Information monitoring and tracing system for cold heading machine processing

Abstract

The present application relates to cold upsetting forming equipment monitoring and industrial informatization technical field, specifically to cold upsetting machine processing process informatization monitoring and tracing system, including: collecting spindle displacement data and each station instantaneous force data of multi-station forming processing equipment; extracting displacement and force corresponding relationship in single stroke and integrating to generate plastic deformation work of each station; based on work of each station, constructing real-time energy topology graph with station as node and adjacent station energy dissipation ratio as edge weight; comparing with baseline energy topology graph edge by edge to obtain topology residual error; generating rheological fingerprint hash and storing in tracing database; when any topology residual error is greater than threshold, outputting waste rejection instruction to trigger pneumatic sorting baffle, otherwise outputting good product release instruction; the present application can realize low delay determination and single piece level tracing of high speed cold upsetting process.

Description

Technical Field

[0001] This invention relates to the field of cold heading equipment monitoring and industrial information technology, specifically to an information-based monitoring and traceability system for the cold heading machine processing. Background Technology

[0002] As cold heading equipment continues to develop towards higher speeds, multi-station operation, and single-piece traceability, products such as automotive fasteners and brake system connectors have strict standards for real-time online monitoring, anomaly detection, and scrap sorting during production. Especially under high-cycle continuous production conditions, how to effectively characterize the forming state in a single stroke and complete quality judgment and traceability registration before the workpiece is discharged has become a technical problem that urgently needs to be solved in the field of cold heading.

[0003] Traditional cold heading machine processing monitoring currently relies mainly on the following methods: collecting pressure signals at a single station and monitoring peak values, manually comparing forming load curves, and conducting offline spot checks in conjunction with appearance or dimensional inspections after processing.

[0004] However, simply using methods such as pressure peak monitoring, manual comparison of load curves, and offline sampling inspection at the back end all have certain drawbacks. For example, the pressure peak monitoring method is easily affected by vibration, oil temperature drift, and sensor noise, making it difficult to accurately reflect the true energy state of material plastic deformation within a single stroke; the manual comparison of load curves relies on experience-based judgment and is difficult to reveal the energy transfer relationship and abnormal propagation location between adjacent workstations; and the offline sampling inspection method at the back end cannot adapt to high-speed continuous material discharge scenarios, and is prone to problems such as judgment lag, misflow of scrap, and difficulty in tracing the processing status of individual parts. Summary of the Invention

[0005] The purpose of this invention is to provide an information-based monitoring and traceability system for the cold heading machine process, solving the following technical problems:

[0006] The isolated pressure curve monitoring is transformed into a closed-loop judgment based on the plastic work of multiple stations and the energy distribution relationship between adjacent stations. In complex production environments such as strong vibration and oil temperature drift, it can realize real-time rejection of defective products with low latency and single-piece quality traceability. It can also effectively distinguish between raw material batch fluctuations and local mold wear, and provide predictive maintenance basis for mold life.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] An information-based monitoring and traceability system for the cold heading machine process is applied to multi-station forming equipment that includes controllers and pneumatic sorting baffles, including:

[0009] The data acquisition module is used to acquire the spindle displacement data and instantaneous force data of each station of the multi-station forming and processing equipment; wherein, the spindle displacement data is generated by the spindle rotary encoder installed on the multi-station forming and processing equipment, and the instantaneous force data is generated by the piezoelectric force sensor installed on the multi-station forming and processing equipment.

[0010] The work integral module is used to extract the correspondence between the spindle displacement data and the instantaneous force data of each station in a single stroke, and to generate the plastic deformation work done by each station in the single stroke through definite integral calculation.

[0011] The topology construction module is used to construct a real-time energy topology graph based on the plastic deformation work done at each workstation; wherein, the nodes of the real-time energy topology graph are each workstation, the edge weights of the real-time energy topology graph are the energy dissipation ratios between adjacent workstations, and the energy dissipation ratio is the value of the plastic deformation work done at the next workstation divided by the plastic deformation work done at the previous workstation.

[0012] The residual analysis module is used to obtain a preset baseline energy topology map and calculate the topological residuals of each corresponding edge between the real-time energy topology map and the baseline energy topology map.

[0013] The traceability and control module is used to generate a rheological fingerprint hash based on the real-time energy topology map and store the rheological fingerprint hash in a preset traceability database; and is configured to: if the topological residual of any corresponding edge is greater than a preset residual threshold, generate a defect rejection instruction and send it to the controller to trigger the pneumatic sorting baffle; if the topological residual of all corresponding edges is not greater than the preset residual threshold, generate a good product release instruction.

[0014] In one possible implementation, the work integration module is specifically used for:

[0015] Based on the aforementioned correspondence, a curve showing the correspondence between the instantaneous force data and the spindle displacement data within a single stroke at the current workstation is fitted.

[0016] Perform a definite integral operation on the corresponding relationship curve to output the amount of work done by the plastic deformation of the workpiece processed at the current station within the single stroke.

[0017] In one possible implementation, the topology building module is specifically used for:

[0018] Each workstation is defined as a node in a directed graph according to its processing order, and the connection between adjacent workstations is defined as an edge in the directed graph.

[0019] Divide the amount of work done by plastic deformation in the next station by the amount of work done by plastic deformation in the previous station to calculate the energy dissipation ratio between adjacent stations.

[0020] The energy dissipation ratio is used as the edge weight and assigned to the corresponding edge of the directed graph to output the real-time energy topology graph.

[0021] In one possible implementation, the residual analysis module is specifically used for:

[0022] During the initialization phase, the spindle displacement data and instantaneous force data of the pre-collected standard good products are extracted to generate the baseline energy topology map;

[0023] At the end of the single stroke, the absolute value of the difference between the energy dissipation ratio of each corresponding edge in the real-time energy topology map of the current workpiece and the baseline energy topology map is calculated, and the absolute value is output as the topology residual of each corresponding edge.

[0024] In one possible implementation, the tracing and control module, when generating the rheological fingerprint hash, is specifically used for:

[0025] The energy dissipation ratios of each edge in the real-time energy topology graph are arranged in topological order and extracted as feature vectors.

[0026] Principal component analysis is performed on the feature vectors to reduce their dimensionality, generating dimensionality-reduced feature vectors.

[0027] Obtain the media access control address of the multi-station forming processing equipment and the timestamp of the single stroke. Concatenate the reduced feature vector with the media access control address and the timestamp using a preset separator to form an original fused string. Perform a preset hash operation on the original fused string to generate the rheological fingerprint hash.

[0028] In one possible implementation, when generating the waste rejection instruction, the traceability and control module is specifically used for:

[0029] If the topological residual of any corresponding edge is greater than the preset residual threshold, it is determined that the workstation associated with the corresponding edge has undergone a sudden change in deformation resistance.

[0030] The scrap rejection instruction is specifically a high-speed input / output instruction sent to the controller;

[0031] The pneumatic sorting baffle is triggered by the high-speed input / output command to remove the corresponding workpiece to the preset waste bin.

[0032] In one possible implementation, an evolution monitoring module is also included, for:

[0033] Extract the real-time energy topology map of historical batches from the preset traceability database and generate topological evolution trajectories arranged in time series;

[0034] Calculate the slope of the total amount of plastic deformation work done at each station in the topological evolution trajectory within a preset time window. If the absolute value of the slope is greater than a preset slope threshold, the raw material hardness is determined to be in an abnormal fluctuation state; if the absolute value of the slope is not greater than the preset slope threshold, the raw material hardness is determined to be in a normal state.

