Visual positioning detection data fusion system for energy storage integrated wire laser welding

Abstract

This invention relates to the field of laser welding and industrial automation control technology, specifically a visual positioning and detection data fusion system for laser welding in energy storage production lines. The system includes: a working condition setting module, which determines parameters such as observation strategy, sensor priority, and sampling rate based on working conditions and system occupancy levels; a multi-source acquisition module, which collects multi-source data such as visual, temperature, acoustic, photoelectric, and pose data and encapsulates them into event packets; a synchronization mapping module, which corrects the time of each event packet and maps it to weld arc length coordinates; a data fusion module, which judges data validity, updates credibility, and generates fusion results and risk scores; a control generation module, which generates scheduling parameters and control command templates based on risk scores and system status; and a control execution recording module, which generates control commands constrained by process conditions, sends them to actuators such as lasers and robots, and records the results. This invention effectively improves the reliability of multi-source data fusion results.

Description

Technical Field

[0001] This invention relates to the field of laser welding and industrial automation control technology, specifically a visual positioning and detection data fusion system for laser welding of energy storage production lines. Background Technology

[0002] The quality monitoring and control of the laser welding station in the energy storage production line essentially relies on multi-source sensors to collect state parameters during the execution process, determine abnormalities, and drive the equipment to adjust its response. Existing data acquisition and monitoring often use multiple sensors such as vision, temperature, acoustics, and photoelectric sensors to obtain physical characteristic data of the operating cycle. In continuous operation scenarios, the aforementioned multi-source acquisition and control faces the following limitations: each sensor channel has different sampling, buffering, processing, and transmission delays, and different reference benchmarks. Direct fusion of heterogeneous data without strict spatiotemporal mapping will introduce serious physical space deviations; sensor saturation, occlusion, voids, frame drops, or communication anomalies often occur during operation, making it impossible to distinguish between data loss and quality degradation. Unconditional fusion and lack of credibility assessment will lead to overall judgment deviations; massive amounts of fine-grained sensor data are prone to causing bus congestion and queuing delays for critical measurement records when the system load increases; relying solely on a single comprehensive risk score to directly drive actuators such as lasers, galvanometers, or robots will cause frequent adjustments and control oscillations when the data is unstable or close to the equipment's limits. Achieving refined spatiotemporal synchronization mapping of multi-source heterogeneous data, dynamically evaluating the reliability of underlying detection quantities while taking into account system occupancy levels, and then generating an anti-oscillation closed-loop control strategy based on process boundaries and interlocking conditions are technical problems that need to be solved in current automatic data acquisition and control applications. Summary of the Invention

[0003] The purpose of this invention is to provide a visual positioning and detection data fusion system for laser welding of energy storage production lines, so as to solve the technical problems in the prior art that the laser welding quality monitoring effect is poor due to inaccurate spatiotemporal mapping of multi-source heterogeneous data, lack of credibility assessment, and easy oscillation of control strategy.

[0004] The objective of this invention can be achieved through the following technical solutions: The visual positioning and detection data fusion system for laser welding in energy storage production lines includes: The operating condition setting module is used to determine the observation strategy, sensor priority, sampling rate, cache budget, and abnormal escalation conditions based on the operating condition information and the system occupancy level represented by the processing queue length, packet loss rate, and cache usage ratio. The multi-source acquisition module communicates with vision sensors, temperature sensors, acoustic sensors, photoelectric sensors, encoders, and robot controllers via communication interfaces. It is used to acquire coaxial images, side view images, 3D contours, temperature, photoelectric data, acoustic emission data, encoder position, and robot pose data as detection quantities and encapsulates them into event packets. The synchronization mapping module, connected to the multi-source acquisition module, is used to correct the timing of event packets and map the event packets to weld arc length coordinates; The data fusion module, connected to the synchronization mapping module, is used to determine the validity of the detection volume within the event packet, update the confidence field, and generate fusion results and risk scores. The control generation module, connected to the data fusion module, is used to generate state switching parameters, sensor scheduling parameters, and instruction templates based on risk scores, updated confidence fields, anomaly duration, and system occupancy levels. The control execution recording module, connected to the control generation module, is used to generate control commands that are restricted by process boundaries and interlocking conditions based on the instruction template, send them to the laser, galvanometer, robot and programmable logic controller via the industrial bus, and record the execution results.

[0005] In one alternative, the event package includes workpiece identifier, weld point identifier, sensor identifier, trigger sequence number, source timestamp, corrected timestamp, encoder position and robot pose data, validity flag, initial confidence and updated confidence fields; The multi-source acquisition module writes the source timestamp, trigger sequence number, encoder position and robot pose data, validity markers and initial confidence level. The validity markers include saturation markers, occlusion markers, hole markers, frame loss markers and communication anomaly markers. The synchronization mapping module writes the corrected timestamp.

[0006] In one alternative approach, the process of the synchronization mapping module correcting the time of the event packet is as follows: obtain the source timestamp of the sensor channel corresponding to each sensor in the multi-source acquisition module; The sampling delay, buffer delay, processing delay, and transmission delay corresponding to the sensor channel are determined respectively; the sampling delay, buffer delay, processing delay, and transmission delay are subtracted from the source timestamp to obtain the corrected timestamp; the sampling delay, buffer delay, processing delay, and transmission delay are updated by sliding estimation based on the hardware trigger pulse.

[0007] In one alternative approach, the synchronization mapping module maps each event packet to weld arc length coordinates as follows: The encoder position and robot pose data are interpolated based on the corrected timestamp; the interpolation results are projected onto the weld reference trajectory, and the weld arc length coordinates corresponding to the event packet are calculated based on the initial calibration parameters used for spatial calibration and the online drift correction parameters used for drift compensation. Perform geometric matching on the tooling reference edge region in the coaxial image or side view image. When the obtained translation deviation, equivalent edge displacement and projection residual deviation are all less than or equal to the spatial allowable range, update the online drift correction parameters according to the obtained geometric matching results. When the translation deviation, equivalent edge displacement, or projection residual deviation exceeds the allowable range in space, the online drift correction parameters are stopped from being updated.

[0008] In one alternative approach, the data fusion module processes the data by acquiring the time deviation, spatial deviation, and multi-sensor trend consistency parameters of the event packets. The time deviation is calculated based on the difference between the corrected timestamp and the recorded time at adjacent locations. The spatial deviation is calculated based on the projection residual deviation and the equivalent edge displacement. The multi-sensor trend consistency parameters are statistically analyzed based on the changing direction of multiple sensor channels within the same weld window. The validity of the detection quantity is determined based on the time deviation, spatial deviation and multi-sensor trend consistency parameters, and each detection quantity in the event packet is marked as valid or invalid. The confidence field is updated based on the event packet quality and multi-sensor trend consistency parameters, and the fusion result is generated based on the effective detection volume and confidence.

[0009] In one alternative approach, the risk score is calculated by extracting the corresponding molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters, and motion characteristic parameters from the fusion results. A risk score is calculated based on the extracted feature parameters. The risk score is compared with a preset risk threshold. When the risk score is greater than or equal to the preset risk threshold, the corresponding weld window divided along the weld arc length coordinate is determined to have a risk. When the risk score is less than the preset risk threshold, the corresponding weld window divided along the weld arc length coordinate is determined to have no risk. Different preset risk thresholds correspond to different trajectory stages.

[0010] In one alternative, the control generation module constructs and runs a state machine consisting of stable welding, suspected anomaly, enhanced observation, parameter suppression, shutdown confirmation, and anomaly recovery. The next state is determined based on the current state, risk score, duration of the anomaly, and system occupancy level. When the risk score is greater than or equal to the preset risk threshold and the duration of the anomaly reaches or exceeds the preset duration threshold, if the system occupancy level is lower than the preset occupancy threshold, the system will switch to enhanced observation; otherwise, it will switch to suspected anomaly. When the anomaly persists and both the main vision sensor and at least one auxiliary sensor determine that there is an anomaly in the same weld window, the process is switched to a shutdown confirmation. When the risk score is lower than the preset recovery threshold and continues for the preset recovery time, and the preset sensor confidence level is higher than the preset confidence level threshold, the system will switch to abnormal recovery.

