the hydraulic cylinder cavity through a calibration coefficient. The corresponding variable type is a continuously changing real-valued load data sequence, which is periodically read and stored by the control system through an analog acquisition module or a digital bus interface.

[0076] The displacement data includes the axial displacement collected by a displacement sensor located in the stroke direction of the slider or pressure head. Specifically, it includes at least: the current position value of the pressure head relative to the initial zero position, the displacement change between adjacent sampling times, and the final displacement value when the final pressing position is reached. The displacement data can be acquired by a grating ruler, linear encoder, or displacement transmitter, and the corresponding variable type is a continuously changing real-number position data sequence, which is synchronously collected during the riveting process via a high-speed counting module or motion control interface.

[0077] The time data includes time-identifying information used to calibrate the sampling order and interval of force and displacement data. Its specific content includes at least: the timestamp corresponding to each sampling point, the sampling interval between adjacent sampling points, and the start and end times of a riveting process. The time data is generated by a clock source or high-precision timer inside the control system, and the corresponding variable type is an incremental time-identifying data sequence, which is synchronously appended while force and displacement data are collected to construct riveting process data with a temporal relationship.

[0078] S1-2, Effective Segment Identification Step: Perform threshold start determination on the force sampling sequence to determine the riveting contact start point, and perform slope change determination on the displacement sampling sequence to determine the forming termination point; using the riveting contact start point and the forming termination point as boundaries, extract the corresponding effective sampling segments from the force sampling sequence and the displacement sampling sequence;

[0079] In the force sampling sequence, several consecutive sampling points during the idle phase of the pressure head are first selected as a baseline segment. The average value and fluctuation range of the force data within the baseline segment are calculated to generate a contact determination threshold. Then, the force sampling sequence is traversed point by point along the time sequence. When the force value of a certain sampling point is detected to exceed the contact determination threshold for the first time and remains higher than the threshold in a preset number of adjacent sampling points, the time index corresponding to the sampling point is determined to be the riveting contact start point, which is used to characterize the moment when the pressure head enters from the idle phase and makes actual contact with the workpiece and rivet and begins to be subjected to force.

[0080] In the displacement sampling sequence, the displacement difference and time difference of adjacent sampling points are divided in chronological order to obtain the corresponding instantaneous displacement slope sequence. The average slope of the instantaneous displacement slope sequence in the stable forming stage is used as the reference slope. When a certain sampling point is detected, the instantaneous displacement slope of several consecutive sampling points continues to decrease and falls below the preset slope threshold, or the slope change changes from negative to near zero and remains stable, the time index corresponding to the sampling point is determined to be the forming termination point, which is used to characterize the boundary position of the pressure head from the continuous forming stage to the final pressure shaping or dwell stage.

[0081] S1-3, Process curve reconstruction step: The force sampling sequence and displacement sampling sequence in the effective sampling segment are paired one by one according to the time index to form a force-displacement data pair sequence driven by the time index, and the force-displacement data pair sequence is used as the riveting process curve corresponding to this riveting.

[0082] In S1-3, it should be noted that: taking the force sampling sequence and displacement sampling sequence within the effective sampling segment extracted in S1-2 as input, and using the time index attached to each sampling point as a unique alignment identifier, the force sampling value and displacement sampling value obtained under the same time index are matched point by point to form a set of force-displacement data pairs arranged strictly in ascending order of time index; during execution, the system traverses each sampling point within the effective sampling segment in ascending order of time index, first reading the force sampling value corresponding to the time index, then reading the displacement sampling value corresponding to the same time index, and combining the two into the same data pair, until the pairing of all sampling points within the entire effective sampling segment is completed;

[0083] Furthermore, by continuously storing all force-displacement data pairs arranged in time index order, an ordered data sequence is formed with the time index as the implicit sorting axis, the displacement sample value as the lateral change, and the force sample value as the vertical change. This ordered data sequence constitutes the riveting process curve. Each data pair fully reflects the instantaneous displacement state and corresponding force state of the pressure head under the corresponding time index. The overall sequence continuously depicts the entire process of force changing with displacement from the riveting contact start point to the forming termination point, making the riveting process curve a basic process data that can be directly used for subsequent fingerprint decomposition, differential calculation, and anomaly prediction.