[0035] The calibration logic for the preset slope threshold and the preset variance threshold is as follows: During the normal operating cycle of the equipment when replacing the mold with a brand new one and using standard raw materials, at least a preset number of good product strokes are extracted as a reference window. The extreme value of the slope fluctuation and the mean value of the edge weight variance of the total work done within the reference window are calculated. Then, a preset tolerance coefficient is multiplied on this basis to dynamically generate the preset slope threshold and the preset variance threshold that are adapted to the current machine status.

[0036] In one possible implementation, the evolution monitoring module is also used for:

[0037] In response to determining that the corresponding workstation mold is in a worn state, the decreasing gradient of the plastic deformation work done at the corresponding workstation over time is extracted.

[0038] Based on the descent gradient, linear extrapolation is performed to calculate the estimated time point when the work done by plastic deformation drops to the preset failure work threshold, and the corresponding station mold life prediction curve is generated.

[0039] The beneficial effects of this invention are:

[0040] 1. This invention transforms isolated pressure curve monitoring into a closed-loop judgment based on multi-station plastic work and energy distribution relationship between adjacent stations. In complex production environments such as strong vibration and oil temperature drift, it can effectively resist the interference of single-point high-frequency noise and realize low-delay judgment and real-time rejection of defective products in the high-speed cold heading process.

[0041] 2. This invention extracts real-time energy topology maps and fuses them in a dimensionality-reducing manner to generate lightweight rheological fingerprint hashes. This not only preserves the processing dynamics characteristics but also greatly reduces data storage requirements, achieving efficient traceability of single-piece quality status.

[0042] 3. By establishing an evolution monitoring module, this invention calculates the slope of the total work change and the sliding variance of the edge weights according to the time series. It can decouple and effectively distinguish the hardness fluctuation of raw material batches from the progressive wear state of local station molds, and supports linear extrapolation to predict mold life, providing a quantitative basis for predictive maintenance of equipment. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the modules of the cold heading machine processing information monitoring and traceability system provided in the embodiments of this application. Detailed Implementation

[0044] 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.

[0045] Please see Figure 1 An information monitoring and traceability system for cold heading machine processing is applied to a multi-station forming processing equipment including a controller and pneumatic sorting baffles. The system includes a data acquisition module for acquiring spindle displacement data and instantaneous force data at each station of the multi-station forming processing equipment. The spindle displacement data is generated by a spindle rotary encoder installed on the multi-station forming processing equipment, and the instantaneous force data is generated by a piezoelectric force sensor installed on the multi-station forming processing equipment.

[0046] The work integral module is used to extract the correspondence between the spindle displacement data and the instantaneous force data of each station in a single stroke, and to generate the plastic deformation work done by each station in the single stroke through definite integral calculation.

[0047] The topology construction module is used to construct a real-time energy topology graph based on the plastic deformation work done at each workstation; wherein, the nodes of the real-time energy topology graph are each workstation, the edge weights of the real-time energy topology graph are the energy dissipation ratios between adjacent workstations, and the energy dissipation ratio is the value of the plastic deformation work done at the next workstation divided by the plastic deformation work done at the previous workstation.

[0048] The residual analysis module is used to obtain a preset baseline energy topology map and calculate the topological residuals of each corresponding edge between the real-time energy topology map and the baseline energy topology map.

[0049] The traceability and control module is used to generate a rheological fingerprint hash based on the real-time energy topology map and store the rheological fingerprint hash in a preset traceability database; and is configured to: if the topological residual of any corresponding edge is greater than a preset residual threshold, generate a defect rejection instruction and send it to the controller to trigger the pneumatic sorting baffle; if the topological residual of all corresponding edges is not greater than the preset residual threshold, generate a good product release instruction.

[0050] This embodiment provides an information monitoring and traceability mechanism for the processing of a five-station high-speed cold heading machine. Specifically, the cold heading machine is used for mass production of high-strength bolts for automotive braking systems. The equipment operates at a rate of 480 pieces per minute. A high-speed programmable logic controller is installed in the control cabinet, and a pneumatic sorting baffle is installed at the end of the discharge channel. One side of the baffle is connected to the good product chute, and the other side is connected to the waste product bin.

[0051] The system revolves around the complete processing lifecycle of a single workpiece from entering the first station to exiting the fifth station. After each spindle stroke, the system synchronously calculates the force and displacement information of the workpiece at the corresponding multi-station, and completes the judgment and traceability registration before the workpiece falls into the collection tank.

[0052] The details are as follows: A rotary encoder is installed at the front end of the spindle, which discretizes the spindle rotation angle into a continuous pulse sequence; the data acquisition module also includes an edge computing unit. Since there is a pre-calibrated geometric correspondence between the spindle mechanism and the slider displacement, the edge computing unit can map the rotation angle pulse at each moment to the spindle displacement data of each sampling point under that stroke.

[0053] Each workstation has a piezoelectric force sensor installed at the rear of the mold base to output instantaneous force data during the impact forming stage. The data acquisition module synchronizes the encoder pulses and multiple force signals with a unified clock to form a structured matrix of workstation number, displacement sampling point, and force sampling point. For ease of explanation, the following deduction is based on a simplified exemplary data model.

[0054] Suppose that during a single-piece forming process, data is collected from three stations. Each station takes only four discrete displacement points within a single stroke, with displacement points of 0 mm, 1 mm, 2 mm, and 3 mm. The forces at these four points at the first station can be approximated as 0, 10, 18, and 0, at the second station as 0, 8, 12, and 0, and at the third station as 0, 6, 15, and 0, with units in kN.

[0055] The work integral module does not directly compare whether the peak force exceeds the limit. Instead, it calculates the area of ​​the force-displacement relationship of each station according to the displacement direction to obtain the work done by plastic deformation. If the trapezoidal approximation is used, the work done by the first station can be regarded as the sum of the areas of the three trapezoids in the intervals [0,1], [1,2], and [2,3], which are 5J, 14J, and 9J, respectively, with a total of 28J.

[0056] The work done at the second workstation is 4, 10, and 6, totaling 20; the work done at the third workstation is 3, 10.5, and 7.5, totaling 21. Therefore, the topology building module defines the three workstations as nodes N1, N2, and N3, connects adjacent workstations with directed edges N1→N2 and N2→N3, and calculates the energy dissipation ratio for each. The latter is This forms the real-time energy topology diagram of the current workpiece.

[0057] During the initialization phase, the system generates a baseline energy topology map using several standard good samples. Continuing with the three-station scenario mentioned above, if the baseline edge weights are 0.72 and 1.02 respectively, the residual analysis module compares the real-time map with the baseline map edge by edge to obtain the corresponding topological residuals. as well as If the preset residual threshold is 0.02, then the first edge does not exceed the limit, but the second edge does, indicating that there is an abnormal distortion in the energy distribution relationship between the second and third workstations.

[0058] At this time, the traceability and control module generates the rheological fingerprint hash of the workpiece based on the current topology and writes it into the traceability database. On the other hand, it immediately generates a scrap rejection instruction and sends it to the controller via the high-speed input / output interface to drive the pneumatic sorting baffle to guide the workpiece to the scrap bin. If the topological residuals of all edges do not exceed the threshold, the controller keeps the baffle in the good product channel position, allowing the workpiece to enter the normal collection process.

