Automobile assembly line communication coordination system and method based on industrial network

Abstract

The invention discloses an automobile assembly line communication coordination system and method based on an industrial network, and relates to the technical field of communication coordination, and the method comprises the steps: calculating a single welding quality score Q in real time for an edge calculation node located at each welding station, generating a welding event fingerprint, and carrying out annular caching; determining whether to upload the high-resolution data and the original waveform to a rear end based on a threshold value and the calculated joint anomaly score; qoS management is carried out on the control flow and the quality return flow; welding event fingerprints reported by the edge are received, backtracking data are stored, a network-quality joint causal analysis model is executed, and model updating is issued; and according to a causal analysis result and a historical model, implementing automatic welding parameter fine tuning, repair welding triggering, local redundancy switching and rollback strategies. According to the method, the problems that the welding process is extremely sensitive to delay jitter and intermittent welding defects and traceability data missing are caused by high-resolution quality data backtracking in an existing automobile assembly production line industrial network are solved.

Description

Technical Field

[0001] This invention relates to the field of communication coordination technology, specifically to a communication coordination system and method for an automobile assembly line based on an industrial network. Background Technology

[0002] As automobile manufacturing transforms towards high automation, flexibility, and digitalization, the requirements for production cycle time, positional accuracy, and quality traceability in assembly lines (including body shops, painting lines, chassis lines, and final assembly lines) have significantly increased. Modern assembly lines consist of a large number of robots, welding units, vision inspection systems, torque guns, AGVs / rail logistics, and edge computing nodes. These devices collaborate through industrial communication networks to complete high-frequency, strictly time-sequential actions. To meet the demands of real-time control and data aggregation, plant networks are evolving from traditional fieldbus to industrial Ethernet and Time-Sensitive Networking (TSN), while introducing new technologies such as OPC UA (including Pub / Sub), edge computing, private 5G, and cloud-edge collaboration, aiming to achieve scalable data services and quality traceability while ensuring real-time performance.

[0003] However, automotive assembly lines still face the following challenges: First, critical processes such as welding are extremely sensitive to timing determinism—the latency jitter in spot welding / arc welding / laser welding for synchronizing robot trajectories and movements is highly unacceptable; even intermittent network jitter can cause weld point misalignment or missed welds, leading to rework or scrapping of the entire vehicle. Second, the demand for quality backtracking is large and granular; each vehicle needs to be associated with a large number of high-resolution visual images, waveform curves, and workstation logs, which poses severe challenges to bandwidth, edge storage, and long-term archiving. Improper handling can affect real-time control or cause backtracking data loss. Furthermore, the coexistence of multiple vendors' equipment in the production environment, the mixing of legacy fieldbus and Ethernet, wireless mobile node switching, and firmware security all introduce additional uncertainties to actual deployment.

[0004] Therefore, the present invention provides a communication coordination system and method for automobile assembly lines based on industrial networks. Summary of the Invention

[0005] The purpose of this invention is to provide a communication coordination system and method for automobile assembly lines based on industrial networks, so as to solve the existing problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a communication and coordination system for an automotive assembly line based on an industrial network, comprising: The welding sensing module includes: at least one set of welding electrical parameter acquisition devices, at least one set of vibration sensors, at least one low frame rate vision camera, and at least one high resolution ring buffer camera. The raw waveform upload decision module is used to calculate the single welding quality score Q in real time for the edge computing nodes located at each welding station, generate welding event fingerprints, and implement circular caching; based on the threshold and the calculated joint anomaly score, it decides whether to upload high-resolution data and raw waveforms to the backend. The Quality Backhaul Management module is used to perform time-sensitive network isolation for critical control flows, implement PTP hardware time synchronization and network metric collection, and perform QoS management for control flows and quality backhaul flows. The causal determination model update module is used to receive welding event fingerprints reported from the edge, store backtracking data, execute the network-quality joint causal analysis model, and issue model updates. The adaptive compensation module is used to implement automatic welding parameter fine-tuning, repair welding triggering, local redundancy switching and rollback strategies based on the single welding quality score Q obtained by the original waveform upload decision module, the causal analysis results of the causal judgment model update module and the historical model.