[0011] In one alternative approach, the operating condition setting module calculates and determines the system occupancy level, and the control generation module calls the system occupancy level. When the system occupancy level is lower than the preset occupancy threshold, the multi-source acquisition module collects and sends data according to the current observation strategy. When the system occupancy level reaches or exceeds the preset occupancy threshold, the sampling rate under stable welding is reduced, and the buffering priority and sending priority of event packets corresponding to risk weld windows are increased. In cases of suspected anomalies or enhanced observation, increase the priority and sampling rate of sensors corresponding to the current anomaly type.

[0012] In one alternative, the process of the control generation module generating control instructions is as follows: acquiring preset process window parameters including power target value, weld offset tolerance zone and temperature target range; The molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters, or motion characteristic parameters in the fusion result are used as target control variables, and the deviation of the target control variables relative to the corresponding process window parameters is calculated. Each target control variable is multiplied by its corresponding process mapping coefficient to obtain the laser power correction, galvanometer offset correction, or robot trajectory correction; the corresponding correction is then limited according to the equipment's allowable lower and upper adjustment limits. After writing the correction amount after the amplitude limit into the instruction template, the control instruction is generated; among them, each process mapping coefficient and process boundary are determined according to the preset calibration parameters.

[0013] In one alternative, the control execution recording module sends control commands to the laser, galvanometer, robot, and programmable logic controller, receives corresponding execution confirmation signals or position feedback status, and records the source timestamp, corrected timestamp, validity mark, confidence level, fusion result, risk score, state transition, control command, and execution result according to the workpiece identification and weld point identification. The operating condition setting module updates the shutdown confirmation sensitivity parameter based on the execution failure rate, and updates the observation strategy, sensor priority, sampling rate, cache budget, and abnormal upgrade conditions for the next welding task based on the execution results, feedback status, and welding quality results.

[0014] The beneficial effects of this invention are: 1. This invention determines the time delay of each item and corrects the source timestamp by using a synchronization mapping module. Then, it interpolates each event packet and maps it to the weld arc length coordinate. This solution solves the time difference problem caused by different sampling, buffering, processing and transmission delays of each sensor channel. It also eliminates the physical space deviation introduced by different reference benchmarks and ensures the accurate fusion of heterogeneous data under the same spatiotemporal benchmark. 2. The multi-source acquisition module of this invention encapsulates validity markers such as saturation, occlusion, holes, and frame loss in the event packet. The data fusion module further updates the confidence field by combining time deviation, spatial deviation, and multi-sensor trend consistency. This mechanism can accurately distinguish between data loss and data quality degradation that occur during continuous operation, thereby avoiding the overall judgment deviation caused by blindly fusing missing or low-quality data and effectively improving the reliability of multi-source data fusion results. 3. The working condition setting module of this invention evaluates the system occupancy level in real time. When the load reaches or exceeds the preset threshold, the system actively reduces the sampling rate of the stable welding section and increases the caching and transmission priority of the risk weld window event packets. This dynamic scheduling mechanism effectively solves the bus congestion problem that is easily caused by massive fine-grained sensor data. It ensures that key measurement records can be processed first under limited system computing and transmission resources, avoiding delays. 4. The control generation module of this invention performs multi-level judgments by constructing a state machine that includes stable welding, suspected abnormality, and parameter suppression, and generates the final correction instruction by combining preset process window parameters and equipment adjustment limits. This solution changes the limitation of relying solely on a single risk score to directly drive the laser or robot and other actuators. It effectively prevents frequent adjustments and control oscillations caused when the detection data is temporarily unstable or the correction amount is close to the physical limit of the equipment, thus ensuring the stability and safety of the welding process. Attached Figure Description

[0015] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a control block diagram of a visual positioning and detection data fusion system for laser welding of an energy storage production line provided in an embodiment of this application. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 A visual positioning and detection data fusion system for laser welding in an energy storage production line includes: The operating condition setting module is used to determine the observation strategy, sensor priority, sampling rate, cache budget, and abnormal escalation conditions based on the operating condition information and the system occupancy level represented by the processing queue length, packet loss rate, and cache usage ratio. The multi-source acquisition module communicates with vision sensors, temperature sensors, acoustic sensors, photoelectric sensors, encoders, and robot controllers via communication interfaces. It is used to acquire coaxial images, side view images, 3D contours, temperature, photoelectric data, acoustic emission data, encoder position, and robot pose data as detection quantities and encapsulates them into event packets. The synchronization mapping module, connected to the multi-source acquisition module, is used to correct the timing of event packets and map the event packets to weld arc length coordinates; The data fusion module, connected to the synchronization mapping module, is used to determine the validity of the detection volume within the event packet, update the confidence field, and generate fusion results and risk scores. The control generation module, connected to the data fusion module, is used to generate state switching parameters, sensor scheduling parameters, and instruction templates based on risk scores, updated confidence fields, anomaly duration, and system occupancy levels. The control execution recording module, connected to the control generation module, is used to generate control instructions based on the instruction template, which are restricted by process boundaries and interlocking conditions. These instructions are then sent to the laser, galvanometer, robot, and programmable logic controller via the industrial bus, and the execution results are recorded. In this embodiment of the invention, the event package includes workpiece identifier, weld point identifier, sensor identifier, trigger sequence number, source timestamp, corrected timestamp, encoder position and robot pose data, validity flag, initial confidence level and updated confidence level fields; The multi-source acquisition module writes the source timestamp, trigger sequence number, encoder position and robot pose data, validity markers and initial confidence level. The validity markers include saturation markers, occlusion markers, hole markers, frame loss markers and communication anomaly markers; the synchronization mapping module writes the corrected timestamp. The process of the synchronization mapping module correcting the time of the event packet is as follows: obtain the source timestamp of the sensor channel corresponding to each sensor in the multi-source acquisition module; The sampling delay, buffer delay, processing delay, and transmission delay corresponding to the sensor channel are determined respectively; the sampling delay, buffer delay, processing delay, and transmission delay are subtracted from the source timestamp to obtain the corrected timestamp; the sampling delay, buffer delay, processing delay, and transmission delay are updated by sliding estimation based on the hardware trigger pulse.