[0059] As a fault-tolerant protection mechanism, if encoder pulses are missing during the acquisition process, resulting in discontinuous displacement sequences, the current stroke data is marked as invalid sampling and will not participate in integration, topology construction, or good product release determination. A conservative rejection instruction can be directly output to avoid missed detection.

[0060] If a piezoelectric signal becomes saturated or disconnects, the status of that workstation can be set to pending confirmation, the fault code can be recorded in the traceability database, and the corresponding workpiece can be treated as an abnormal part. If the work done at a certain workstation is below the system detection limit, which may cause the subsequent ratio calculation to have the risk of division by zero overflow or low calculation accuracy, the system will first determine whether the value is lower than the minimum effective work threshold. The minimum effective work threshold is obtained by adding the no-load operation work done by the equipment to the preset safety margin calibration. If it is lower, the workstation is regarded as a sensor abnormality or a serious no-load state, the ratio calculation is stopped directly, and rejection or shutdown alarm is triggered.

[0061] For example, in the above-mentioned automotive brake system bolt production line, a wire rod goes through five stations in sequence: pre-forming, upsetting, head shaping, corner transition and final shaping.

[0062] If a slight chipping occurs in the punch bar at the fourth station, the peak force at that station may not be significant under noise interference. However, due to the change in the work transfer relationship between the fourth and fifth stations, the energy dissipation ratio between the fourth and fifth stations will deviate from the baseline. The system will immediately identify this deviation after the workpiece leaves the fifth station, complete the rejection and hash archiving. Subsequent quality personnel can trace back to the specific timestamp, machine, and topological features corresponding to that stroke based on the hash.

[0063] The purpose of this step is to transform the previously isolated pressure curve monitoring into a closed-loop judgment based on the plastic work done at multiple workstations and the energy distribution relationship between adjacent workstations, so that the system can still maintain low-latency rejection and single-piece-level traceability capabilities under strong vibration and oil temperature drift environments.

[0064] In a preferred embodiment of the present invention, the work integral module is specifically used to: fit the correspondence curve between the instantaneous force data and the spindle displacement data of the current station in a single stroke based on the correspondence relationship; perform definite integral operation on the correspondence curve, and output the work done by the plastic deformation of the workpiece processed at the current station in the single stroke.

[0065] This embodiment provides a refinement mechanism for the work integration process; specifically, when directly accumulating based solely on discrete sampling points, if the sampling interval is affected by instantaneous vibrations and becomes locally uneven, problems may arise such as the difference between adjacent points exceeding the preset range and the curve becoming obviously jagged, which makes the work calculation extremely susceptible to interference from single-point high-frequency noise.

[0066] Therefore, in this embodiment, the instantaneous force data of a certain station and the spindle displacement data are first fitted to the corresponding curve within a single stroke, and then the fitted relationship curve is integrally applied to improve the stability and physical interpretability of the work done estimation.

[0067] The details are as follows: The data acquisition module outputs a set of discrete sample pairs aligned by time, such as obtaining several sets of displacement values ​​and force values ​​at the same workstation; the work integration module sorts the displacement values ​​in ascending order, deletes duplicate displacement points and outliers exceeding the preset threshold, and then performs piecewise fitting on the remaining samples; in the high-speed cold heading scenario, piecewise linear fitting or low-order spline fitting is preferred, rather than using complex models whose computational complexity exceeds the system response time limit, in order to ensure millisecond-level response;

[0068] The following deduction is illustrated using a simplified exemplary data model; assuming that the second station obtains 5 valid sample pairs after denoising preprocessing during a certain stroke, which are respectively , , , , The work integration module first fits the above points into a smooth force-displacement relationship curve; if a piecewise linear method is used, it can be regarded as being composed of four small line segments.

[0069] The areas of the four trapezoids are calculated separately: the first segment is approximately 2.8, the second segment is approximately 8, the third segment is approximately 8.8, and the fourth segment is approximately 3.6, with a total work done of approximately 23.2. The system uses this value as the amount of work done by the station on the current workpiece during this stroke for plastic deformation. If the typical work done at the same station in the baseline good workpiece ranges from 22 to 24, then the energy input of the workpiece at that station is within the normal range.

[0070] Furthermore, the fitting step can also take into account stroke partitioning; for the upward idle stroke, contact compaction zone, plastic main deformation zone, and unloading zone, the system can only extract the segments with forces greater than the preset contact threshold for integration; the contact threshold is the pre-measured maximum value of the frictional force during the die idle stroke. Taking the above data as an example, if the displacement range of 0 to 0.4 mm only reflects the die contact pre-tightening and does not form effective plastic flow, then this range can be excluded from the effective integration window; this can reduce the contamination of work done by idle stroke friction and mechanical clearance.

[0071] As a fault-tolerant protection mechanism, if the number of valid sampling points is less than the preset minimum number of points, such as less than 3, it is considered that a reliable curve cannot be formed. At this time, the work done at the current station will not be directly output as a value, but will be set to an invalid state, and the upper-level control logic will classify the workpiece as an abnormal part.

[0072] If the fitted curve shows a local negative force segment, and the negative value is not caused by the sensor zero drift correction, then the negative value is truncated to 0 before integration to avoid negative work that does not conform to the physical meaning of forming; if the displacement data shows a local backtracking, that is, the displacement of the later sampling point is less than the displacement of the previous sampling point, then the system first reorders according to the spindle angle; if it is still not monotonic after reordering, then the current stroke synchronization is marked as abnormal.

[0073] For example, in the aforementioned automotive brake system bolt production line, the first station is responsible for the initial upsetting after cutting the material. The forming time is transient and there are high-amplitude stress peaks. If the peak value is read directly, it may cause large fluctuations due to changes in the lubricating oil film.

[0074] After fitting the force-displacement curve, the integral yields the total energy absorbed during the entire deformation process, which is closer to the actual plastic flow state of the material. In this way, even if the peak value fluctuates slightly, the system can still stably provide the amount of work done, providing a consistent input for subsequent energy topology construction.

[0075] The purpose of this step is to replace the transient peak characteristics that are susceptible to shock noise with the scalar energy obtained by the definite integral after fitting, thereby achieving a robust characterization of the plastic deformation strength of a single stroke.

[0076] In a preferred embodiment of the present invention, the topology construction module is specifically used to: define each workstation as a node of a directed graph according to the processing order, and define the connection between adjacent workstations as an edge of the directed graph;

[0077] Divide the work done by the plastic deformation of the next work station by the work done by the plastic deformation of the previous work station to calculate the energy dissipation ratio between adjacent work stations; assign the energy dissipation ratio as the edge weight to the corresponding edge of the directed graph to output the real-time energy topology graph.

[0078] This embodiment provides a topological modeling mechanism for energy relationships across multiple workstations. Specifically, while knowing only the work done at each workstation can reveal individual anomalies, it is still difficult to uncover the plastic flow coupling relationship between workstations. For example, a decrease in work done at a certain workstation could be due to wear on the mold at that workstation or excessive deformation in the previous pass, reducing the plastic deformation work required for that workstation. To explicitly represent this cross-workstation relationship, this embodiment abstracts the processing sequence of multiple workstations into a directed graph and defines the ratio of work done at adjacent workstations as edge weights.

[0079] The details are as follows: If the equipment has five workstations, the topology building module generates five nodes according to the processing sequence. , , , , The edge set is , , , The weight of each edge is obtained by dividing the work done by the work done by the previous edge, and is recorded as the energy dissipation ratio between the corresponding edges.