[0084] S1-4, Event Binding Generation Step: The riveting process curve is uniformly bound with the corresponding equipment identifier, process parameter version, mold life, material batch, and timestamp to generate the riveting event corresponding to this riveting operation. The equipment identifier represents the unique number of the specific riveting equipment or workstation performing the riveting operation, used to distinguish process differences arising from different equipment. The process parameter version represents the version number of a set of control parameters configured in the current riveting process, used to identify the specific process setting state corresponding to this riveting operation. The mold life represents the cumulative number of uses or work cycles completed by the pressure head or mold before this riveting operation, used to reflect the mold wear stage. The material batch represents the material batch number of the rivets or connected parts used in this riveting operation, used to distinguish performance differences between materials from different sources. The timestamp represents the specific time point when the riveting event occurs, used to determine the temporal position of the riveting event in the overall production process.

[0085] S2 includes:

[0086] S2-1, Section division steps: Obtain the corresponding riveting process curve in the riveting event, and determine the curve segment between the riveting contact start point and the forming termination point determined in S1-2 as the loading segment, and determine the curve segment after the forming termination point until the force value drops to the preset unloading threshold as the return segment.

[0087] S2-2, Back-through inversion mapping step: The force-displacement data pair sequence in the back-through section is rearranged in reverse order according to the time index, and the displacement values ​​after reversal are symbolically mapped in a manner consistent with the direction of displacement change in the loading section, so as to obtain a back-through mapping sequence aligned with the displacement direction of the loading section.

[0088] S2-3, Equipment fingerprint calculation steps: Align the force-displacement data pair sequence within the loading section with the return mapping sequence point by point according to the time index, and take the average of the corresponding force values ​​under the same time index to form the equipment projection curve, and determine the equipment projection curve as the equipment fingerprint of the riveting event;

[0089] S2-4, Material fingerprint extraction step: Using the device fingerprint as a reference, perform force value difference operation point by point on the riveting process curve in the loading section according to the time index to form a material residual curve, and determine the material residual curve as the material fingerprint of the riveting event;

[0090] S2-5, Process fingerprint acquisition steps: Use the material fingerprint and equipment fingerprint as the process fingerprint of the riveting event;

[0091] In the segment division step of S2-1, it should be noted that: taking the riveting process curve as input, by directly reading the time index corresponding to the riveting contact start point and the forming termination point determined in S1-2, the continuous force-displacement data pair sequence between the two is extracted as the loading segment, which is used to represent the process of the pressure head continuously applying load to the workpiece and completing the material forming process; then, taking the curve part after the forming termination point as candidates, a preset unloading threshold is set and the force value is compared point by point along the time index. When the force value is continuously lower than the preset unloading threshold for the first time, it is determined that the pressure head has entered the unloading return stage, thereby determining the curve segment between the forming termination point and the trigger point as the return segment. The preset unloading threshold is taken as the average value of the force value of the idle segment before the riveting contact start point plus a preset margin, so as to ensure that the return segment is only included when the load drops significantly.

[0092] In S2-2, it should be noted that: taking the force-displacement data pair sequence within the return segment determined in S2-1 as input, the sequence is rearranged in descending order of time index, transforming the unloading process that originally increases with time into a "reverse loading" process represented by increasing time index, thus achieving reverse time index rearrangement; subsequently, the displacement value of each data pair in the reversed sequence is compared with the displacement change direction of the loading segment. If the displacement change direction is opposite, the displacement value is multiplied by a negative sign to change its sign, so that all displacement values ​​are consistent with the loading segment in the direction of numerical change, thus completing the sign mapping, and finally obtaining the return mapping sequence that is aligned with the loading segment in both time index order and displacement change direction;

[0093] In S2-3, it should be noted that: taking the force-displacement data pair sequence within the loading section obtained in S2-1 and the return mapping sequence obtained in S2-2 as input, the data pairs with the same time index in the two sequences are read point by point in ascending order of time index. The loading force value and the return mapping force value corresponding to the same time index are added together and divided by two to obtain the projected force value at that time index. This projected force value is then combined with the corresponding displacement value to form a new force-displacement data pair. Subsequently, the projected force values ​​and displacement values ​​formed under all time indices are arranged continuously according to the time index to form the device projection curve. The device projection curve is used as the device fingerprint of the riveting event to characterize the structural and driving response characteristics jointly exhibited by the riveting equipment during loading and unloading.