[0007] A further improvement of this invention lies in that the single welding quality score Q in the original waveform upload decision module is constructed by collecting electrical waveform signals, vibration signals, labeled images, and manual quality judgments from N car samples to build a training set; and the original current... With voltage Short-time feature extraction is employed to calculate peak value, energy, duration, and root mean square (RMS), and each feature is normalized to obtain the normalized current feature. and voltage characteristics ; Bandpass filtering is applied to the vibration signal, the principal components of the energy spectrum are calculated and normalized to obtain the normalized vibration characteristics. ; A lightweight CNN forward inference is performed on the labeled image signal to obtain the image score 𝑆; Current characteristics and voltage characteristics Vibration characteristics Image score The single welding quality score Q is obtained by linear weighting and then Sigmoid mapping.

[0008] A further improvement of this invention is that the process of generating the welding event fingerprint includes: after each welding event is completed at the edge, the electrical waveform signal, vibration signal and labeled image are extracted according to the time window; the energy spectrum is extracted by Fourier transform and the first k principal component vectors are extracted by PCA; the first k principal component vectors are used to generate a summary by LSH, and the samples are combined into a welding event fingerprint F and reported.

[0009] A further improvement of this invention is that the quality backhaul management module adopts a PTP boundary clock and a transparent clock that support hardware timestamps, and edge nodes perform statistical network metrics on key flows, including one-way delay sequences. Shaking Packet loss rate PTP offset original amount and the offset after exponential smoothing .

[0010] A further improvement of this invention is that the joint anomaly score is calculated based on an aligned PTP timeline, and the calculation formula is expressed as follows: ;in, , , These represent the jitter standard value, packet loss standard value, and PTP offset standard value, respectively. Simultaneously, time-series causal analysis is used to confirm whether changes in network metrics pre-influence a decrease in the single welding quality score. If determined to be yes, anomaly scoring is also considered. If the score exceeds the set anomaly score threshold, it is determined to be a network-related fault; otherwise, it is determined to be a process-related fault.

[0011] A further improvement of this invention is that the circular caching strategy calculates the single welding quality score Q in real time and generates a welding event fingerprint F, while storing high-resolution frames and the original waveform in a short time window locally. The feedback is determined based on semantic gating logic. When the single welding quality score Q is less than the set quality score threshold or the combined abnormal score is less than the threshold, the feedback is determined. When the score exceeds the set anomaly threshold, high-resolution data for the corresponding time period is uploaded from the circular cache to the central storage; otherwise, only the welding event fingerprint F is uploaded.

[0012] A further improvement of this invention is that when the original waveform upload decision module determines that the fault is process-related, a process attribution process is executed, including defining an adjustment parameter set, including current I, dwell time dT, and welding speed v, setting a minimum / maximum adjustment range and step size for each parameter, and performing micro-step parameter adjustments at the edges according to the following formula and evaluating the effect: ,in, Indicates the standard welding current. The coefficient is an empirical factor, and the single welding quality score Q curve is monitored for N welding events after adjustment. If the average welding quality score Q improves, it is retained and recorded; if there is no improvement or it deteriorates, it is rolled back and an alarm is triggered.

[0013] A further improvement of this invention is that when the original waveform upload decision module determines that the fault is process-related, it executes a network attribution process, including switching the key control from remote control mode to local control mode of the local PLC; rolling back the most recent network configuration change; and prioritizing the reporting of the labeled image signal and waveform data of the corresponding welding position for troubleshooting.

[0014] A further improvement of this invention is that the network-quality joint causal analysis model in the causal determination model update module is used to use all labeled events as training samples, periodically train and update the welding quality scoring model and joint anomaly discrimination weights, and after acceptance, it is distributed to the edge nodes in a versioned manner. After receiving the new model, the edge nodes perform gray-scale deployment and report the effect indicators to determine whether to replace the entire model.