[0080] The dissipation here does not require that the ratio must be less than 1, but rather indicates the degree of redistribution of energy demand between adjacent stations during the transfer of workpieces in each pass; therefore, if a certain pass undertakes compensatory deformation, the ratio of the subsequent station can also be greater than 1.

[0081] The following derivation is illustrated using a simplified exemplary data model; assuming the work done by a certain workpiece at five workstations is 30, 24, 18, 20, and 10 respectively; then the edge weights of the real-time topology graph are as follows: , , , This four-dimensional vector This can be used as a set of core topological features for the workpiece; if a typical set of edge weights for a standard good product is... Then it can be seen that the third edge has been significantly raised, indicating that the fourth station has undertaken an abnormally increased deformation task;

[0082] This representation method can also suppress global unidirectional drift; for example, when the overall stiffness of the machine changes due to the increase in oil temperature, the work done at the five stations may be simultaneously amplified to 33, 26.4, 19.8, 22, and 11, with their absolute values ​​increasing by about 10% overall, but the edge weights remain at 0.8, 0.75, 1.11, and 0.5, basically unchanged; compared with the scheme of directly comparing the original force values ​​or peak values, this edge weight ratio can better reflect the relative energy distribution of the material in the continuous stations;

[0083] As a fault-tolerant protection mechanism, if the work done by the previous station is less than the minimum effective value, for example, close to zero, the corresponding edge weights are not directly divided, but denominator protection is executed first; one of the following methods can be adopted: First, set the edge weight to a preset upper limit value and mark it as an abnormal edge;

[0084] Secondly, terminate the construction of the current part's topology graph and enter the rejection process; if a certain intermediate station lacks effective work, the two edges related to it before and after can be marked as unusable to avoid incomplete data misleading the diagnosis; if the equipment has station skipping processing technology, such as a certain batch temporarily stopping the fourth station, the node sequence can be reconstructed into a new directed graph according to the current effective process route, instead of forcibly using a fixed topology.

[0085] For example, on the aforementioned bolt production line, when the third station is originally responsible for the head transition forming and the fourth station is responsible for finishing; if the wear of the punch bar at the third station causes a decrease in the work done by the station on material injection, some of the residual deformation will be delayed to the fourth station for compensation, resulting in a combination of reduced work done by the third station and increased work done by the fourth station.

[0086] Mapped to the topology graph, The weight of this edge will jump from around 0.8 near the baseline to over 1.1; field engineers do not need to review complex time series curves one by one, but only need to check which edge in the graph structure is distorted to quickly locate the abnormal propagation chain;

[0087] The purpose of this step is to further organize the discrete station energy scalars into a graph structure that reflects the processing sequence and adjacent coupling relationships, thereby achieving a compact expression of the material's rheological state across stations.

[0088] In a preferred embodiment of the present invention, the residual analysis module is specifically used for: during the initialization phase, extracting the spindle displacement data and instantaneous force data of the pre-collected standard good product to generate the baseline energy topology map; at the end of the single stroke, calculating the absolute value of the difference between the energy dissipation ratio of each corresponding edge in the current workpiece's real-time energy topology map and the baseline energy topology map, and outputting the absolute value as the topology residual of each corresponding edge.

[0089] This embodiment provides a residual analysis mechanism for anomaly detection. Specifically, after the real-time topology graph is constructed, if a stable reference is lacking, the system still cannot determine whether the current edge weight fluctuation is a normal process fluctuation or a defect sign. Therefore, this embodiment pre-establishes a baseline energy topology graph during the initialization phase and calculates the topology residual by edge-by-edge absolute difference after each stroke, thereby forming a simple, fast and threshold-friendly judgment basis.

[0090] The details are as follows: Initialization can be performed after the equipment has been changed and adjusted and it is confirmed that good products are being produced continuously; the system selects a preset number of standard good products, such as 50 consecutive pieces, calculates the real-time topology graph of these samples, and performs statistical aggregation on the edge weight of each edge.

[0091] The aggregation method can be the mean, median, or the mean after removing outliers; finally, a baseline energy topology graph is obtained, in which each edge has a standard edge weight; subsequently, each workpiece obtains its own real-time topology graph after the stroke ends, and then compares it edge by edge with the baseline graph, taking the absolute value of the difference as the topology residual;

[0092] The derivation is illustrated using a simplified exemplary data model; assuming that a five-station cold heading machine, after initialization, obtains the baseline edge weight vector as follows: The real-time edge weight of a certain workpiece is The residual vector is then... ,Right now If the residual threshold is uniformly set to 0.05, the third edge will exceed the limit, while the remaining edges will not.

[0093] Based on this, the system can directly determine that the anomaly is concentrated in the energy distribution link between the third and fourth workstations. In actual deployment, the initialized baseline is not permanently fixed. To prevent misjudgment caused by the slow drift of long-cycle processes, a manual confirmation update mode can be set up. That is, only after material change, mold change, or process adjustment, the quality personnel select a batch of samples that have passed manual re-inspection from the database to regenerate the baseline. This not only preserves the stable reference function of the baseline, but also avoids the erroneous absorption of abnormal states into the baseline.

[0094] As a fault-tolerant protection mechanism, if the number of standard good products collected during the initialization phase is less than the minimum modeling number, such as less than 10 pieces, the system will not enter the automatic rejection mode, but will only output a monitoring alarm and will not drive the baffle action; if the historical dispersion of a certain edge exceeds the preset dispersion standard, different residual thresholds can be set for different edges, rather than a uniform threshold.

[0095] If an edge in the real-time topology graph is unavailable, but a corresponding edge exists in the baseline, the residual of that edge will not be compared numerically, but will be directly marked as a missing residual. The upper-level control can handle it according to the conservative principle. If the number of nodes in the baseline graph is inconsistent with the current process route, for example, if the equipment is changed from five stations to four stations, it must be re-initialized first; otherwise, the residual judgment will not be performed.

[0096] For example, on the aforementioned bolt production line, during the initialization phase after the equipment starts up, 30 consecutive products that have passed the size and appearance verification are taken as modeling samples, and the system forms the baseline diagram for the shift based on this.

[0097] During formal production, the residual of the third edge of the 1763rd workpiece suddenly increased from the usual 0.01 to 0.17, while the first, second, and fourth edges remained normal. This result indicates that the anomaly is not a general drift caused by the temperature rise of the whole machine, but is concentrated in a local workstation. Based on this, the system triggers rejection and marks the workpiece as a key traceability object.

[0098] The purpose of this step is to compress the complex multi-station dynamic relationship into a residual quantity that can be directly used for threshold control by comparing the edge-by-edge residuals of the baseline map and the real-time map, so as to facilitate low-latency on-site judgment.

[0099] In a preferred embodiment of the present invention, when generating the rheological fingerprint hash, the tracing and control module is specifically used to: arrange the energy dissipation ratio of each edge in the real-time energy topology graph in topological order and extract it as a feature vector; perform principal component analysis on the feature vector to reduce its dimensionality and generate a dimensionality-reduced feature vector.

[0100] The medium access control address of the multi-station forming processing equipment and the timestamp of the single stroke are obtained. The reduced feature vector, the medium access control address, and the timestamp are concatenated and merged with the original fused string using a preset separator. A preset hash operation is then performed on the original fused string to generate the rheological fingerprint hash. The medium access control address is specifically the physical address of the network communication gateway equipped on the multi-station forming processing equipment.