[0018] The system is deployed at the laser welding station of the entire production line for energy storage cells, busbars or connecting pieces. The laser, galvanometer, robot, programmable logic controller and edge computing node are connected through industrial bus and communication interface. Welding tasks are executed continuously in the order of workpiece and weld point. Before each welding task begins, the working condition setting module first reads the current workpiece identifier, weld type, material combination, trajectory stage, historical defect record, equipment temperature and node load, and combines the processing queue length, packet loss rate and cache usage ratio to form the current system occupancy level; Based on the above information, the operating condition setting module writes the observation strategy corresponding to this task into the shared configuration area, including sensor priority, sampling rate, cache budget and abnormal upgrade conditions, so that subsequent acquisition and control links can directly call it to avoid subsequent links repeatedly reading upstream operating condition information. In this workstation, the multi-source acquisition module maintains communication connections with the coaxial camera, side-view camera, 3D profilometer, temperature sensor, photoelectric sensor, acoustic emission sensor, encoder, and robot controller. In this embodiment, the acoustic sensor is implemented as an acoustic emission sensor. After the welding enters the operation cycle, each sensor outputs corresponding data under the drive of laser triggering, trajectory position, or programmable logic controller events. The multi-source acquisition module does not force the reconstruction of various types of data in a unified format. Instead, it encapsulates the data in an event package manner, writing the workpiece identifier, weld point identifier, sensor identifier, trigger sequence number, source timestamp, encoder position or robot pose data into the same record, while also writing the validity flag and initial confidence level. The validity marker is used to retain whether there is saturation, occlusion, holes, frame loss or communication anomaly within the period; the subsequent judgment module distinguishes between data missing and data quality degradation based on the validity marker, and removes the detection quantity that does not meet the preset judgment conditions. To avoid the event package remaining only at the abstract container level, the multi-source acquisition module sequentially processes the acquisition results of each channel into five running states: pending writing, pending time calibration, pending mapping, pending fusion, and executed record. Each channel only retains a sequential cache covering the most recent consecutive weld seam windows locally, and writes them incrementally according to the trigger sequence number. When a new event packet will cause the cache of this channel to exceed the current cache budget, the system will not clear it completely. Instead, it will prioritize retaining event packets that have been marked as risk windows, pending position record windows, and those that have not yet completed fusion judgment. The earliest event packet in the stable welding segment that has completed execution records and has no traceability requirement will be truncated first. If multiple channels arrive simultaneously in the same processing cycle, they will first enter the main processing link according to the solder joint order, then according to the risk window priority, and finally according to the sensor priority. When contention occurs under the same priority, the earlier trigger sequence number takes precedence; the system executes the lifecycle allocation, eviction, and congestion retention control of event packets from acquisition to fusion to control records in the order of the trigger sequence number; When a sensor does not return data in the current cycle, the multi-source acquisition module does not terminate the generation of the entire packet. Instead, it retains the workpiece identifier, weld point identifier, sensor identifier and trigger sequence number in the corresponding event packet and sets a frame loss flag or communication abnormality flag. Even if location information is available but measurement values ​​are lacking, the event packet can still proceed to subsequent processes to maintain data continuity on the weld coordinates. If both the encoder position and the robot pose are unavailable, the current event packet will only retain the source timestamp and the exception flag, and will be transferred to the waiting queue in the synchronization mapping module. After time correction is completed, the event packet is only allowed to complete the interpolation mapping if two valid location records located before and after its corrected timestamp are obtained simultaneously in the subsequently recovered location records. If only one side of the location record is restored, or if the corrected timestamp falls outside the range of currently available location records, then extrapolation will not be performed, and we will only continue to wait for the location records to be completed. If a valid position record before and after the corrected timestamp cannot be formed within the preset waiting range, the event packet exits the main processing link, but the record information is still retained for the execution recording module to trace; the preset position interpolation algorithm is executed on the event packet waiting to be completed through the aforementioned configuration. After receiving the event packet, the synchronization mapping module performs time correction. Specifically, it obtains the source timestamp for each sensor channel and reads the sampling delay, buffer delay, processing delay, and transmission delay corresponding to that channel. The aforementioned latency values ​​are derived from a combination of equipment calibration values, historical statistical values, and online update values. The online update part is based on hardware trigger pulses and is continuously corrected in a sliding manner, so that the channel latency can be adjusted in response to buffer backlog, bus congestion, or changes in processing load during the soldering process. The synchronization mapping module subtracts the aforementioned delays from the source timestamp to obtain the corrected timestamp, and writes this time back to the event packet. After this processing, data from different channels within the same weld window can be grouped into the same time range closer to the sampling position. If a channel is missing some delay parameters in the current period, the system does not directly use null values ​​in the calculation. Instead, it prioritizes reading the stable record from the previous valid period of the channel as a substitute and retains the corresponding mark in the event packet. If a usable delay update value cannot be obtained for several consecutive cycles, the event packet of that channel can still be retained, but its initial credibility will be reduced in order to reduce the impact of that channel during subsequent fusion; based on the set initial credibility, the weight contribution of abnormal time data of a single channel to the overall packet judgment result is controlled. In this implementation, the initial confidence level is not arbitrarily assigned, but is written in stages according to the health status of the sensor channel and the completeness of the event packets in this cycle; Specifically, when the channel communication is normal, the trigger sequence number is continuous, the location information is complete and there are no saturation, obstruction, holes, frame loss and communication abnormality markers, write a high initial confidence level; When a single preset general anomaly flag exists but the main measurement value can still be obtained, write the initial confidence level to medium; when there are cases of frame loss, communication anomalies, missing replacements for continuous delays, or incomplete location information, write the initial confidence level to low. The aforementioned distinctions between high-end, mid-range, and low-end equipment are determined by the calibration records at the time of equipment delivery and the continuous operation records on site, and remain fixed within the same production line. They are only recalibrated after the replacement of sensors, communication links, or significant changes in the process cycle. The initial confidence levels of each channel obtained by the subsequent fusion module have a unified assignment standard; To avoid the high-end, mid-end, and low-end ratings remaining merely textual descriptions, under the same calibration version on the same production line, the initial confidence level is fixed as a three-level discrete value: 0.9 for high-end, 0.6 for mid-end, and 0.3 for low-end. When multiple exception flags appear in an event packet at the same time, multiple levels are no longer stacked; instead, the most unfavorable level is written directly. Correspondingly, a single minor anomaly refers to the occurrence of only one of the following situations and the main measurement value is still retained: the trigger sequence number is discontinuous for a single time but recovers later, a single-cycle delay is substituted, the difference between the main location source and the verification location source exceeds the allowable range but projection can still be completed, or the image has local occlusion but the target area still retains continuous edges; If any of the following conditions exist, the initial confidence level will be directly written to a low level: frame loss flag, communication abnormality flag, time delay replacement for two or more consecutive cycles, encoder position and robot pose are both unavailable, or the main measurement area is occluded to the point that the main measurement value of the current cycle cannot be obtained. During the event packet formation phase, based on the aforementioned fixed triggering conditions and fixed write values, an initial credibility write operation is performed. Sampling delay is primarily determined based on sensor specifications and trigger time synchronization records; buffering delay is determined based on the statistical difference between the time a packet enters and leaves the buffer; processing delay is determined based on the statistical difference between the time a packet enters the algorithm processing queue and the time a result is generated; transmission delay is determined based on the statistical difference between the time a packet is sent and the time a packet is received. For each channel, if the online statistical value falls within the most recent continuous stable range of that channel, the online statistical value is used as the current cycle delay; if the difference between the online statistical value and the most recent continuous stable range is greater than the preset fluctuation threshold, the previous effective cycle delay is used for correction, and that cycle is marked as a delay to be confirmed. If the deviation continues for a preset number of consecutive cycles, the channel delay recalibration is triggered. The preset number of consecutive cycles is determined by the channel sampling rate and welding cycle time, so that both high-sampling channels and low-sampling channels can complete the anomaly confirmation in a similar time. The most recent continuous stable interval is determined by the statistical results of the most recent continuous effective period of the same channel, specifically using the delay records of no less than five recent effective periods; when the current online statistical value is between the maximum and minimum values ​​of this set of records, it is considered to fall into the stable interval. When the deviation exceeds this range but does not occur consecutively, it is considered a single-cycle deviation; when three consecutive cycles exceed this range, it is considered a continuous deviation and triggers time delay recalibration; correspondingly, the corrected timestamp is determined by the following formula: ; For cycles where encoder position and robot pose are available simultaneously, this implementation prioritizes using a position source consistent with the current weld reference trajectory calibration source as the primary position source, and another position source as the verification position source. When the projection difference between the primary position source and the verification position source within the current weld window is within the allowable range, the event packet is written according to the primary position source. When the difference between the two exceeds the allowable range, instead of directly averaging and merging, the write result of the primary location source is retained and the initial confidence of the event packet is lowered. At the same time, the location source inconsistency mark is written into the event packet for further processing in the subsequent synchronization mapping and fusion stages. To address the issue of concurrent entry and processing blockage of continuous heterogeneous data streams, the synchronization mapping module only performs time correction on event packets that are in the pending time state and whose trigger sequence numbers have not been rolled back. For event packets in the same channel whose trigger sequence number difference is greater than the preset sequence number threshold, the original record is retained and mapping is temporarily suspended. Only after obtaining a continuous record that can prove that the sequence number difference is a real beat rather than a duplicate to the packet will it be restored to enter the main processing link. If there are event packets in the waiting queue that have completed time synchronization but have not yet acquired enough position records, these event packets will remain in the corrected time order and will not be inserted in reverse order with newly arrived stable window event packets. When the number of waiting events approaches the cache budget, the system prioritizes releasing abnormal event packets that have exited the current solder joint main judgment window and are only used for tracing. The system prioritizes retaining event packets that need to be mapped based on the preset congestion release rules, and allocates the execution priority of the acquisition, time synchronization and waiting to be mapped stages. In this implementation, the working condition setting module, multi-source acquisition module, synchronous mapping module, data fusion module, control generation module, and control execution record module work sequentially according to the above connection relationship; Event packets are generated during the acquisition phase, and corrected timestamps are added during the synchronization phase before entering the subsequent weld coordinate mapping and detection judgment process. Through this processing sequence, a clear division of labor is maintained between acquisition, time correction and subsequent judgment. The previous step only needs to write the data fields it is responsible for, and the next step can directly read the existing fields to continue processing. This is suitable for continuous, multi-batch task operation in the entire welding line.