[0015] On the other hand, the present invention provides a communication coordination method for an automotive assembly line based on an industrial network, comprising the following steps: S1. For each edge computing node located at the welding station, calculate the single welding quality score Q in real time, generate welding event fingerprints, and implement circular caching; based on the threshold and the calculated joint anomaly score, decide whether to upload high-resolution data and original waveforms to the backend. S2. Perform time-sensitive network isolation on critical control flows, implement PTP hardware time synchronization and network indicator collection, and perform QoS management on control flows and quality return flows. S3, Receive welding event fingerprints reported from the edge, store backtracking data, execute the network-quality joint causal analysis model and issue model updates; S4. Based on the single welding quality score Q obtained by the original waveform upload decision module, the causal analysis results of the causal judgment model update module and the historical model, implement automatic welding parameter fine-tuning, repair welding triggering, local redundancy switching and rollback strategies.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first places the control commands for spot welding, arc welding, and laser welding machines into independent time-sensitive network channels, and uses hardware timestamps to accurately calibrate end-to-end delays and jitter, thereby ensuring the determinism of robot trajectory control; it can not only minimize the impact of network jitter on weld point positioning, but also significantly reduce intermittent weld point defects and weld discontinuities, ensuring the accuracy and consistency of vehicle body assembly. 2. By performing real-time preprocessing and compression of high-resolution visual inspection data at the edge nodes of the assembly station, and establishing a hierarchical storage and key time window caching strategy, anomaly and summary information are uploaded first, rather than all raw data. This not only alleviates network bandwidth and storage pressure, but also ensures the integrity and consistency of quality traceability data in high-frequency acquisition scenarios, improving the reliability of subsequent quality analysis and accountability. Attached Figure Description

[0017] Figure 1 This is a framework diagram of the communication and coordination system for an automobile assembly line based on an industrial network, as described in this invention. Figure 2 This is a flowchart of the communication coordination method for an automobile assembly line based on an industrial network, as described in this invention. Detailed Implementation

[0018] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0019] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] Example 1 Figure 1 This embodiment illustrates a framework diagram of a communication and coordination system for an automotive assembly line based on an industrial network, including: The welding sensing module includes: at least one set of welding electrical parameter acquisition devices (for acquiring current and voltage waveforms), at least one set of vibration sensors, at least one low frame rate vision camera, and at least one high-resolution ring buffer camera. The raw waveform upload decision module is used to calculate the single welding quality score Q in real time for the edge computing nodes located at each welding station, generate welding event fingerprints, and implement circular caching; based on the threshold and the calculated joint anomaly score, it decides whether to upload high-resolution data and raw waveforms to the backend. The Quality Backhaul Management module is used to perform time-sensitive network isolation for critical control flows, implement PTP hardware time synchronization and network metric collection, and perform QoS management for control flows and quality backhaul flows. The causal determination model update module is used to receive welding event fingerprints reported from the edge, store backtracking data, execute the network-quality joint causal analysis model, and issue model updates. The adaptive compensation module is used to implement automatic welding parameter fine-tuning, repair welding triggering, local redundancy switching and rollback strategies based on the single welding quality score Q obtained by the original waveform upload decision module, the causal analysis results of the causal judgment model update module and the historical model.

[0021] It realizes a closed-loop process with edge detection, semantic gating backhaul, and network-quality joint causal analysis as the core, which can not only ensure the determinism and security of critical control, but also enable post-event backtracking and rapid root cause location under bandwidth-limited conditions, thereby reducing rework rate, optimizing bandwidth usage and shortening fault location time (MTTR).