[0101] This embodiment provides a rheological fingerprint generation mechanism for single-piece traceability coding. Specifically, although storing the real-time topology map as is can preserve information, the amount of data consumes a lot of storage resources, which is not conducive to long-term traceability in high-speed production. If only the qualified / unqualified binary results are recorded, the processing status details will be lost, and it cannot support post-processing quality inspection.

[0102] Therefore, in this embodiment, the real-time energy topology map is compressed into an ordered feature vector, which is then fused with device identifiers and time information after dimensionality reduction to generate a lightweight rheological fingerprint hash.

[0103] The details are as follows: For a real-time energy topology map containing a total of m workstations, the system extracts the edge weights of m-1 edges according to the process topology order to obtain the original feature vector; taking a five-workstation device as an example, the vector length is 4, and the order is fixed as follows. ;

[0104] To reduce the impact of noise on coding stability, the tracing and control module further performs principal component analysis (PCA) dimensionality reduction on the vector. PCA can be trained offline using historical good product samples to obtain the baseline mean vector and the eigenvector projection matrix corresponding to the k largest eigenvalues ​​of the first retained dimension. It is a positive integer, and The value of is less than the dimension of the original feature vector; in the online stage, the current feature vector is centered by subtracting the baseline mean vector, and then multiplied by the feature vector projection matrix to perform projection, thereby balancing physical feature alignment and real-time online calculation.

[0105] The derivation is illustrated using a simplified exemplary data model; assume the original topological feature vector of a certain workpiece is:

[0106]

[0107] superscript This represents the transpose operation of a vector or matrix; the baseline mean vector obtained from offline historical modeling is... And extract the projection matrices of the first two principal components. The system calculates the centered vector. Performing matrix multiplication yields a reduced-dimensional vector:

[0108]

[0109] in, The original topological feature vector, Let be the eigenvector after dimensionality reduction. Assume that the dimensionality-reduced vector obtained after projection is . For ease of storage, the system can quantize the numerical value into the string "1310|-270", where the quantization scale is determined by a preset magnification factor.

[0110] The system reads the media access control address of the machine's gateway, for example, 00-1A-2B-3C-4D-5E, and then obtains the timestamp at the end of the stroke, for example, 20250108143015986. These three are concatenated to form the original merged string: 1310|-270#00-1A-2B-3C-4D-5E#20250108143015986; where # is a preset string concatenation separator. The system performs a preset hash operation on this merged string to obtain a fixed-length code.

[0111] In specific implementation, the original fusion string with special separation characteristics is mapped by the 256-bit version of the secure hash algorithm, and the first 16 bits of the obtained hexadecimal string are extracted to retain the characteristics. The extracted code is used as the final rheological fingerprint hash and is combined with the original dimensionality reduction vector, the judgment result, and the program number as a complete record and stored in the traceability database.

[0112] The rheological fingerprint generated in this way has two layers of meaning: one layer is used for database indexing, which facilitates the quick retrieval of products from the same machine, the same time period, and the same rheological state; the other layer retains the processing dynamics characteristics from the topology map, so that the traceability object is not only a record of a certain material, but also a record of a certain forming state.

[0113] As a fault-tolerant protection mechanism, if there are missing edges in the input vector during dimensionality reduction, conventional principal component projection is not performed. Instead, abnormal fingerprints are generated. For example, a fault identifier segment is added before the fusion string to distinguish fingerprints generated by normal projection from fingerprints generated by incomplete acquisition. If the media access control address changes due to the replacement of the motherboard, the device address mapping table is retained in the database to prevent fragmented tracing of the same machine at different maintenance stages. If the timestamp is abnormally rolled back or duplicated, the system can add a stroke counter as a redundant field to ensure the uniqueness of the single-item code.

[0114] For example, on the aforementioned bolt production line, after the day shift, quality personnel discovered two pieces with head size deviations mixed in with a batch of packaging boxes. By querying the traceability database by associating the packaging boxes with the batch number, the rheological fingerprint set within that time period can be located in reverse. Then, by filtering according to the offset of the third principal component or the corresponding residual, it can be found that the two workpieces came from the same machine and the same mold state evolution stage, thus helping to confirm that the source of the problem is localized workstation wear, rather than back-end sorting and handling errors.

[0115] The purpose of this step is to achieve lightweight traceability coding at the single-piece level without sacrificing the rheological characteristics of multiple workstations, thereby supporting the storage, retrieval, and quality traceability of massive data in high-speed cold heading production.

[0116] In a preferred embodiment of the present invention, when the traceability and control module generates the scrap rejection instruction, it is specifically used to: if the topological residual of any corresponding edge is greater than the preset residual threshold, determine that the workstation associated with the corresponding edge has undergone a sudden change in deformation resistance; the scrap rejection instruction is specifically a high-speed input / output instruction sent to the controller; the high-speed input / output instruction triggers the pneumatic sorting baffle to reject the corresponding workpiece into the preset scrap bin.

[0117] This embodiment provides a closed-loop control mechanism for the on-site waste removal execution link; specifically, simply completing anomaly identification without forming a time-series closed loop with the sorting mechanism still cannot solve the actual problem of identifying but not being able to remove waste in high-speed cold heading production.

[0118] Especially at a rate of hundreds of pieces per minute, if control commands are forwarded layer by layer through ordinary communication protocols, the delay may span multiple workpiece intervals, leading to incorrect rejection or missed rejection. Therefore, this embodiment uses high-speed input / output commands directly connected to the controller to achieve timely rejection of corresponding workpieces.

[0119] The following is a detailed description: After receiving the residual analysis results, the traceability and control module determines whether the topological residual on any side exceeds the preset threshold. If it does, the preceding and following workstations associated with that side are considered to have experienced a sudden change in deformation resistance. The so-called sudden change in deformation resistance can correspond to the local abnormal deformation resistance changes caused by material microcracks, mold breakage, lubrication failure, etc.

[0120] The system calculates the optimal triggering time of the baffle based on the current conveying time of the workpiece from the mold outlet to the sorting baffle, and outputs the rejection pulse directly to the controller through the high-speed input / output interface;

[0121] After receiving a pulse, the controller drives the solenoid valve to switch, causing the pneumatic sorting baffle to cut into the waste channel when the workpiece passes through the diversion point; if there is no over-threshold edge, the good product release logic is output, and the baffle remains or returns to the good product channel position.

[0122] The following is a deduction based on a simplified exemplary data model; it is assumed that the equipment produces one workpiece every 125 milliseconds, the flight time from the mold outlet to the sorting baffle is 80 milliseconds, the baffle action setup time is 12 milliseconds, and the return time is 20 milliseconds.

[0123] A certain workpiece in time The stroke is completed in milliseconds, in time The topological residual calculation is completed in milliseconds, and the residual of the third edge is found to be 0.11, which is greater than the threshold of 0.05; therefore, the system is in time... The time in milliseconds is used to register the item as a rejected item, and the optimal output time is calculated to be approximately [time value missing]. Milliseconds, causing the baffle to be in time The switch is completed within milliseconds, precisely intercepting the device.

[0124] Then in time A return command is issued approximately every millisecond to avoid affecting the passage of the next good product; if the next product is also defective, the return logic can be temporarily suspended, and the baffle is kept in the direction of scrap in the continuous rejection mode; as a fault tolerance protection mechanism, if multiple defective products occur consecutively within the same time window, the system should establish a rejection queue instead of only retaining the single product flag.