[0094] In S2-4, it should be noted that: taking the device projection curve corresponding to the device fingerprint determined in S2-3 as a reference, the force values ​​in the riveting process curve within the loading section are read point by point in ascending order of time index, and the projected force values ​​in the device projection curve are read under the same time index. The two are subtracted to obtain the residual force value at that time index. Then, the residual force value is combined with the corresponding displacement value to form a new force-displacement data pair. Subsequently, the residual force values ​​and displacement values ​​obtained under all time indices are arranged continuously according to the time index to form the material residual curve. The material residual curve is used as the material fingerprint of the riveting event to characterize the mechanical change characteristics caused by the plastic forming of the material during the riveting process that are not explained by the device response.

[0095] S3 includes:

[0096] S3-1, Sequential Writing Step: According to the time sequence of riveting occurrence, write the currently generated riveting event to the end of the event sequence, and record the event number of the current riveting event in the event sequence; writing the currently generated riveting event to the end of the event sequence is to ensure that the event sequence strictly corresponds to the actual occurrence order of the riveting process in physical order, thereby ensuring that the subsequent summary calculation, chain association and backtracking operation based on the "previous riveting event" are based solely on the time sequence, avoiding distortion of the event summary dependency relationship due to insertion or out-of-order;

[0097] S3-2, Current Content Summary Generation Steps: For the equipment identifier, process parameter version, mold life, material batch, timestamp, and riveting process curve bound to the current riveting event, extract the summary sequentially and concatenate them in a preset order to generate the current content summary. Here, "summary extraction" refers to: performing deterministic encoding and compression mapping on the equipment identifier, process parameter version, mold life, material batch, timestamp, and riveting process curve respectively, converting their original values ​​into fixed-length summary values. The equipment identifier, process parameter version, mold life, material batch, and timestamp are directly encoded and then used in the summary calculation to obtain corresponding summaries. The riveting process curve's force-displacement data sequence is traversed by time index and a curve summary calculation is performed to obtain a curve summary. Finally, the above summary values ​​are concatenated in a preset order to form the current content summary, which simultaneously represents the context information and process data characteristics of the riveting event.

[0098] S3-3, Chain-based basic summary calculation steps: If the current riveting event is the first event, then use the current content summary as the basic input string to perform a summary operation to generate a basic event summary; otherwise, read the event summary recorded in the previous riveting event in the event sequence as the preceding summary, concatenate the preceding summary with the current content summary to form a basic input string, and perform a summary operation on the basic input string to generate a basic event summary;

[0099] In S3-3, it should be noted that: using the current content summary generated in S3-2 as the basic data source, when the current riveting event is the first event, the current content summary is directly used as the basic input string and input into a preset summary operation function for operation. The summary operation function is used to perform iterative compression mapping on the input data in byte order and output a summary value of fixed length, thereby generating a basic event summary that corresponds one-to-one with the current content summary. When the current riveting event is not the first event, the event summary recorded in the riveting event immediately preceding the current riveting event is first read from the event sequence and used as the preceding summary. Then, the preceding summary and the current content summary are concatenated in a preset order to form a basic input string. The basic input string is then input into the summary operation function to perform the same iterative compression mapping operation to generate a basic event summary that is constrained by both the preceding summary and the current content summary. This ensures that the basic event summary represents the content of the current riveting event and maintains a chain association with the previous riveting event.

[0100] The digest operation function includes one or more known deterministic hash digest algorithms such as SHA-256, SM3, and MD5, or an equivalent iterative compression mapping function, with the purpose of mapping an arbitrary length base input string to a fixed length digest value.

[0101] S3 also includes:

[0102] S3-4, Steps for constructing a skip-level summary: Based on the event number of the current riveting event, determine whether the preset level conditions are met; if so, backtrack in the event sequence to select a historical riveting event with an interval of a preset step size from the event number, and read its event summary as the level pre-sequence summary; concatenate the level pre-sequence summary with the basic event summary to form a level input string, and perform a summary operation on the level input string to generate a level event summary; otherwise, directly use the basic event summary as the level event summary.