[0022] The original waveform upload decision module uses a single welding quality score Q, which is constructed by collecting electrical waveform signals, vibration signals, labeled images, and manual quality judgments from N car samples to build a training set; the original current... With voltage Short-time feature extraction is employed to calculate peak value, energy, duration, and root mean square (RMS), and each feature is normalized to obtain the normalized current feature. and voltage characteristics ; Bandpass filtering is applied to the vibration signal, the principal components of the energy spectrum are calculated and normalized to obtain the normalized vibration characteristics. ; A lightweight CNN forward inference is performed on the labeled image signal to obtain the image score 𝑆; Current characteristics and voltage characteristics Vibration characteristics Image score The single welding quality score Q is obtained by linear weighting and then Sigmoid mapping.

[0023] Feature normalization is used because different physical quantities have large differences in dimensions (current mA, acoustic energy dB, and image scores are dimensionless), and normalization makes linear weighting numerically comparable. Using min-max or z-score re-linear mapping can balance robustness and edge computing efficiency.

[0024] The benefit of weight training is that it can quantify human experience rules into a data-driven discriminator, reducing false alarms / missed detections caused by the failure of a single sensor.

[0025] By fusing multimodal features to obtain a more robust online quality score, anomalies can be detected efficiently without uploading the entire large image, thereby saving bandwidth and ensuring backtracking capabilities.

[0026] The process of generating the welding event fingerprint includes: after each welding event is completed, the electrical waveform signal, vibration signal and labeled image are extracted according to the time window at the edge; the energy spectrum is extracted by Fourier transform and the first k principal component vectors are extracted by PCA; the first k principal component vectors are used to generate a summary by LSH, and combined into a welding event fingerprint F and reported.

[0027] The fingerprint structure enables the central terminal to efficiently retrieve, index, and compare historical similar events and quickly locate relevant network logs without uploading raw big data, thereby significantly reducing root cause analysis time and storage costs.

[0028] The quality backhaul management module uses a PTP boundary clock and a transparent clock that support hardware timestamps. Edge nodes perform statistical network metrics on key flows, including one-way delay sequences. Shaking Packet loss rate PTP offset original amount and the offset after exponential smoothing .

[0029] Using hardware timestamps ensures nanosecond / microsecond-level accuracy in measurements, making it easy to align network events with welding events by time.

[0030] Jitter, measured by standard deviation, reflects the dispersion of delay distribution. Welding control is extremely sensitive to jitter, and the threshold of J is an important indicator for judging the network's impact.

[0031] Exponential smoothing is used to filter out short-term noise and retain trend shifts, with α values ​​ranging from 0.8 to 0.99 to balance response speed and stability.

[0032] Provide quantifiable and alignable network metrics so that the central end can use these metrics to perform causal analysis with Q, thereby avoiding misclassifying process defects as network problems or vice versa.

[0033] The joint anomaly score is calculated based on the aligned PTP timeline, and the calculation formula is as follows: ;in, , , These represent the jitter standard value, packet loss standard value, and PTP offset standard value, respectively. Simultaneously, time-series causal analysis (such as Granger causality or transfer entropy) is used to confirm whether changes in network metrics pre-influence a decrease in the single weld quality score. If determined to be yes, anomaly scoring is combined. If the score exceeds the set anomaly score threshold, it is determined to be a network-related fault; otherwise, it is determined to be a process-related fault.

[0034] Incorporate 1−Q to represent the intensity of mass decrease; include J, L, Standardization and summation allow indicators with different dimensions to be compared in the same score; the weight λ is used to reflect the relative impact of different indicators on quality.

[0035] The Granger test and other causal analyses are used to avoid drawing conclusions based solely on correlation (correlation ≠ causation) and to improve the accuracy and interpretability of judgments.

[0036] By combining joint scoring with causal testing, the false positive rate is reduced, enabling the system to more accurately adopt different response strategies such as network redundancy switching or process parameter adjustment, thus avoiding unnecessary downtime or incorrect adjustments.