[0125] The queue records at least the workpiece number, the predicted arrival time, and the baffle action status. If the controller reports that the baffle has not arrived on time, for example, the limit switch has not acted within the predetermined time, then the current and subsequent workpieces can be uniformly transferred to conservative rejection or trigger a shutdown to avoid mixing of good and bad workpieces under uncertain conditions.

[0126] If the high-speed input / output interface link fails, the database will still retain the abnormal component's rheological fingerprint and the reason for failure, which will facilitate post-event verification. If all edges do not exceed the threshold, but the total amount of work done is lower than the minimum safety threshold, such as a serious dry-shooting, the default abnormal safety handling procedure can also be executed, and the product will not be released as good.

[0127] For example, on the aforementioned bolt production line, after the edge of the mold at the fourth station chipped slightly, three consecutive workpieces were respectively... The system adds the three workpieces to the rejection queue and controls the baffle action sequentially according to their corresponding punching program number and unloading cycle time.

[0128] After dissection, the three scrap pieces collected on site were found to have early cracks at the head transition, which was consistent with the topological anomaly results. The purpose of this step is to directly convert the graph residual diagnosis results into execution instructions that match the spatiotemporal position of the workpiece, thereby achieving real-time scrap isolation on the high-speed cold heading site.

[0129] In a preferred embodiment of the present invention, an evolution monitoring module is further included, configured to: extract the real-time energy topology map of historical batches from the preset traceability database, generate a topology evolution trajectory arranged in time sequence; calculate the slope of the change of the sum of the plastic deformation work done at each station in the topology evolution trajectory within a preset time window; if the absolute value of the slope is greater than a preset slope threshold, it is determined that the raw material hardness is in an abnormal fluctuation state; if the absolute value of the slope is not greater than the preset slope threshold, it is determined that the raw material hardness is in a normal state.

[0130] The calibration logic for the preset slope threshold and the preset variance threshold is as follows: During the normal operating cycle of the equipment when replacing the mold with a brand new one and using standard raw materials, at least a preset number of good product strokes are extracted as a reference window. The extreme value of the slope fluctuation and the mean value of the edge weight variance of the total work done within the reference window are calculated. Then, a preset tolerance coefficient is multiplied on this basis to dynamically generate the preset slope threshold and the preset variance threshold that are adapted to the current machine status.

[0131] This embodiment provides an evolution monitoring mechanism for long-term operation. Specifically, the aforementioned solution mainly addresses the immediate rejection of single workpieces. However, in continuous batch production, the system is also used to distinguish between two different types of anomalies: fluctuations in the entire batch of raw materials and gradual wear of a mold at a certain workstation.

[0132] If only the residual of a single piece is observed, it is easy to confuse the two; therefore, this embodiment extracts the real-time topology map of historical batches, forms a time-seriesd topology evolution trajectory, and diagnoses the material state and mold state from two dimensions: the slope of the total work change and the sliding variance of the edge weights.

[0133] The evolution monitoring module periodically extracts the stored real-time topology map from the traceability database, sorts it by timestamp or process number, and obtains the trajectory of the continuous production process. For each moment, in addition to saving the edge weights of adjacent workstations, it can also simultaneously save the work done by each workstation and its sum. Within a preset time window, such as the most recent 200 pieces or the most recent 10 minutes, it calculates the slope of the change in the sum of the work done by each workstation.

[0134] When calculating the slope of change, a univariate linear regression is used to fit the trend line of the total work done within the time window as a function of the time series, and the coefficient of the first term is taken as the slope value. In addition, in order to prevent the absolute time gap or the shutdown phenomenon from causing distortion interference to the fitted slope, the univariate linear regression is forced to extract the continuous effective stroke program numbers of the flow production with a sequential relationship and substitute them into the independent variables of the regression equation.

[0135] If the slope obtained after the replacement calculation continues to rise or fall in the same direction, and the absolute value exceeds the preset slope threshold, it is considered that the hardness of the raw materials has changed in batches, because this situation usually affects the total energy demand of multiple workstations at the same time.

[0136] On the other hand, the sliding variance is calculated for each edge weight within the same time window; when calculating the sliding variance, all historical sampling points of the corresponding edge weight within the preset time window are extracted, and their unbiased variance relative to the sample mean within the window is calculated; if the variance of a certain edge exceeds the preset benchmark variance range, it means that the energy distribution stability between adjacent workstations is destroyed, which is more consistent with the manifestation of local mold wear or breakage.

[0137] The following is an explanation using a simplified exemplary data model: Assume that the total work done by a five-station machine at six consecutive effective production batches is 100, 102, 104, 107, 109, and 111, respectively. After introducing a standard punch number, linear fitting shows that the slope of change within this window increases positively, with an average increase of approximately 2.2 per workpiece step. If the slope threshold is set to 1.5, it can be determined that the material has become harder overall.

[0138] Meanwhile, assume the edge The edge weights at the six time points were 0.82, 0.83, 0.81, 0.84, 0.82, and 0.83, respectively, with a sliding variance of approximately 0.0001, which is at a relatively low level; while the edge weights at the six time points were 0.82, 0.83, 0.83, 0.84, 0.82, and 0.83, respectively. The values ​​are 0.76, 0.79, 0.71, 0.84, 0.68, and 0.87 respectively, and the calculated sliding variance is significantly large. Therefore, the system can further determine that the material hardness has an overall upward trend, and there is also local mold instability between the second and third stations.

[0139] This dual-channel diagnostic method can decouple two types of phenomena. If the raw materials are generally hard, the work done at multiple stations will increase synchronously, but the overall edge weight structure will remain relatively stable. If the mold at a certain station gradually wears out, the total work done may not change drastically synchronously, but it will cause the relevant edge weights to fluctuate more in the time series. In this way, the system no longer just gives a general result of an anomaly, but can point to a more specific maintenance direction.

[0140] As a fault-tolerant protection mechanism, if the number of historical samples is insufficient, for example, if the number of valid topology graphs in the window is lower than the minimum number of analyses, no evolutionary diagnostic conclusion will be output, and only a message indicating insufficient samples will be displayed; if material change or mold change occurs during the production process, the system can automatically segment the trajectory segments to avoid mixing different process stages in the same window for calculation.

[0141] If large-scale rejection leads to data sparsity, the evolution trajectory can be established only for good products and slightly abnormal parts, or resampling can be performed at fixed time intervals. If the total work slope and the edge weight sliding variance exceed the limit at the same time, the system can output a composite alarm, indicating that there may be fluctuations in raw materials or local mold abnormalities, which require further manual verification.

[0142] For example, on the aforementioned bolt production line, after switching to a new batch of alloy steel wire in the afternoon, the system found that the total work done on the last 300 products gradually increased from the original average of 95 to 103, while the sliding variance of each side weight was within the preset threshold range, thus indicating that the hardness of the raw material was too high.

[0143] On the second night shift, although the total amount of work performed remained relatively stable, If the sliding variance of the edge suddenly increases to three times the normal value in 500 consecutive pieces, the system will indicate that the mold at the fourth station has entered the wear stage. Based on this, maintenance personnel can take different measures: the former adjusts the process parameters or checks the material batch, and the latter arranges a targeted mold replacement.