[0103] S3-5, Segment Fusion Summary Construction Steps: Select a preset number of historical riveting events in the event sequence, with the current riveting event as the endpoint; read their current content summaries in sequence and concatenate them in chronological order to form a segment input string; concatenate the segment input string with the hierarchical event summary to form a fusion input string; and perform a summary operation on the fusion input string to generate the current event summary.

[0104] S3-6, Summary Writing and Consistency Judgment Steps: Write the current event summary and the previous summary into the current riveting event; then select a preset number of riveting events, including the current riveting event, from the event sequence, and repeat the calculations of S3-3 to S3-5 for each selected riveting event to generate a recalculated event summary; if any recalculated event summary is inconsistent with the event summary recorded in the corresponding riveting event, the event chain is determined to be invalid at the riveting event and marked as a chain abnormal event; otherwise, the event chain remains valid;

[0105] In the S3-4 skip-level summary construction step, it should be noted that: the event sequence number recorded for the current riveting event in S3-1 is used as the input index, where the event sequence number represents the sequential position of the current riveting event in the event sequence; the event sequence number is compared with preset level conditions to determine whether level construction is triggered, wherein the preset level conditions include the event sequence number satisfying zero modulo a preset level cardinality or reaching a preset level interval value, used to characterize that the current riveting event is at a position where a skip-step association needs to be established; when the preset level conditions are met, the sequence number position corresponding to the preset step size is traced back in the event sequence according to the event sequence number, and the riveting event corresponding to that sequence number position is selected as the historical riveting event, wherein the preset step size... The length is a pre-defined positive integer used to represent the interval distance of the hierarchical jump, and the recorded event summary is read from the historical riveting event as the hierarchical preorder summary; then, the hierarchical preorder summary and the basic event summary obtained in S3-3 are concatenated in a preset order to form a hierarchical input string, and the same summary operation function as in S3-3 is input to perform the operation to generate a hierarchical event summary that is simultaneously constrained by the basic event summary and the stepping historical event summary; when the preset hierarchical conditions are not met, the basic event summary is directly determined as the hierarchical event summary, wherein the hierarchical event summary is used to characterize the final summary result of the current riveting event under the jump hierarchical structure, and serves as the input for subsequent segment fusion and chain consistency judgment;

[0106] In S3-5, it should be noted that: taking the current riveting event's position in the event sequence as the anchor point, selecting a preset number of historical riveting events backward is to include "a continuous process context near the current riveting event" within the same verifiable range, so that any content change within this continuous segment will affect the subsequently generated summary results; during execution, the system reads the current content summaries generated by each of the historical riveting events in chronological order, and concatenates these current content summaries in chronological order to form a segment input string, so that the segment input string simultaneously carries the summary information of the equipment identifier, process parameter version, mold life, material batch, timestamp, and riveting process curve within the continuous segment;

[0107] The segment input string is then concatenated with the hierarchical event summary obtained in S3-4 to form a fused input string. The hierarchical event summary provides stride history constraints, and the segment input string provides neighborhood evolution constraints. The superposition of the two can avoid local forgery caused by relying solely on adjacent events. Finally, the fused input string is fed with the same summary operation function as described above to generate the current event summary. This allows the current event summary to be subject to the dual constraints of "adjacent continuous segments" and "jump hierarchical association", providing a recalcible basis for subsequent chain consistency judgment and tracing.

[0108] In S3-6, it should be noted that: using the current event summary generated in S3-5 as the final result, the current event summary and the preceding summary used to form a chain association are first written into the current riveting event, so that the riveting event has a summary basis that can be referenced and recalculated later in the event sequence; then, taking the current riveting event as the endpoint, a preset number of riveting events are selected backward from the event sequence in order to perform a local recalculation and verification on the "most recently formed sub-chain of events with hierarchical and segmental fusion relationships". During the execution, each selected riveting event is reread. The current content summary and the previous summary are processed, and the fused input string is reconstructed step by step according to the same order and summary operation rules in S3-3 to S3-5 to generate a recalculated event summary. Then, each recalculated event summary is compared with the event summaries recorded in the riveting event. If any pair is inconsistent, it indicates that the riveting event or its previous dependencies have changed, thus determining that the event chain has failed at the riveting event and marking it as a chain anomalous event. Otherwise, when all comparisons are consistent, it is determined that the current event chain remains complete and valid in the segment and its chain structure remains unchanged.