[0019] In this embodiment of the invention, the process by which the synchronization mapping module maps each event packet to the weld arc length coordinate is as follows: interpolating the encoder position and robot pose data according to the corrected timestamp; projecting the interpolation result onto the weld reference trajectory; and calculating the weld arc length coordinate corresponding to the event packet based on the initial calibration parameters used for spatial calibration and the online drift correction parameters used for drift compensation. Perform geometric matching on the tooling reference edge region in the coaxial image or side view image. When the obtained translation deviation, equivalent edge displacement and projection residual deviation are all less than or equal to the spatial allowable range, update the online drift correction parameters according to the obtained geometric matching results. When the translation deviation, equivalent edge displacement, or projection residual deviation exceeds the allowable range in space, stop updating the online drift correction parameters; The data fusion module processes data by acquiring the time and space deviations of the event packets and the trend consistency parameters of the multiple sensors. The time deviation is calculated based on the difference between the corrected timestamp and the recorded time at adjacent locations. The spatial deviation is calculated based on the projection residual deviation and the equivalent edge displacement. The multi-sensor trend consistency parameters are statistically analyzed based on the changing direction of multiple sensor channels within the same weld window. The validity of the detection quantity is determined based on the time deviation, spatial deviation and multi-sensor trend consistency parameters, and each detection quantity in the event packet is marked as valid or invalid. The confidence field is updated based on the event packet quality and multi-sensor trend consistency parameters, and the fusion result is generated based on the effective detection volume and confidence. The risk score calculation process is as follows: extract the corresponding molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters, and motion characteristic parameters from the fusion results; calculate the risk score based on each extracted characteristic parameter; The risk score is compared with a preset risk threshold. When the risk score is greater than or equal to the preset risk threshold, it is determined that there is a risk in the corresponding weld window divided along the weld arc length coordinate. Different trajectory stages correspond to different preset risk thresholds.

[0020] When the laser welding station of the energy storage production line is running continuously, time calibration alone is not enough to support the common use of data from different sensors. After obtaining the corrected timestamp, the synchronization mapping module extracts two valid position records before and after the timestamp from the encoder position sequence or robot pose sequence, and performs interpolation to obtain the position result corresponding to the event packet. Let the corrected timestamp of the previous location record be... The corrected timestamp of the next location record is The corresponding trajectory positions are respectively and When performing position correction on the event packet, the system decomposes the robot pose into translational and rotational quaternion components, and performs interpolation calculations independently for each. The interpolation calculation formula for the translational component is as follows: ; In the formula, The corrected timestamp for the event packet; , These are the corrected timestamps for the previous and next position records, respectively; , These are the translation components corresponding to the positions p1 of the previous trajectory and p2 of the next trajectory, respectively. The translation component is calculated after interpolation of the current event packet; For the rotation quaternion part of the attitude, the spherical linear interpolation algorithm is used to calculate the corrected time step. The corresponding spatial attitude is used to ensure the mathematical orthogonality constraints and physical effectiveness of the three-dimensional rotation; when lie in and Perform the above interpolation when there is a gap; when If the record is earlier than the earliest available location record or later than the latest available location record, extrapolation is not performed. Instead, the event packet is marked as insufficient location interpolation and the process is redirected to a branch that waits for completion or traces back to the record. The interpolated position is projected onto the pre-established weld reference trajectory. The initial calibration parameters include fixed spatial transformation parameters between the camera, tooling, encoder, and robot coordinate systems. The online drift correction parameters include translation correction and rotation correction. The projected nearest point position is converted into weld arc length coordinates using the above parameters and written back to the event packet; the event packet only enters the main judgment window of the weld point when the obtained arc length coordinates fall within the preset start and end arc length range of the current weld point; event packets that exceed this range do not enter the main judgment window, but their records are retained for traceability. For coaxial images and side view images, the system extracts edge segments within the pre-calibrated tooling reference edge area and performs geometric matching between the edge segments of the current cycle and the calibrated reference edge segments to obtain translational deviation, rotational deviation, and projection residual deviation. The rotational deviation is converted into equivalent edge displacement according to the reference distance of the tooling reference edge area, and the spatial deviation is taken as the larger value between the projection residual deviation and the equivalent edge displacement. Both are recorded using length units. When translation deviation When the equivalent edge displacement and the remaining projected deviation do not exceed the spatial allowable range corresponding to the current weld point type, the online drift correction parameters are updated using the matching results of this cycle. The specific pose update formula is as follows: ; in, This is the updated drift correction vector for this period. This is the previous cycle drift correction vector, which includes translation and rotation components. The system's preset two-dimensional diagonal gain matrix includes translation gain coefficients and rotation gain coefficients. Due to translational deviation The two-dimensional heterogeneous space deviation column vector formed by the equivalent edge displacement is obtained through the gain matrix. The dimensional transformation ensures the consistency of dimensions in matrix multiplication and the closed loop of the algorithm; If any of these parameters exceeds the preset spatial deviation threshold, the online drift correction parameters of the previous effective period are retained, and the matching results of the current period are not used for updating. If the coaxial image cannot form a valid match but the side view image can form a valid match, only the side view image result is used for updating, and vice versa. If neither of them can form a valid match, only the arc length coordinate conversion is performed. After receiving the event packet with the weld arc length coordinates, the data fusion module checks the validity marker, time deviation, spatial deviation, and multi-sensor trend consistency in sequence. The time deviation is recorded as 0 when the corrected timestamp is between the two position records used for interpolation; when it is outside the interval, it is recorded as the absolute time difference of the time when the corrected timestamp exceeds the nearest boundary position record. The trend consistency of multiple sensors is determined based on the direction of change of each effective channel within the same weld window relative to the most recent stable window; each channel is marked as rising, falling, deviating to the left, deviating to the right, expanding, contracting, or remaining basically unchanged, and then the number of channels pointing to the same abnormal direction is counted; When the main visual channel and at least one auxiliary channel point to the same abnormal direction, the trend consistency is judged to be high; when only one channel points to the abnormal direction and the remaining effective channels remain basically unchanged, the trend consistency is judged to be low; when two or more effective channels point to opposite abnormal directions, the trend consistency is judged to be conflicting. For each detection quantity, if it has one of the following flags: saturation, occlusion, hole, frame loss, or communication anomaly, it is marked as invalid; if it does not have the above flags and the time and space deviations are within the allowable range, and the trend consistency is high, it is marked as valid. If only one deviation slightly exceeds the allowable range, does not have the aforementioned abnormality marker, and at least one other valid channel points to the same abnormal direction, then it is marked as valid for deweighting. If the deviation is greater than the preset second deviation threshold, there is a trend consistency conflict, or there are no other valid channels that can be verified and the detection quantity belongs to a vulnerable channel, it is marked as invalid. The detection quantities are traversed in the following order: main vision detection quantity, auxiliary vision detection quantity, temperature detection quantity, photoelectric detection quantity, acoustic emission detection quantity, and motion detection quantity; if one type of detection quantity is invalid, the remaining detection quantities are evaluated. Event packet quality is determined sequentially by communication status, location mapping status, and the completeness of the original measurement; when there are frame loss markers or communication anomaly markers, the event packet quality is low. The event packet quality is medium when one of the aforementioned markers is present, such as insufficient location interpolation, continuous delay substitution, or inconsistent location source markers; and the event packet quality is high when the communication status is normal, the location mapping is normal, and the main measurements are complete. The updated confidence level uses a three-level discrete value: 0.9 for high confidence level, 0.6 for medium confidence level, and 0.3 for low confidence level; the number of detections that are completely effective and have high trend consistency is increased by one level from the initial confidence level. The number of valid detections to be downgraded should be kept at the original level or downgraded by one level; the update confidence of invalid detections should be set to 0; the adjustment should be limited to the adjacent levels of high, medium and low, not exceeding 0.9 and not lower than 0; when the event packet quality is low, the update confidence should not be adjusted to high. The fusion result is generated only from the detection quantities marked as valid or to be downgraded to valid; to enable detection quantities with different dimensions to participate in the judgment, each detection quantity is first converted into a dimensionless deviation level relative to its corresponding stability window: If the deviation falls within the allowable range of the process, it is recorded as 0; if it exceeds the allowable range but does not reach the preset deviation threshold, it is recorded as 1; if it reaches the preset deviation threshold or two consecutive weld windows deviate in the same direction, it is recorded as 2. The level is multiplied by the update confidence level and used as the fusion contribution value of the detection quantity. When there are multiple effective channels for the same detection feature, the fusion value of the feature is obtained by dividing the sum of the fusion contribution values ​​of each channel by the sum of the corresponding update confidence levels. When the denominator is 0, the fusion value of the feature is not generated and it is marked as insufficient observation conditions. Therefore, the molten pool boundary, weld position, temperature, photoelectric, acoustic emission and motion characteristics each form independent fusion values, and different physical quantities are not directly added together; In the formation of risk scores, the data fusion module extracts molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters and motion characteristic parameters from the fusion results, and judges whether each characteristic item falls within its corresponding process allowable range. Items falling within the allowable range are scored 0 points; items exceeding the allowable range with a difference less than or equal to the first deviation threshold are scored 1 point; items with two consecutive weld windows deviating in the same direction or a single window exceeding the allowable range with a difference greater than the first deviation threshold are scored 2 points. The current risk score for the weld window is the sum of the scores for the above six items, with a maximum risk score of 12. The allowable range and significant deviation limit for each item are determined based on stable welding records from the same production line, the same weld type, and the same trajectory stage. First, extract the original physical measurement values ​​of each effective sensor channel in the normal historical records, and statistically analyze the continuous and stable distribution range under the true physical dimensions. Then, take the upper or lower boundary of the physical range as the process allowable range boundary of the feature item, and take the physical threshold corresponding to the historical defect sample as the preset deviation threshold. When changing the tooling, solder joint type, material combination or sensor installation position, redetermine the corresponding range. Each trajectory stage saves a corresponding preset risk threshold, which is then called in the working condition setting module based on the workpiece identification, weld point type, and trajectory stage. When the risk score reaches or exceeds the preset risk threshold for the current trajectory stage, the corresponding weld window is marked as having risk; if a single weld position parameter or molten pool boundary parameter reaches 2 points, a local high-risk candidate mark is written at the same time. If the mark remains within a continuous weld window, or if the geometric matching spatial deviation continuously exceeds a preset spatial deviation threshold, then the window is marked as a local high-risk area. If the current effective detection volume is insufficient to simultaneously cover a primary detection volume and an auxiliary detection volume, only the partial fusion result formed by the existing effective detection volume will be output, the confidence level will be downgraded, and an insufficient observation condition marker will be added. A definitive risk conclusion will not be forcibly formed based on this.