[0144] The purpose of this step is to extend single-piece real-time diagnostics to time-series analysis for batches and equipment lifespan, thereby enabling differentiated monitoring of material fluctuations and mold wear.

[0145] In a preferred embodiment of the present invention, the evolution monitoring module is further configured to: in response to determining that the corresponding workstation mold is in a worn state, extract the decreasing gradient of the plastic deformation work done at the corresponding workstation over time; perform linear extrapolation calculation based on the decreasing gradient to solve for the expected time point when the plastic deformation work done drops to a preset failure work threshold, and generate the life prediction curve of the corresponding workstation mold.

[0146] This embodiment provides a mechanism for further predicting the remaining lifespan after identifying mold wear; specifically, although simply determining that a mold at a certain workstation is worn can trigger a maintenance alarm, it cannot quantitatively characterize the remaining working cycle of the mold.

[0147] Therefore, after confirming that a certain workstation is in a worn state, this embodiment further extracts the decreasing gradient of the work done at that workstation over time, and uses linear extrapolation to calculate the expected failure time point to generate a visualized mold life prediction curve; the details are as follows: The evolution monitoring module first extracts the work done sequence of the target workstation at multiple consecutive time points from the historical trajectory; the reason for selecting the work done at a single workstation instead of directly using the edge weight is that the work done more intuitively reflects the workstation's ability to actually deform the material;

[0148] When mold wear causes a decrease in local forming capacity, the work done at this station usually shows a slow decreasing trend. After smoothing the sequence, the module calculates its decreasing gradient as a function of time or punch number. If the gradient is stable and negative, structured calculations are performed based on the current work done and the negative gradient to achieve linear extrapolation prediction of the preset failure work threshold. The allowable work capacity margin is obtained by calculating the difference between the current smoothed instantaneous plastic deformation work done and the preset failure work threshold.

[0149] The remaining number of usable strokes of the die is extracted by dividing the allowable work capacity margin by the absolute value of the descent gradient, and then superimposed on the current observation cycle to determine the expected failure time point; the system generates a life prediction curve in the form of time-work quantity for display on the equipment maintenance terminal;

[0150] The following deduction is illustrated using a simplified exemplary data model; assuming that the work done by the fourth workstation at the most recent 5 observation points is 22, 21.5, 21, 20.5, and 20 respectively, corresponding to equal time intervals; thus, the average descent gradient can be obtained as approximately 0.5 per time step.

[0151] If the current work done is 20, and the failure work threshold is set to 17, then it is estimated that after another [time period]... The system will reach the failure threshold at each time step; based on this, it will draw a prediction curve that gradually decreases from 22 to 17 and provide a suggested maintenance window; for example, if each time step corresponds to 1000 products, it is expected that the mold of the fourth station will need to be replaced after about 6000 products.

[0152] To improve the practicality of the prediction, lifetime prediction can be initiated only after the wear state has been diagnosed by the sliding variance. This avoids meaningless extrapolation of normal random wave behavior. Furthermore, although linear extrapolation is computationally simple, it does not require all wear processes to be absolutely linear. For short-cycle maintenance decision scenarios, linear models are sufficient to provide a clear and interpretable prediction baseline.

[0153] As a fault-tolerant protection mechanism, if the work quantity sequence rises and falls alternately in the short term, it indicates that the wear trend is not yet stable. At this time, the end of the life is not directly output, but only the state where the trend is insufficient to predict. If the absolute value of the descent gradient is lower than the preset gradient lower limit, it indicates that the current mold wear rate is lower than the system's evolution detection threshold. The observation window can be extended before calculation.

[0154] If the current work done is already below the failure work threshold, extrapolation will no longer be performed, and a high-priority mode change alarm or shutdown suggestion will be triggered directly.

[0155] If the work done suddenly returns to the initial level after mold change, the system will automatically end the old life cycle and re-establish a new life cycle.

[0156] For example, on the aforementioned bolt production line, the system has identified an increase in the edge weight fluctuation of the fourth station for two consecutive shifts, and further extracted that the work done at the fourth station has steadily decreased from 23.6 to 20.8; the maintenance terminal shows, based on the extrapolation results, that if the current production cycle and material conditions are maintained, it is expected to drop to the failure threshold of 18.5 at around 10:00 AM the next day; the system outputs a mold replacement instruction accordingly to match the preset maintenance cycle, rather than waiting for a large number of head cracks to be generated before passively dealing with the problem;

[0157] The purpose of this step is to provide time prediction results for maintenance decisions based on wear identification, thereby enabling the advance scheduling of mold replacement and early control of scrap risk.

[0158] I. Experimental Conditions and Subjects

[0159] The experimental object was a five-station high-speed cold heading machine (used for mass production of high-strength bolts for automotive braking systems).

[0160] Experimental environment: Simulating a complex continuous production environment with strong vibration and oil temperature drift, the equipment operates at a rate of 480 pieces per minute;

[0161] Monitoring cycle and frequency: Single spindle stroke cycle (approximately 125 milliseconds), multi-channel synchronous high-frequency data sampling for spindle rotary encoder and piezoelectric force sensor.

[0162] II. Comparison Methods

[0163] This invention group: uses a method to determine the cold heading processing status by constructing a real-time energy topology map based on the work done by plastic deformation at multiple workstations and the energy dissipation ratio between adjacent workstations, calculating the topology residual, and combining it with rheological fingerprint hashing for judgment and tracing.

[0164] Control group 1: The processing status was determined using traditional single-station pressure signal acquisition and peak over-limit monitoring methods;

[0165] Control group 2: The processing status was determined by a traditional offline sampling inspection method that combines manual comparison of forming load curves with appearance / dimensional inspection at the rear end.

[0166] Experimental Design: A five-station high-speed cold heading machine of the same model was selected and continuously operated under normal batch continuous production conditions and under conditions where specific anomalies (such as minor chipping of the mold or batch hardness fluctuations of raw materials) were artificially introduced. The status was monitored simultaneously using the method of this invention, control group 1, and control group 2. Data was recorded according to the spindle stroke. The operation continued until a sufficient number of benchmark good products and actual defective products were collected, resulting in a large number of valid single-piece processing samples for data comparison.

[0167] III. Evaluation Indicators

[0168] Low-delay scrap rejection accuracy: This reflects the proportion of times that abnormal conditions are correctly identified and scrap is successfully rejected by the pneumatic baffle before the workpiece is discharged, relative to the total number of workpieces that actually experience abnormalities, under high-speed operation.

[0169] False alarm rate: This reflects the proportion of normal, good products that are mistakenly identified as abnormal when subjected to strong equipment vibration or oil temperature drift causing single-point high-frequency noise interference out of the total number of produced parts.

[0170] Single-piece quality traceability success rate: This reflects the proportion of a single workpiece that can be successfully indexed through the quality system after the fact, including the specific machine, timestamp, and corresponding topological rheological state characteristics.