[0109] S4 includes:

[0110] S4-1, Similarity Sequence Calculation Steps: In the event chain, select two adjacent riveting events sequentially according to time order. Read the material fingerprint and device fingerprint of the preceding riveting event and the material fingerprint and device fingerprint of the following riveting event, respectively. Subtract the corresponding items from the material fingerprints of the preceding and following riveting events to obtain a material difference sequence. Calculate the average absolute value of the material difference sequence as the material similarity value. Subtract the corresponding items from the device fingerprints of the preceding and following riveting events to obtain a device difference sequence. Calculate the average absolute value of the device difference sequence as the device similarity value. Register the material similarity value and the device similarity value as similarity records for corresponding adjacent riveting event pairs.

[0111] The purpose of S4-1 is to transform multiple changes that are originally scattered in material fingerprints and device fingerprints into a unified metric that can be continuously compared along the event chain, thereby providing a basis for subsequent mutation localization. In specific execution, by comparing only two adjacent rivetings in time, the difference between corresponding positions in each pair of fingerprints is compressed into an average absolute difference, so that each pair of events corresponds to a stable and cumulative change intensity index, which avoids the amplification of single-point fluctuations and ensures that the evolution between events can be continuously tracked and entered into the index structure.

[0112] S4-2, Mutation Index Construction Steps: Using the similarity records as index entries, write the event sequence number, material similarity value and equipment similarity value of adjacent riveting event pairs into the mutation index, so that the mutation index can be searched according to the time sequence corresponding to the event sequence number, and located and searched according to the material similarity value and equipment similarity value;

[0113] The purpose of S4-2 is to write the similarity record obtained in S4-1 into the mutation index so that subsequent steps no longer need to repeatedly recalculate the similarity of the event chain, but can directly read and locate it in the mutation index in sequence and by value.

[0114] In practice, an index entry is generated for each pair of adjacent riveting events. The event sequence number of the adjacent riveting event pair and the corresponding material similarity value and equipment similarity value are written into the mutation index. The mutation index is then searched one by one according to the time sequence corresponding to the event sequence number. The adjacent riveting event pairs with similarity values ​​exceeding the threshold are directly located by the material similarity value or equipment similarity value.

[0115] S4 also includes:

[0116] S4-3, Continuity Monitoring and Mutation Determination Steps: Traverse the similarity records in the mutation index sequentially along the event sequence number, compare the material similarity value with the first preset mutation threshold, and compare the equipment similarity value with the second preset mutation threshold; if the material similarity value is greater than the first preset mutation threshold or the equipment similarity value is greater than the second preset mutation threshold, then it is determined that there is a structural mutation in the process fingerprint of the corresponding adjacent riveting event, and the subsequent riveting event is determined to be a mutation event; otherwise, it is determined that the process fingerprint of the corresponding adjacent riveting event remains continuous and the mutation index remains unchanged;

[0117] The purpose of S4-3 is to perform pairwise continuous monitoring of the event chain using the similarity records registered in the mutation index, and to provide a clear mutation judgment result when the threshold conditions are met. Specifically, during execution, similarity records are read sequentially according to the time sequence corresponding to the event number: first, the material similarity value in the record is compared with a first preset mutation threshold. If the material similarity value is greater than the first preset mutation threshold, it indicates that the material fingerprint difference between two adjacent riveting operations has exceeded the allowable range, triggering a mutation condition from the perspective of material forming characteristics; then, the equipment similarity value in the record is compared with a second preset mutation threshold. If the equipment similarity value is greater than the second preset mutation threshold, it indicates that the equipment fingerprint difference between two adjacent riveting operations has exceeded the allowable range, triggering a mutation condition from the perspective of the structural response and driving characteristics of the riveting equipment.