[0021] In this embodiment of the invention, the control generation module constructs and runs a state machine consisting of stable welding, suspected anomaly, enhanced observation, parameter suppression, shutdown confirmation, and anomaly recovery; and determines the next state based on the current state, risk score, anomaly duration, and system occupancy level. When the risk score is greater than or equal to the preset risk threshold and the duration of the anomaly reaches or exceeds the preset duration threshold, if the system occupancy level is lower than the preset occupancy threshold, the system will switch to enhanced observation; otherwise, it will switch to suspected anomaly. When the anomaly persists and both the main vision sensor and at least one auxiliary sensor determine that there is an anomaly in the same weld window, the process is switched to a shutdown confirmation. When the risk score is lower than the preset recovery threshold and continues for the preset recovery time, and the preset sensor confidence level is higher than the preset confidence level threshold, the system will switch to abnormal recovery. The operating condition setting module calculates and determines the system occupancy level, and the control generation module calls the system occupancy level; When the system occupancy level is lower than the preset occupancy threshold, the multi-source acquisition module collects and sends data according to the current observation strategy. When the system occupancy level reaches or exceeds the preset occupancy threshold, the sampling rate under stable welding conditions is reduced, and the buffering priority and transmission priority of event packets corresponding to risk weld windows are increased; under suspected abnormal or enhanced observation conditions, the sensor priority and sampling rate corresponding to the current abnormality type are increased. The process of generating control commands by the control generation module is as follows: acquiring preset process window parameters including power target value, weld offset tolerance zone and temperature target range; taking the molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters or motion characteristic parameters in the fusion result as target control quantities, and calculating the deviation of the target control quantities relative to the corresponding process window parameters; Each target control variable is multiplied by its corresponding process mapping coefficient to obtain the laser power correction, galvanometer offset correction, or robot trajectory correction; the corresponding correction is then limited according to the equipment's allowable lower and upper adjustment limits. After writing the correction amount after the amplitude limit into the instruction template, the control instruction is generated; among them, each process mapping coefficient and process boundary are determined according to the preset calibration parameters; The control execution recording module sends control commands to the laser, galvanometer, robot, and programmable logic controller, receives corresponding execution confirmation signals or position feedback status, and records the source timestamp, corrected timestamp, validity mark, confidence level, fusion result, risk score, state transition, control command, and execution result according to the workpiece identification and weld point identification. The operating condition setting module updates the shutdown confirmation sensitivity parameter based on the execution failure rate, and updates the observation strategy, sensor priority, sampling rate, cache budget, and abnormal upgrade conditions for the next welding task based on the execution results, feedback status, and welding quality results.