[0037] The circular caching strategy calculates the single welding quality score Q in real time and generates a welding event fingerprint F, while storing high-resolution frames and short time windows of the original waveform locally. The feedback is determined based on semantic gating logic. When the single welding quality score Q is less than the set quality score threshold or the combined abnormal score is less than the threshold, the feedback is determined. When the score exceeds the set anomaly threshold, high-resolution data for the corresponding time period is uploaded from the circular cache to the central storage; otherwise, only the welding event fingerprint F is uploaded.

[0038] While ensuring backtracking capabilities, it greatly reduces bandwidth and storage usage under normal conditions, so that high-frequency vision does not occupy real-time control network resources and ensures stable production cycle.

[0039] When the original waveform upload decision module determines that the fault is process-related, the process attribution process is executed, including defining an adjustment parameter set, including current I, dwell time dT, and welding speed v. Minimum / maximum adjustment range and step size are set for each parameter, and micro-step parameter adjustments are performed at the edges according to the following formula, with the effect evaluated: ,in, Indicates the standard welding current. The coefficient is an empirical factor, and the single welding quality score Q curve is monitored for N welding events after adjustment. If the average welding quality score Q improves and the average value is greater than the initial Q, it is retained and recorded; if there is no improvement or the quality deteriorates, it is rolled back and an alarm is triggered.

[0040] The fine-tuning formula is driven by 1−Q; the lower the Q, the larger the adjustment, making the adjustment magnitude proportional to the severity of the anomaly. Control the adjustment rate to prevent new process deviations from occurring.

[0041] Adopting a limited number of assessments plus rollback strategy can automatically correct short-term deviations without taking significant risks, reducing the frequency of manual intervention.

[0042] When the original waveform upload decision module determines that the fault is process-related, the network attribution process is executed, including switching the key control from remote control mode to local control mode of the local PLC; rolling back the most recent network configuration change; and prioritizing the reporting of the labeled image signal and waveform data of the corresponding welding position for troubleshooting.

[0043] Local control mode prevents network anomalies from further impacting the process, ensuring production safety or allowing production to continue according to a minimized failure strategy, reducing the occurrence of defects. Automatic rollback and QoS adjustment can quickly restore the determinism of the control plane. Control is quickly rolled back to a reliable channel or safe state, minimizing large-scale defects or downtime losses caused by network fluctuations.

[0044] The network-quality joint causal analysis model in the causal determination model update module is used to train all labeled events as training samples, periodically train and update the welding quality scoring model and joint anomaly discrimination weights, and after acceptance, it is distributed to the edge nodes in a versioned manner. After receiving the new model, the edge nodes perform gray-scale deployment and report the effect indicators to determine whether to replace it completely.

[0045] The online learning model can adapt to changes in the production environment and materials. Versioning and canary deployment mechanisms reduce deployment risks and ensure model reliability.

[0046] The causal determination model update module further includes a similar event retrieval module based on fingerprint F. This module indexes the historical fingerprint set (based on LSH or a vector database) to achieve approximate retrieval of newly reported abnormal fingerprints and automatically returns process parameters, network logs, and handling records of historical events to assist in rapid location and judgment. First, the central end establishes an LSH index or vectorizes the historical F. When a new F is reported and judged as abnormal, it automatically retrieves the k most similar historical events and presents the corresponding root cause and handling effect. This leverages historical experience to achieve rapid knowledge transfer, reducing repetitive investigation time and improving handling efficiency.

[0047] Each triggered upload of high-resolution images, raw waveforms, and fingerprints is indexed and stored using the Vehicle Identification Number (VIN) and cycle ID. Each handling record (such as welding repair, parameter adjustment, and line stop records) and corresponding data are stored as associated entries in the Traceability DB for subsequent traceability and responsibility attribution. The implementation steps include: generating a unique cycle ID for each workstation and each vehicle in the production cycle system and ensuring the binding of PTP time and cycle ID; the central terminal writing data to the DB using VIN + cycle ID as the primary key when receiving uploaded data; and providing a query interface for searching by VIN, time, workstation, fingerprint, and other dimensions. This embodiment provides a complete and accurate quality traceability chain, facilitating quality management, recall processing, and responsibility determination.