[0118] When any mutation condition in the above two types of comparisons is triggered, the judgment logic of "the material similarity value is greater than the first preset mutation threshold or the equipment similarity value is greater than the second preset mutation threshold" is satisfied, thereby determining that there is a structural mutation in the process fingerprint of the adjacent riveting event pair, and determining the latter riveting event in the adjacent riveting event pair as the mutation event as the starting event for subsequent prediction and tracing; otherwise, when the material similarity value is not greater than the first preset mutation threshold and the equipment similarity value is not greater than the second preset mutation threshold, it is determined that the process fingerprint of the adjacent riveting event pair remains continuous, and only the original similarity record in the mutation index is retained without adding a new mutation marker.

[0119] The working principle of this invention is as follows: During each riveting process, force, displacement, and time data are simultaneously collected to construct a force-displacement riveting process curve. This curve is then bound to contextual information such as equipment identification, process parameters, mold life, and material batch to generate riveting events. Subsequently, the riveting process curve is decomposed to extract material fingerprints and equipment fingerprints that respectively reflect the material forming characteristics and equipment structural response characteristics. Each riveting event is written into an event chain with associated verification relationships in the order of occurrence, ensuring that each riveting event forms a recalcible and continuous relationship with the preceding and following events. Based on this, the material fingerprints and equipment fingerprints of adjacent riveting events are further analyzed. The system calculates the similarity of equipment fingerprints and constructs a mutation index. It monitors whether fingerprint changes exceed a threshold along the event sequence to locate key riveting events where structural mutations occur in the process fingerprint. When a mutation is detected, the system uses the riveting event as a reference to backtrack and select several preceding riveting events. It then performs a difference analysis between the historical fingerprint and the reference fingerprint to further calculate the magnitude and trend of the offset. By combining these analyses, it generates a prediction risk of quality anomalies in the subsequent riveting process. Finally, it recalculates the above prediction rules in reverse order along the event chain to find the earliest riveting event that triggers the prediction risk and outputs its corresponding equipment, process, mold, and material information as the traceability result.

[0120] Through the above steps, the present invention enables real-time prediction of quality anomalies based on the original riveting process data, and can accurately trace back to the specific source of the problem when an anomaly occurs, so that the entire riveting process has a monitorable, predictable and traceable operating mechanism.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting and tracing quality anomalies in the riveting process, characterized in that, include: S1. Acquire force data, displacement data and time data collected by the riveting equipment during a single riveting process, and construct the riveting process curve corresponding to the riveting process based on the collected data; The riveting process curve is bound to the corresponding equipment identifier, process parameter version, mold life, material batch and timestamp to generate the riveting event corresponding to the riveting. S2. Perform process decomposition calculation on the riveting process curve in the riveting event, decomposing the riveting process curve into material fingerprint and equipment fingerprint; the material fingerprint is used to characterize the material forming characteristics during the riveting process, and the equipment fingerprint is used to characterize the structural response and driving characteristics of the riveting equipment. The material fingerprint and the equipment fingerprint are used as the process fingerprints for the riveting event; S3. Append the currently generated riveting events to the event sequence according to the time sequence in which the riveting occurs; An event summary is generated based on the summary value of the previous riveting event in the event sequence and the content of the current riveting event, and the event summary is recorded in the current riveting event, so that an event chain with an associated verification relationship is formed between adjacent riveting events; S4. Based on the process fingerprints of each riveting event in the event chain, perform similarity calculations on the material fingerprints and equipment fingerprints respectively, and construct a mutation index with the material fingerprints and equipment fingerprints as index keys; and use the mutation index to perform continuity monitoring on the process fingerprints of adjacent riveting events in the event chain to determine whether there are structural mutations in the process fingerprints.