[0022] When the laser welding station of the energy storage production line continuously processes multiple weld points, risk scoring alone is not enough to directly drive the equipment to operate; this is because the risk level, duration, and node load of different weld windows vary. The control generation module runs a state machine consisting of stable welding, suspected anomaly, enhanced observation, parameter suppression, shutdown confirmation and anomaly recovery, and then decides whether to adjust the sampling and equipment parameters. The control generation module reads the risk score and confidence level from the data fusion module, reads the system occupancy level from the working condition setting module, and judges the status change of the current weld window in combination with the duration of the anomaly. When the risk score reaches or exceeds the preset risk threshold for the corresponding trajectory stage, and the duration of the anomaly reaches or exceeds the preset duration threshold, the system further determines the current system occupancy level; if the system occupancy level is lower than the preset occupancy threshold, the state machine switches to enhanced observation. If the system occupancy level has reached or exceeded the preset occupancy threshold, the state machine will enter a suspected abnormal state and maintain the current sampling frequency. If the anomaly persists in subsequent weld seam windows, and the main vision sensor and at least one auxiliary sensor both give an anomaly judgment for the same weld seam window, the state machine can further transition to a shutdown confirmation; otherwise, when the risk score is lower than the preset recovery threshold and maintains the preset recovery time, and the preset sensor confidence is higher than the preset confidence threshold, the state machine transitions to anomaly recovery and then gradually returns to a stable welding state. The state machine executes judgments in a fixed sequence, rather than jumping arbitrarily at the same time. Specifically, after each weld window is reached, it first checks whether the shutdown confirmation condition is met; if not, it checks whether the enhanced observation or suspected anomaly condition is met; it checks whether the parameter suppression condition is met; it checks whether the anomaly recovery condition is met; if none of these conditions are met, the original state is maintained. Correspondingly, stable welding can only be transferred to suspected anomaly or enhanced observation. Suspected anomaly can be transferred to enhanced observation, parameter suppression, shutdown confirmation or anomaly recovery. Enhanced observation can be transferred to parameter suppression, shutdown confirmation or anomaly recovery. Parameter suppression can be transferred to enhanced observation, shutdown confirmation or anomaly recovery. Anomaly recovery can be transferred back to stable welding, transferred to suspected anomaly again or transferred to shutdown confirmation when the shutdown confirmation conditions are met. The judgment of shutdown confirmation conditions takes precedence over other state switching conditions; when the current state is suspected anomaly, enhanced observation, parameter suppression or anomaly recovery, as long as the anomaly persistence condition and the consistent anomaly condition of the main vision sensor and at least one auxiliary sensor are met at the same time, the shutdown confirmation will be initiated according to this priority order. The state machine executes the determined state transitions and switching controls based on the above decision sequence; In this embodiment, the parameter suppression state is used to bridge situations where risks have been identified but it is not advisable to further increase the adjustment amount at this time; Specifically, when the risk score has risen continuously, but the current number of effective detections is insufficient, the reliability of key sensors has decreased, the target control quantity is close to the upper or lower limit of equipment adjustment, or the feedback of the previous cycle shows that the following error exceeds the set threshold, the control generation module does not directly increase the correction range, but instead switches to parameter suppression. When in this state, the system maintains enhanced observation or sampling scheduling under suspected anomalies, while applying stricter limits to newly generated laser power correction, galvanometer offset correction, or robot trajectory correction. If necessary, the control template of the previous effective cycle remains unchanged. After the subsequent weld window obtains a more stable fusion result, it switches back to enhanced observation, shutdown confirmation, or anomaly recovery. The following error exceeds the set threshold when the execution confirmation signal has been returned, but the position feedback status and the target value in the instruction template fail to converge to the allowable band for two consecutive weld windows. Approaching the upper or lower limit of equipment adjustment means that the correction amount after the previous window limit has reached 90% or more of the boundary of the control item; If the shutdown confirmation conditions are not met, but any of the above conditions are met, parameter suppression takes precedence over further amplification adjustment in the enhanced observation state; if the shutdown confirmation conditions are met simultaneously, shutdown confirmation will still be prioritized according to the aforementioned state judgment order. The operating condition setting module continuously calculates the system occupancy level and provides it for use by the control generation module and the multi-source acquisition module. When the system occupancy level is lower than the preset occupancy threshold, the multi-source acquisition module collects and transmits data normally according to the current observation strategy. When the system occupancy level reaches or exceeds the preset occupancy threshold, the system prioritizes reducing the sampling rate of stable welding sections, while increasing the buffering priority and sending priority of event packets corresponding to risky weld windows, so that limited processing and transmission resources are concentrated on weld positions that require more judgment. In cases of suspected anomalies or enhanced observation, the priority and sampling rate of the corresponding sensors are increased according to the current anomaly type. For example, channels that are more sensitive to weld position shifts and channels that are more sensitive to changes in heat input are specifically enhanced. The system occupancy level is divided into three levels: low occupancy, medium occupancy, and high occupancy, which are directly called by the control generation module. High occupancy corresponds to reaching or exceeding the preset occupancy threshold. The working condition setting module updates the occupancy level once before the start of each welding task and once after the end of each consecutive preset window. If at least two of the processing queue length, packet loss rate, and cache usage ratio exceed the warning range consecutively, then write usage is high; if only one of them enters the warning range and the critical event packets have not exceeded the allowed waiting time, then write usage is medium; if all three are within a stable range, then write usage is low. After receiving high occupancy results, the multi-source acquisition module only downsamples stable welding sections and does not reduce the sampling rate of weld windows marked as risky, thereby ensuring that critical event packets pass through the main processing link first. In terms of equipment instruction generation, the control generation module does not directly use the risk score itself as the control quantity, but reads the preset process window parameters, including the power target value, weld offset tolerance zone and temperature target range. The target control quantity in the fusion result is compared with the corresponding process window parameter to obtain the deviation, and then multiplied by the corresponding process mapping coefficient to form the laser power correction, galvanometer offset correction or robot trajectory correction. After the above correction amount is formed, it is necessary to limit the amplitude according to the lower and upper limits of the adjustment allowed by the equipment. The result after limiting is written into the instruction template, and finally a control instruction that can be issued is generated. The deviation is obtained by comparing the target control quantity with the center value of the corresponding process window or the center line of the allowable zone; when the target control quantity falls within the allowable zone, the corresponding deviation is recorded as zero, and no correction is generated for this control item. When the target control quantity exceeds the allowable range, the sign is retained according to the direction of the excess to distinguish between power increase, power decrease, left correction, or right correction; the process mapping coefficients are fixed under the same weld point type and do not change temporarily due to a single weld window; Correspondingly, the laser power correction, galvanometer offset correction, and robot trajectory correction are calculated independently and do not substitute for each other; if the required input detection quantity for a certain control item is insufficient, only that control item is frozen, without affecting the continued generation of other control items; To ensure that the deviation of the target control quantity relative to the corresponding process window parameter has a definite source, this embodiment imposes a fixed limitation on the correspondence between the control item and the input detection quantity: when the weld position parameter in the fusion result deviates relative to the center line of the weld offset tolerance zone, it is only used as the input of the galvanometer offset correction quantity or the robot trajectory correction quantity. When the temperature characteristic parameters in the fusion result deviate from the target temperature range, they are only used as input for laser power correction. When the molten pool boundary parameters and temperature characteristic parameters both indicate insufficient or excessive heat input, laser power correction is generated first. Weld position control items remain independent and are not directly replaced by molten pool boundary parameters to generate trajectory correction. Under the same weld point type, the position parameter of the weld seam is not randomly selected for which position actuator, but is fixedly determined by the preset control permission table of the weld point: when the current weld seam window is within the fine tracking range that the galvanometer can execute, only the galvanometer offset correction amount is generated; When the current weld window is within the robot trajectory following range, or when the current workstation is not configured with galvanometer position correction permission, only robot trajectory correction amount is generated; when both the galvanometer and the robot have position correction permission, the galvanometer offset correction amount only applies to the local offset of the spot in the next weld window, and the robot trajectory correction amount is only written to the subsequent trajectory points that have not yet been executed as feedforward compensation. Within the same weld window, two position correction commands with the same direction and the same purpose are not issued at the same time for the same position deviation. For the power target value, the corresponding process boundary is not generated temporarily, but is formed with the power target value as the center and the upper and lower boundaries according to the allowable fluctuation in the preset calibration parameters. Therefore, the direction and magnitude of the power-related deviation are determined by using the power target value as a comparison benchmark, while the final limit is still based on the upper and lower boundaries; If a certain target control quantity lacks the necessary fusion results to support it, for example, the weld position information is insufficient to support trajectory correction, while the temperature information can still support power correction, then the control generation module will only write the correction amount for control items that meet the conditions, and keep the original template value unchanged for control items that lack the basis. If all control items lack reliable data, the system may only perform state switching and sampling adjustments without outputting new device adjustment values. To avoid conflicts between state switching and control execution within the same weld window, the processing order for each weld window in this embodiment is fixed as follows: first, complete the fusion result and risk score writing for the window, then complete the state judgment and instruction template generation, and apply the generated result to the sampling scheduling and regular control instruction issuance of the next weld window. The current window only records the judgment result and does not write back to change the acquisition and fusion that has been completed in the window; only when the shutdown confirmation is entered and the programmable logic controller safety link allows immediate disconnection, will the programmable logic controller issue a shutdown or prohibition of continued light emission command for subsequent welding actions that have not yet been executed, without retroactively canceling the completed actions of the current window; To ensure that the control issuance process forms a deterministic physical closed loop, the control execution record module processes each type of control instruction in the following order: generating the instruction to be issued, sending it, waiting for execution confirmation, verifying feedback, and resending or freezing after timeout. Specifically, after a laser power command is sent, priority is given to waiting for confirmation from the laser; after a galvanometer offset command is sent, priority is given