[0048] The system also includes the following engineering deployment details: key switches and network interface cards support hardware PTP timestamps; PTP boundary clocks are deployed at the workstation aggregation switches; edge device computing power supports at least lightweight CNN inference and STFT / PCA operations; NPU or GPU is optional; and the system is first deployed in a demonstration unit and jitter injection tests are conducted for calibration. and T_; Redundant power supplies and local fast rollback mechanisms are tested at the edge; Information models (UANodeSet) are ensured to be consistent across different vendors to enable data semantic interoperability. This embodiment provides practical implementation parameters and conditions, facilitating rapid engineering pilot projects and large-scale rollout after small-scale validation.

[0049] The threshold and weight settings can be set by default according to the present invention, or they can be set by those skilled in the art.

[0050] Example 2 Figure 2 The flowchart of the communication coordination method for an automotive assembly line based on an industrial network according to the present invention is shown. Based on the same inventive concept as Embodiment 1, the present invention provides a communication coordination method for an automotive assembly line based on an industrial network, including the following steps: S1. For each edge computing node located at the welding station, calculate the single welding quality score Q in real time, generate welding event fingerprints, and implement circular caching; based on the threshold and the calculated joint anomaly score, decide whether to upload high-resolution data and original waveforms to the backend. S2. Perform time-sensitive network isolation on critical control flows, implement PTP hardware time synchronization and network indicator collection, and perform QoS management on control flows and quality return flows. S3, Receive welding event fingerprints reported from the edge, store backtracking data, execute the network-quality joint causal analysis model and issue model updates; S4. Based on the single welding quality score Q obtained by the original waveform upload decision module, the causal analysis results of the causal judgment model update module and the historical model, implement automatic welding parameter fine-tuning, repair welding triggering, local redundancy switching and rollback strategies.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A communication and coordination system for an automotive assembly line based on an industrial network, characterized in that: include: The welding sensing module includes: at least one set of welding electrical parameter acquisition devices, at least one set of vibration sensors, at least one low frame rate vision camera, and at least one high resolution ring buffer camera. The raw waveform upload decision module is used to calculate the single welding quality score Q in real time for the edge computing nodes located at each welding station, generate welding event fingerprints, and implement circular caching; based on the threshold and the calculated joint anomaly score, it decides whether to upload high-resolution data and raw waveforms to the backend. The Quality Backhaul Management module is used to perform time-sensitive network isolation for critical control flows, implement PTP hardware time synchronization and network metric collection, and perform QoS management for control flows and quality backhaul flows. The causal determination model update module is used to receive welding event fingerprints reported from the edge, store backtracking data, execute the network-quality joint causal analysis model, and issue model updates. The adaptive compensation module is used to implement automatic welding parameter fine-tuning, repair welding triggering, local redundancy switching and rollback strategies based on the single welding quality score Q obtained by the original waveform upload decision module, the causal analysis results of the causal judgment model update module and the historical model.

2. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 1, characterized in that: The original waveform upload decision module uses a single welding quality score Q, which is constructed by collecting electrical waveform signals, vibration signals, labeled images, and manual quality judgments from N car samples to build a training set; the original current... With voltage Short-time feature extraction is employed to calculate peak value, energy, duration, and root mean square (RMS), and each feature is normalized to obtain the normalized current feature. and voltage characteristics ; Bandpass filtering is applied to the vibration signal, and the principal components of the energy spectrum are calculated and normalized to obtain the normalized vibration characteristics. ; A lightweight CNN forward inference is performed on the labeled image signal to obtain the image score 𝑆; Current characteristics and voltage characteristics Vibration characteristics Image score The single welding quality score Q is obtained by linear weighting and then Sigmoid mapping.

3. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 2, characterized in that: The process of generating the welding event fingerprint includes: after each welding event is completed, the electrical waveform signal, vibration signal and labeled image are extracted according to the time window at the edge; the energy spectrum is extracted by Fourier transform and the first k principal component vectors are extracted by PCA; the first k principal component vectors are used to generate a summary by LSH, and combined into a welding event fingerprint F and reported.

4. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 3, characterized in that: The quality backhaul management module uses a PTP boundary clock and a transparent clock that support hardware timestamps. Edge nodes perform statistical network metrics on key flows, including one-way delay sequences. Shaking Packet loss rate PTP offset original amount and the offset after exponential smoothing .

5. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 4, characterized in that: The joint anomaly score is calculated based on the aligned PTP timeline, and the calculation formula is as follows: ;in, , , These represent the jitter standard value, packet loss standard value, and PTP offset standard value, respectively. Simultaneously, time-series causal analysis is used to confirm whether changes in network metrics pre-influence a decrease in the single welding quality score. If determined to be yes, anomaly scoring is also considered. If the score exceeds the set anomaly score threshold, it is determined to be a network-related fault; otherwise, it is determined to be a process-related fault.

6. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 5, characterized in that: The circular caching strategy calculates the single welding quality score Q in real time and generates a welding event fingerprint F, while storing high-resolution frames and short time windows of the original waveform locally. The feedback is determined based on semantic gating logic. When the single welding quality score Q is less than the set quality score threshold or the combined abnormal score is less than the threshold, the feedback is determined. When the score exceeds the set anomaly threshold, high-resolution data for the corresponding time period is uploaded from the circular cache to the central storage; otherwise, only the welding event fingerprint F is uploaded.

7. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 6, characterized in that: When the original waveform upload decision module determines that the fault is process-related, the process attribution process is executed, including defining an adjustment parameter set, including current I, dwell time dT, and welding speed v. Minimum / maximum adjustment range and step size are set for each parameter, and micro-step parameter adjustments are performed at the edges according to the following formula, with the effect evaluated: ,in, Indicates the standard welding current. The coefficient is an empirical factor, and the single welding quality score Q curve is monitored for N welding events after adjustment. If the average welding quality score Q improves, it is retained and recorded; if there is no improvement or it deteriorates, it is rolled back and an alarm is triggered.

8. The communication and coordination system for an automotive assembly line based on an industrial network according to claim 7, characterized in that: When the original waveform upload decision module determines that the fault is process-related, the network attribution process is executed, including switching the key control from remote control mode to local control mode of the local PLC; rolling back the most recent network configuration change; and prioritizing the reporting of the labeled image signal and waveform data of the corresponding welding position for troubleshooting.

9. The communication and coordination system for an automobile assembly line based on an industrial network according to claim 8, characterized in that: The network-quality joint causal analysis model in the causal determination model update module is used to train all labeled events as training samples, periodically train and update the welding quality scoring model and joint anomaly discrimination weights, and after acceptance, it is distributed to the edge nodes in a versioned manner. After receiving the new model, the edge nodes perform gray-scale deployment and report the effect indicators to determine whether to replace it completely.

10. A communication coordination method for an automobile assembly line based on an industrial network, used to execute the communication coordination system for an automobile assembly line based on an industrial network as described in any one of claims 1-9, characterized in that: Includes the following steps: S1. For each edge computing node located at the welding station, calculate the single welding quality score Q in real time, generate welding event fingerprints, and implement circular caching; based on the threshold and the calculated joint anomaly score, decide whether to upload high-resolution data and original waveforms to the backend. S2. Perform time-sensitive network isolation on critical control flows, implement PTP hardware time synchronization and network indicator collection, and perform QoS management on control flows and quality return flows. S3, Receive welding event fingerprints reported from the edge, store backtracking data, execute the network-quality joint causal analysis model and issue model updates; S4. Based on the single welding quality score Q obtained by the original waveform upload decision module, the causal analysis results of the causal judgment model update module and the historical model, implement automatic welding parameter fine-tuning, repair welding triggering, local redundancy switching and rollback strategies.

Images