2. The method for predicting and tracing quality anomalies based on the riveting process according to claim 1, characterized in that: Also includes S5: S5-1. When the mutation index detects a structural mutation of the process fingerprint in the event chain or the offset between process fingerprints is greater than the preset offset threshold, the corresponding riveting event is determined as the target riveting event. In the event chain, a preset number of historical riveting events are selected in chronological order, with the target riveting event as the endpoint, to form a historical event set; for each historical riveting event in the historical event set, its material fingerprint and equipment fingerprint are read respectively. Using the material fingerprint and device fingerprint corresponding to the target riveting event as the reference fingerprint, the material fingerprint of each historical riveting event is subjected to a term-by-term difference operation to obtain a material difference sequence arranged in chronological order; and the device fingerprint of each historical riveting event is subjected to a term-by-term difference operation to obtain a device difference sequence arranged in chronological order. S5-2. Calculate the increment of change between adjacent terms in the material difference sequence in chronological order to form a material change increment sequence; calculate the average absolute value of the material change increment sequence as the material offset amplitude, and calculate the difference between the first and last terms as the material offset trend. The change increment between adjacent two items is calculated sequentially for the equipment difference sequence to form the equipment change increment sequence; the average absolute value of the equipment change increment sequence is calculated as the equipment offset amplitude, and the difference between the first and last items is calculated as the equipment offset trend; the material offset amplitude, material offset trend, equipment offset amplitude, and equipment offset trend are combined to form the offset evolution set. S5-3. Compare the material offset amplitude in the offset evolution set with a first amplitude threshold, and generate a material amplitude anomaly flag when it exceeds the first amplitude threshold. The material offset trend is compared with a first trend threshold, and a material trend anomaly flag is generated when it exceeds the first trend threshold; The device offset amplitude is compared with a second amplitude threshold, and a device amplitude abnormality flag is generated when it exceeds the second amplitude threshold; The device offset trend is compared with a second trend threshold, and a device trend anomaly flag is generated when it exceeds the second trend threshold; A combined judgment is performed on the material amplitude abnormality flag, material trend abnormality flag, equipment amplitude abnormality flag and equipment trend abnormality flag. When the preset combination conditions are met, the quality abnormality prediction risk of the subsequent riveting process is generated as the trigger state, and the target riveting event is determined as the prediction starting point event. S5-4. Starting from the predicted starting event, select backtracking riveting events one by one in reverse chronological order in the event chain. For each backtracking riveting event, repeat the fingerprint differential sequence construction in S5-1 and the offset evolution calculation in S5-2, and regenerate the corresponding predicted risk according to the predicted risk generation rule in S5-3. When a backtracking riveting event with the predicted risk in the triggered state occurs for the first time, the backtracking riveting event is identified as the source riveting event. It also reads the equipment identifier, process parameter version, mold life and material batch bound in the source riveting event, and outputs the traceability results corresponding to the quality anomaly prediction risk.

3. The method for predicting and tracing quality anomalies based on the riveting process according to claim 2, characterized in that: S1 includes: S1-1. During a riveting process, force data, displacement data and time data are synchronously sampled according to a unified sampling clock. The collected force data and displacement data are respectively appended with corresponding time indices to form a force sampling sequence and a displacement sampling sequence that correspond one-to-one according to the time index. S1-2. Perform threshold start determination on the force sampling sequence to determine the riveting contact start point, and perform slope change determination on the displacement sampling sequence to determine the forming termination point; using the riveting contact start point and the forming termination point as boundaries, extract the corresponding effective sampling segments from the force sampling sequence and the displacement sampling sequence; S1-3. The force sampling sequence and displacement sampling sequence within the effective sampling segment are paired one-to-one according to the time index to form a force-displacement data pair sequence driven by the time index, and the force-displacement data pair sequence is used as the riveting process curve corresponding to this riveting. S1-4. Bind the riveting process curve with the corresponding equipment identifier, process parameter version, mold life, material batch and timestamp to generate the riveting event corresponding to the riveting.

4. The method for predicting and tracing quality anomalies based on the riveting process according to claim 3, characterized in that: S2 includes: S2-1. Obtain the corresponding riveting process curve in the riveting event. The curve segment between the riveting contact start point and the forming termination point determined in S1-2 is defined as the loading segment, and the curve segment after the forming termination point until the force value drops to the preset unloading threshold is defined as the return segment. S2-2. The force-displacement data pairs in the return section are rearranged in reverse order according to the time index, and the displacement values ​​after reversal are symbolically mapped in a manner consistent with the direction of displacement change in the loading section, so as to obtain the return mapping sequence aligned with the displacement direction of the loading section. S2-3. Align the force-displacement data pair sequence within the loading section with the return mapping sequence point by point according to the time index, and take the average of the corresponding force values ​​under the same time index to form a device projection curve, and determine the device projection curve as the device fingerprint of the riveting event. S2-4. Using the device fingerprint as a reference, perform force value difference calculation on the riveting process curve in the loading section point by point according to the time index to form a material residual curve, and determine the material residual curve as the material fingerprint of the riveting event. S2-5. Use the material fingerprint and the equipment fingerprint as the process fingerprint of the riveting event.