to waiting for feedback on the galvanometer's position. After a robot trajectory command is sent, priority is given to waiting for robot position feedback or trajectory reception confirmation; if confirmation is received within the allowable confirmation time corresponding to the current weld window and the feedback value enters the allowable zone, the control item is recorded as executed successfully. If an acknowledgment is received but the feedback value does not enter the allowed band, it is recorded as executed but not converged, and the next window will determine whether to switch to parameter suppression. If no confirmation is received within the timeout period, the control item will be resent only once. If no confirmation is received after the resentment, the control item will not be resent indefinitely, but the control item will be frozen and the execution result of this window will be recorded as a failure. When multiple control items exist simultaneously within the same weld window, the safety priority of the shutdown or prohibition of continued light emission command is higher than that of the conventional power correction, the power correction is higher than the position feedforward correction, and the position feedforward correction is higher than the sampling rate adjustment. If a high-priority control item has already triggered a security prohibition, then a low-priority control item will no longer be issued; through this cascading order, the execution priority in case of rule conflicts is fixed; After receiving the control commands issued by the control generation module, the control execution recording module sends them to the laser, galvanometer, robot or programmable logic controller via the industrial bus, and receives the corresponding execution confirmation signal or position feedback status. Before the instruction is issued, the control execution record module also checks the interlock conditions related to the current device action. The interlock conditions include at least the laser enable state, cooling and protective gas state, robot or galvanometer axis ready state, and programmable logic controller safety permission signal. When any interlock condition is not met, the current instruction is not executed; instead, the result is recorded as not executed due to interlock restrictions, and the corresponding state transition and instruction template content are retained. The control execution record module writes the source timestamp, corrected timestamp, validity mark, confidence level, fusion result, risk score, state transition, control instruction, and execution result into the record area according to the workpiece identifier and weld point identifier. This record can reflect the detection and control process that a weld point undergoes during the welding cycle, and can also serve as the basis for extracting update strategies for subsequent working condition settings modules. Before the start of subsequent welding tasks, the working condition setting module can update the shutdown confirmation sensitivity parameters based on the execution failure rate, and, in combination with the execution results, feedback status and welding quality results, revise the observation strategy, sensor priority, sampling rate, buffer budget and abnormal upgrade conditions for the next welding task. If a certain type of control command repeatedly fails to execute in the preceding task, the operating condition setting module will increase the shutdown confirmation sensitivity parameter before the next task begins, so that the shutdown confirmation condition is moved forward accordingly; if a certain type of solder joint can still maintain stable quality at a low sampling rate, the next task can continue to use the lower load configuration. The execution record of the previous task does not directly change the instructions generated by the current task, but is used as the basis for updating the working conditions of the next task, so that the operating parameters of the entire welding line are gradually corrected according to the workpiece and weld point. In this embodiment, the preset duration threshold, preset recovery time, preset occupancy threshold, preset recovery threshold, preset confidence threshold, and shutdown confirmation sensitivity parameter are all determined according to the solder joint type, trajectory stage, and production line cycle time. Specifically, the preset duration threshold is the shortest number of windows that triggers a state transition only when multiple consecutive weld windows maintain an abnormal trend, in order to filter out single-window fluctuations; the preset recovery time is the shortest number of windows where the risk score continues to fall and remains stable, in order to avoid repeated switching immediately after recovery. The preset occupancy threshold is determined by the combined operation records of processing queue length, packet loss rate, and cache occupancy ratio, so that the system reduces the sampling rate of preset non-critical channels before the occupancy level reaches the preset congestion threshold; the preset recovery threshold and preset confidence threshold are jointly determined by the risk score and confidence distribution in the stable welding records; The sensitivity parameter for shutdown confirmation is adjusted up or down according to the failure rate and subsequent welding quality results. The above parameters are kept consistent under the same weld type and are only recalibrated when the equipment cycle time, sensor configuration or process window is adjusted. To enable the state switching conditions to be executed directly, the preset duration threshold and preset recovery time are recorded in terms of the number of weld windows under the same weld point type and trajectory stage; the abnormal duration reaching the preset duration threshold means that the risk scores of the current and previous consecutive weld windows have reached or exceeded the corresponding preset risk threshold. The continuous preset recovery time refers to the situation where the risk scores of several consecutive weld windows are lower than the preset recovery threshold and the update confidence of the main vision sensor and at least one auxiliary sensor are higher than the preset confidence threshold. For calculating system occupancy levels, this implementation method uses a tiered approach rather than relying on a single indicator. Specifically, when the processing queue length, packet loss rate, and cache occupancy ratio are all within their respective stable operating ranges, the system is considered to be under low occupancy. When any one of the indicators enters the warning range but the critical event package can still be processed in a timely manner according to the schedule, it is judged as medium occupancy; when at least two indicators exceed the warning range consecutively, or when the critical event package has been queued for longer than the allowed waiting time, it is judged as high occupancy. The control generation module distinguishes between at least medium occupancy and high occupancy, with high occupancy corresponding to reaching or exceeding a preset occupancy threshold in the embodiment. This clarifies when to maintain only the current observation strategy and when it is necessary to reduce the sampling rate of stable welding segments and increase the priority of risk window event packets. When writing control quantities into the instruction template, this implementation method classifies and marks correction quantities according to the duration of deviation, but the classification does not constitute a second calculation path independent of the process mapping coefficient calculation. For each weld window, the target control quantity is first compared with the corresponding process window parameters to obtain the signed deviation quantity. The deviation quantity is multiplied by the same process mapping coefficient to obtain the basic correction quantity. The current correction quantity is obtained by quantifying according to the minimum executable adjustment step size of the equipment and adjusting the upper and lower limits of the control item in the order of limiting the amplitude. When the target control value falls within the allowable zone, the corresponding correction value is written as zero; when the deviation exceeds the allowable zone for the first time in a single weld window, the current correction value is marked as a first-level correction value and written into the instruction template. When the deviation remains in the same direction for at least two consecutive weld windows, the correction amount after quantization and amplitude limiting will be recalculated based on the current window deviation and marked as the secondary correction amount. The absolute value of the secondary correction amount shall not exceed the upper limit of the preset multiple of the absolute value of the most recent valid primary correction amount, nor shall it exceed the process boundary and equipment boundary of the control item; if it exceeds, the smaller absolute value among the above boundaries shall be taken. If the current solder joint does not have the most recent valid first-level correction value before entering the continuous unidirectional superband, the first superband window will only write the first-level correction value and use this window as the starting point for subsequent continuous unidirectional superband comparisons; only when the subsequent continuous windows still maintain unidirectional superband will a second-level correction value be generated and the above-mentioned first-level correction value be used as the comparison benchmark for the preset multiple upper limit. When the deviation is close to the process boundary but has not yet triggered a shutdown confirmation, it enters the parameter suppression state and keeps the current control item at the effective limit value already executed in the previous window, and no longer increases it; Therefore, Level 1 and Level 2 are only used to represent the deviation persistence state and additional limiting conditions. The correction amount itself still comes solely from the deviation amount, the corresponding process mapping coefficient, the minimum executable adjustment step size of the equipment, and the sub-item boundary. If the system enters an abnormal recovery state, the control items are not immediately cleared to zero. Instead, while maintaining the original correction direction, the absolute value of the correction amount is reduced window by window according to the minimum executable adjustment step size of the equipment until it returns to zero correction. If the remaining correction amount is less than a minimum executable adjustment step size, the next window is directly written as zero. The above step-by-step reduction method is used to avoid reverse over-adjustment during the abnormal recovery process. To ensure that the corresponding process mapping coefficients do not constitute vague limitations, under the same weld joint type, the laser power correction only calls the power mapping coefficient, the galvanometer offset correction only calls the galvanometer offset mapping coefficient, and the robot trajectory correction only calls the trajectory mapping coefficient. The three are stored in preset calibration parameters and read in a fixed manner according to the control item name. The same control item will not be temporarily switched to other mapping coefficients within the same weld point type and the same trajectory stage; when the control generation module generates the instruction template, it processes the control item name in a fixed order: read the corresponding deviation amount, read the same mapping coefficient, quantize according to the minimum executable adjustment step size of the equipment, execute the limit, and write back the template field, so that the generation path of each control item can be reproduced item by item. The limiting of different control items adopts sub-item boundaries; the laser power correction amount shall not cause the power to exceed the upper and lower boundaries of the process window corresponding to the power target value; the galvanometer offset correction amount shall not cause the spot center to exceed the equipment's allowed scanning boundary outside the weld offset allowable zone; The robot trajectory correction amount must not cause the trajectory points to exceed the robot's permissible speed, acceleration, and position boundaries; when a control item reaches its sub-item boundary, the control item is frozen at the current boundary value and will not continue to increase even if other control items still have margin. This sub-item limiting mechanism is used to prevent a single overall score from driving all implementing agencies to amplify synchronously when the overall risk is high, so as to prevent local control from overstepping its limits. For shutdown confirmation, this implementation also sets an anomaly rejection rule; if the main vision sensor determines an anomaly, but the corresponding auxiliary sensor has communication anomalies, frame drops, or confidence levels below the preset confidence threshold in a continuous window, then it does not directly enter the shutdown confirmation process, but maintains the suspected anomaly or enhances observation and increases the priority of the relevant channels. If the main vision sensor and at least one auxiliary sensor are both effective and consistently determine an anomaly within a continuous window, then proceed to the shutdown confirmation according to the aforementioned fixed judgment sequence. When the corresponding control item has reached the restricted boundary under the parameter suppression state, the sensitivity of shutdown confirmation can be increased, so that the number of consecutive windows required to reach the abnormality persistence condition is reduced according to the pre-calibrated sensitivity level. However, the restricted boundary is not a necessary condition for entering shutdown confirmation. The above multi-source joint confirmation rule is used to meet the shutdown logic's requirements for multi-source evidence and to prevent shutdown from being triggered by occasional abnormalities in a single channel.