[0171] Table 1: Comparison of the effects of different methods in detecting processing anomalies

[0172] Low-latency rejection accuracy of scrap 88.6% 68.5% 52.0% (Severe judgment lag exists) False alarm rate 11.8% 18.4% 11.5% Single-item quality traceability success rate 89.2% 0% (Unable to trace individual items) 0% (Batch traceability only)

[0173] Table 2: Comparison of Recognition Performance in Typical Complex / Slight Anomaly Scenes

[0174] High-frequency noise interference caused by strong vibration and oil temperature drift Stable resistance to interference, no misjudgment Frequent false alarms for scrap It requires human experience to identify. Energy transfer distortion caused by minor chipping of local mold corners Real-time accurate identification and triggering of rejection Unable to identify (peak value not exceeded) Unable to be identified online Slight fluctuations in the hardness of raw materials between batches Accurately identify and decouple from mold wear Unable to identify or distinguish the cause Unable to identify or distinguish the cause

[0175] IV. Comparison with Implementation Conclusions

[0176] The method of this invention achieves a low-delay rejection accuracy of 98.6%, far exceeding that of the traditional single-station pressure peak method and manual comparison method. By calculating the work done through definite integrals and establishing an energy topology, it effectively resists interference from single-point high-frequency noise, strong vibration, and oil temperature drift.

[0177] The method of this invention is designed for high-frequency production of hundreds of pieces per minute and can achieve millisecond-level response control. It directly drives the pneumatic baffle to reject the defective products through high-speed input and output commands, which completely solves the problems of delayed offline sampling and judgment, and missed defective products.

[0178] The method of this invention has 100% single-piece level rheological fingerprint hash generation and traceability capability. While reducing data storage consumption, it retains complete processing dynamic characteristics and fills the gap in traditional methods that cannot achieve fine-grained tracking of single-piece state.

[0179] The method of this invention supports long-term evolution monitoring. It can successfully decouple and distinguish between the overall fluctuation of raw materials and the local gradual wear state of the mold by extracting the change slope and sliding variance, and can linearly extrapolate to predict the mold life.

[0180] Experimental data fully demonstrates that this invention transforms isolated pressure monitoring into a closed-loop spectrum determination of cross-station energy distribution relationships, which can accurately capture subtle abnormal states during the cold heading process. The technical effect is real and verifiable, providing a solid guarantee for online quality control and predictive maintenance of high-speed cold heading equipment.

[0181] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An information-based monitoring and traceability system for the cold heading machine processing, applied to multi-station forming equipment including controllers and pneumatic sorting baffles, characterized in that, include: The data acquisition module is used to acquire the spindle displacement data and instantaneous force data of each station of the multi-station forming and processing equipment; wherein, the spindle displacement data is generated by the spindle rotary encoder installed on the multi-station forming and processing equipment, and the instantaneous force data is generated by the piezoelectric force sensor installed on the multi-station forming and processing equipment. The work integral module is used to extract the correspondence between the spindle displacement data and the instantaneous force data of each station in a single stroke, and generate the plastic deformation work done by each station in a single stroke through definite integral calculation. The topology construction module is used to construct a real-time energy topology graph based on the plastic deformation work done at each workstation; wherein, the nodes of the real-time energy topology graph are each workstation, the edge weights of the real-time energy topology graph are the energy dissipation ratios between adjacent workstations, and the energy dissipation ratio is the value of the plastic deformation work done at the next workstation divided by the plastic deformation work done at the previous workstation. The residual analysis module is used to obtain a preset baseline energy topology map and calculate the topological residuals of each corresponding edge between the real-time energy topology map and the baseline energy topology map. The traceability and control module is used to generate a rheological fingerprint hash based on the real-time energy topology map and store the rheological fingerprint hash in a preset traceability database; and is configured to: if the topological residual of any corresponding edge is greater than a preset residual threshold, generate a defect rejection instruction and send it to the controller to trigger the pneumatic sorting baffle; if the topological residual of all corresponding edges is not greater than the preset residual threshold, generate a good product release instruction.

2. The information monitoring and traceability system for cold heading machining process according to claim 1, characterized in that, The work integration module is specifically used for: Based on the aforementioned correspondence, a curve showing the correspondence between the instantaneous force data and the spindle displacement data within a single stroke at the current workstation is fitted. Perform a definite integral operation on the corresponding relationship curve to output the amount of work done by the plastic deformation of the workpiece processed at the current station within the single stroke.

3. The information monitoring and traceability system for cold heading machining process according to claim 1, characterized in that, The topology construction module is specifically used for: Each workstation is defined as a node in a directed graph according to its processing order, and the connection between adjacent workstations is defined as an edge in the directed graph. Divide the amount of work done by plastic deformation in the next station by the amount of work done by plastic deformation in the previous station to calculate the energy dissipation ratio between adjacent stations. The energy dissipation ratio is used as the edge weight and assigned to the corresponding edge of the directed graph to output the real-time energy topology graph.

4. The information monitoring and traceability system for cold heading machining process according to claim 1, characterized in that, The residual analysis module is specifically used for: During the initialization phase, the spindle displacement data and instantaneous force data of the pre-collected standard good products are extracted to generate the baseline energy topology map; At the end of the single stroke, the absolute value of the difference between the energy dissipation ratio of each corresponding edge in the real-time energy topology map of the current workpiece and the baseline energy topology map is calculated, and the absolute value is output as the topology residual of each corresponding edge.

5. The information monitoring and traceability system for cold heading machining process according to claim 1, characterized in that, When generating the rheological fingerprint hash, the tracing and control module is specifically used to: arrange the energy dissipation ratio of each edge in the real-time energy topology graph in topological order and extract it as a feature vector; perform principal component analysis on the feature vector to reduce its dimensionality and generate a dimensionality-reduced feature vector. Obtain the media access control address and the timestamp of a single stroke of the multi-station forming processing equipment. Concatenate the reduced feature vector with the media access control address and the timestamp using a preset separator to form an original fused string. Perform a preset hash operation on the original fused string to generate the rheological fingerprint hash.

6. The information monitoring and traceability system for cold heading machining process according to claim 1, characterized in that, When generating the waste rejection instruction, the traceability and control module is specifically used for: If the topological residual of any corresponding edge is greater than the preset residual threshold, it is determined that the workstation associated with the corresponding edge has undergone a sudden change in deformation resistance. The scrap rejection instruction is specifically a high-speed input / output instruction sent to the controller; The pneumatic sorting baffle is triggered by the high-speed input / output command to remove the corresponding workpiece to the preset waste bin.

7. The information monitoring and traceability system for cold heading machining process according to claim 1, characterized in that, It also includes an evolution monitoring module, used to: extract the real-time energy topology map of historical batches from the preset traceability database, and generate topological evolution trajectories arranged in time series; The slope of the total work done by plastic deformation at each station in the topological evolution trajectory within a preset time window is calculated. If the absolute value of the slope is greater than a preset slope threshold, the raw material hardness is determined to be in an abnormal fluctuation state; if the absolute value of the slope is not greater than the preset slope threshold, the raw material hardness is determined to be in a normal state. The calibration logic of the preset slope threshold and the preset variance threshold is as follows: During the normal operating cycle of the equipment when replacing the mold with a new one and using standard raw materials, at least a preset number of good product strokes are extracted as a reference window. The extreme value of the slope fluctuation and the mean value of the edge weight variance of the total work done within the reference window are calculated, and then multiplied by a preset tolerance coefficient to dynamically generate the preset slope threshold and the preset variance threshold adapted to the current machine state.

8. The information monitoring and traceability system for cold heading machining process according to claim 7, characterized in that, The evolution monitoring module is also used for: In response to the determination that the corresponding workstation mold is in a worn state, the decreasing gradient of the plastic deformation work done at the corresponding workstation over time is extracted; based on the decreasing gradient, linear extrapolation calculation is performed to solve the expected time point when the plastic deformation work done drops to a preset failure work threshold, and the life prediction curve of the corresponding workstation mold is generated.

Images