5. The method for predicting and tracing quality anomalies based on the riveting process according to claim 4, characterized in that: S3 includes: S3-1. According to the time sequence of riveting, write the currently generated riveting event to the end of the event sequence, and record the event number of the current riveting event in the event sequence; S3-2. Extract the summary of the equipment identifier, process parameter version, mold life, material batch, timestamp and riveting process curve bound in the current riveting event in sequence, and splice them in the preset order to generate the current content summary; S3-3. If the current riveting event is the first event, then use the current content summary as the basic input string to perform a summary operation to generate a basic event summary; otherwise, read the event summary recorded in the previous riveting event in the event sequence as the preceding summary, concatenate the preceding summary with the current content summary to form a basic input string, and perform a summary operation on the basic input string to generate a basic event summary.

6. The method for predicting and tracing quality anomalies based on the riveting process according to claim 5, characterized in that: S3 also includes: S3-4. Based on the event sequence number of the current riveting event, determine whether the preset hierarchical conditions are met. If met, backtrack in the event sequence to select a historical riveting event with an interval of a preset step size from the event sequence number, and read its event summary as the hierarchical pre-sequence summary. Concatenate the hierarchical pre-sequence summary with the basic event summary to form a hierarchical input string, and perform a summary operation on the hierarchical input string to generate a hierarchical event summary. Otherwise, use the basic event summary directly as the hierarchical event summary. S3-5. Select a preset number of historical riveting events in the event sequence, with the current riveting event as the endpoint, read their current content summaries in sequence and concatenate them in chronological order to form a segment input string; concatenate the segment input string with the hierarchical event summary to form a fusion input string, and perform a summary operation on the fusion input string to generate the current event summary; S3-6. Write the current event summary and the previous summary into the current riveting event; then select a preset number of riveting events, including the current riveting event, from the event sequence, and repeat the calculations of S3-3 to S3-5 for each selected riveting event to generate a recalculated event summary; if any recalculated event summary is inconsistent with the event summary recorded in the corresponding riveting event, the event chain is determined to be invalid at the riveting event and marked as a chain abnormal event; otherwise, the event chain is maintained as valid.

7. The method for predicting and tracing quality anomalies based on the riveting process according to claim 6, characterized in that: S4 includes: S4-1. In the event chain, select two adjacent riveting events sequentially according to time order. Read the material fingerprint and equipment fingerprint of the previous riveting event and the material fingerprint and equipment fingerprint of the subsequent riveting event respectively. Subtract the corresponding items of the material fingerprint of the previous riveting event and the material fingerprint of the subsequent riveting event to obtain a material difference sequence. Calculate the average of the absolute values ​​of the material difference sequence as the material similarity value. Subtract the corresponding items of the equipment fingerprint of the previous riveting event and the equipment fingerprint of the subsequent riveting event to obtain a device difference sequence. Calculate the average of the absolute values ​​of the device difference sequence as the device similarity value. Register the material similarity value and the device similarity value as similarity records for corresponding adjacent riveting event pairs. S4-2. Using the similarity record as an index entry, write the event number, material similarity value and equipment similarity value of adjacent riveting event pairs into the mutation index, so that the mutation index can be searched according to the time sequence corresponding to the event number, and located and searched according to the material similarity value and equipment similarity value.

8. The method for predicting and tracing quality anomalies based on the riveting process according to claim 7, characterized in that: S4 also includes: S4-3. Traverse the similarity records in the mutation index in sequence along the event number, compare the material similarity value with the first preset mutation threshold, and compare the equipment similarity value with the second preset mutation threshold; if the material similarity value is greater than the first preset mutation threshold or the equipment similarity value is greater than the second preset mutation threshold, it is determined that there is a structural mutation in the process fingerprint of the corresponding adjacent riveting event, and the subsequent riveting event is determined as a mutation event; otherwise, it is determined that the process fingerprint of the corresponding adjacent riveting event remains continuous and the mutation index remains unchanged.

Images