Claims

1. A visual positioning and detection data fusion system for laser welding in an energy storage production line, characterized in that, include: The operating condition setting module is used to determine the observation strategy, sensor priority, sampling rate, cache budget, and abnormal escalation conditions based on the operating condition information and the system occupancy level represented by the processing queue length, packet loss rate, and cache usage ratio. The multi-source acquisition module communicates with vision sensors, temperature sensors, acoustic sensors, photoelectric sensors, encoders, and robot controllers via communication interfaces. It is used to acquire coaxial images, side view images, 3D contours, temperature, photoelectric data, acoustic emission data, encoder position, and robot pose data as detection quantities and encapsulates them into event packets. A synchronization mapping module, connected to the multi-source acquisition module, is used to correct the time of the event packet and map the event packet to the weld arc length coordinates; The data fusion module, connected to the synchronization mapping module, is used to determine the validity of the detection quantity within the event packet, update the confidence field, and generate fusion results and risk scores. The control generation module, connected to the data fusion module, is used to generate state switching parameters, sensor scheduling parameters and instruction templates based on the risk score, the update confidence field, the anomaly duration and the system occupancy level. The control execution recording module, connected to the control generation module, is used to generate control instructions restricted by process boundaries and interlocking conditions according to the instruction template, send them to the laser, galvanometer, robot and programmable logic controller via industrial bus, and record the execution results.

2. The visual positioning and detection data fusion system for laser welding of energy storage production lines according to claim 1, characterized in that, The event package includes workpiece identifier, weld point identifier, sensor identifier, trigger sequence number, source timestamp, corrected timestamp, encoder position and robot pose data, validity flag, initial confidence level and updated confidence level fields; The multi-source acquisition module writes the source timestamp, trigger sequence number, encoder position and robot pose data, validity flag and initial confidence level. The validity flag includes saturation flag, occlusion flag, hole flag, frame loss flag and communication anomaly flag. The synchronization mapping module writes the corrected timestamp.

3. The visual positioning and detection data fusion system for laser welding of energy storage production lines according to claim 2, characterized in that, The process by which the synchronization mapping module corrects the time of the event packet is as follows: obtain the source timestamp of the sensor channel corresponding to each sensor in the multi-source acquisition module; The sampling delay, buffer delay, processing delay, and transmission delay corresponding to the sensor channel are determined respectively; the corrected timestamp is obtained by subtracting the sampling delay, buffer delay, processing delay, and transmission delay from the source timestamp. Based on the hardware trigger pulse, the sampling delay, the buffer delay, the processing delay, and the transmission delay are updated by sliding estimation.

4. The visual positioning and detection data fusion system for laser welding of energy storage production lines according to claim 3, characterized in that, The process by which the synchronization mapping module maps each event packet to weld arc length coordinates is as follows: The encoder position and robot pose data are interpolated based on the corrected timestamp; the interpolation result is projected onto the weld reference trajectory, and the weld arc length coordinates corresponding to the event packet are calculated based on the initial calibration parameters for spatial calibration and the online drift correction parameters for drift compensation. Geometric matching is performed on the tooling reference edge region in the coaxial image or side view image. When the obtained translation deviation, equivalent edge displacement and projection residual deviation are all less than or equal to the spatial allowable range, the online drift correction parameters are updated according to the obtained geometric matching results. When the translation deviation, equivalent edge displacement, or projection residual deviation exceeds the allowable spatial range, the online drift correction parameters are stopped from being updated.

5. The visual positioning and detection data fusion system for laser welding of an energy storage production line according to claim 4, characterized in that, The data fusion module processes the following steps: acquiring the time deviation, spatial deviation, and multi-sensor trend consistency parameters of the event packets; The time deviation is calculated based on the difference between the corrected timestamp and the recorded time at adjacent locations. The spatial deviation is calculated based on the remaining projection deviation and the equivalent edge displacement. The multi-sensor trend consistency parameters are statistically analyzed based on the changing direction of multiple sensor channels within the same weld window. The validity of the detection quantity is determined based on the time deviation, the spatial deviation, and the multi-sensor trend consistency parameter, and each detection quantity in the event packet is marked as valid or invalid. The update confidence field is updated based on the event packet quality and the multi-sensor trend consistency parameter, and the fusion result is generated based on the effective detection volume and confidence.

6. The visual positioning and detection data fusion system for laser welding of energy storage production lines according to claim 5, characterized in that, The risk score is calculated by extracting the corresponding molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters, and motion characteristic parameters from the fusion result. The risk score is calculated based on the extracted feature parameters; the risk score is compared with a preset risk threshold, and when the risk score is greater than or equal to the preset risk threshold, it is determined that there is a risk in the corresponding weld window divided along the weld arc length coordinate. When the risk score is less than the preset risk threshold, the corresponding weld window divided along the weld arc length coordinate is determined to be risk-free; wherein, different trajectory stages correspond to different preset risk thresholds.

7. The visual positioning and detection data fusion system for laser welding of an energy storage production line according to claim 6, characterized in that, The control generation module constructs and runs a state machine consisting of stable welding, suspected anomaly, enhanced observation, parameter suppression, shutdown confirmation, and anomaly recovery. The next state is determined based on the current state, risk score, duration of the anomaly, and system occupancy level. When the risk score is greater than or equal to the preset risk threshold, and the abnormal duration reaches or exceeds the preset duration threshold, if the system occupancy level is lower than the preset occupancy threshold, the system will switch to enhanced observation; otherwise, it will switch to suspected abnormality. When the anomaly persists and both the main vision sensor and at least one auxiliary sensor determine that there is an anomaly in the same weld window, the process is switched to a shutdown confirmation. When the risk score is lower than the preset recovery threshold and continues for the preset recovery time, and the preset sensor confidence level is higher than the preset confidence level threshold, the process switches to abnormal recovery.

8. The visual positioning and detection data fusion system for laser welding of an energy storage production line according to claim 7, characterized in that, The system occupancy level is calculated and determined by the operating condition setting module, and the control generation module calls the system occupancy level. When the system occupancy level is lower than the preset occupancy threshold, the multi-source acquisition module collects and sends data according to the current observation strategy. When the system occupancy level reaches or exceeds the preset occupancy threshold, the sampling rate under stable welding is reduced, and the caching priority and sending priority of the event packets corresponding to the risk weld window are increased. In cases of suspected anomalies or enhanced observation, increase the priority and sampling rate of sensors corresponding to the current anomaly type.

9. The visual positioning and detection data fusion system for laser welding of an energy storage production line according to claim 8, characterized in that, The process of the control generation module generating control commands is as follows: acquiring preset process window parameters including power target value, weld offset tolerance zone and temperature target range; The molten pool boundary parameters, weld position parameters, temperature characteristic parameters, photoelectric characteristic parameters, acoustic emission characteristic parameters, or motion characteristic parameters in the fusion result are used as target control quantities, and the deviation of the target control quantities relative to the corresponding process window parameters is calculated. Each target control variable is multiplied by its corresponding process mapping coefficient to obtain the laser power correction, galvanometer offset correction, or robot trajectory correction; the corresponding correction is then limited according to the equipment's allowable lower and upper adjustment limits. The control command is generated after the amplitude limit correction amount is written into the command template; wherein, each process mapping coefficient and the process boundary are determined according to preset calibration parameters.

10. The visual positioning and detection data fusion system for laser welding of an energy storage production line according to claim 9, characterized in that, The control execution recording module sends the control commands to the laser, galvanometer, robot, and programmable logic controller, receives the corresponding execution confirmation signal or position feedback status, and records the source timestamp, corrected timestamp, validity mark, confidence level, fusion result, risk score, state switch, control command, and execution result according to the workpiece identifier and weld point identifier. The operating condition setting module updates the shutdown confirmation sensitivity parameter based on the execution failure rate, and updates the observation strategy, sensor priority, sampling rate, cache budget, and abnormal upgrade conditions for the next welding task based on the execution result, feedback status, and welding quality result.

Images