A multi-machine cooperation scene and space-time coupling sensing big data analysis method and system for 3C manufacturing

Abstract

This invention proposes a method and system for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis in 3C manufacturing. It belongs to the fields of intelligent manufacturing, industrial big data analysis, and production process control. The method includes: constructing a three-level sensor network covering equipment, workstations, and production lines; deploying sensor nodes on equipment in the 3C manufacturing production line; generating sensor data; and determining the topological relationships and physical constraints between devices based on the sensor data, forming a dataset of device relationships and constraints. By constructing a three-level sensor network covering equipment, workstations, and production lines, and employing a series of innovative analysis methods, the diagnostic accuracy of 3C manufacturing production lines is greatly improved, enabling precise location of the root causes of multi-machine collaborative failures.

Description

Technical Field

[0001] This invention proposes a method and system for multi-machine collaborative scenarios and spatiotemporal coupled sensing big data analysis for 3C manufacturing, belonging to the fields of intelligent manufacturing, industrial big data analysis and production process control technology. Background Technology

[0002] In the 3C manufacturing sector, with the continuous increase in product precision and production complexity, the intelligent upgrading of production lines is urgently needed. However, current 3C manufacturing production lines face many challenging problems in data analysis and process control.

[0003] At the data level, various devices such as SMT placement machines and screw-fastening robots collect data independently, creating serious data silos. The data from these devices lacks spatiotemporal alignment, making collaborative analysis difficult. For example, a placement misalignment caused by vibration from device A being transmitted to device B is difficult to detect because the data cannot be examined from a collaborative perspective.

[0004] In terms of diagnostics, existing systems offer coarse-grained diagnostics, relying heavily on single-device threshold alarms, and are ineffective in handling multi-device coupled faults. For example, complex faults such as conveyor belt speed fluctuations and sudden acceleration changes in the robotic arm causing camera focusing failure are difficult to diagnose accurately.

[0005] In terms of process adjustment, process parameters are usually manually adjusted by shift or batch, and cannot be dynamically fine-tuned based on real-time sensor data, making it difficult to compensate for the impact of environmental temperature and humidity and material batch differences.

[0006] In algorithmic applications, publicly available time-series models fail to consider the unique equipment topology and physical constraints of 3C production lines, leading to prediction distortion. Moreover, existing technologies often remain at the level of generalized promotion such as "industrial big data platforms," ​​failing to deeply integrate multi-machine collaborative topology modeling and spatiotemporal coupled sensor fusion. This hinders the realization of an intelligent closed loop from collaborative anomaly root cause localization to process self-healing, severely restricting the efficient and stable operation of 3C manufacturing production lines. Summary of the Invention

[0007] This invention provides a method and system for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis in 3C manufacturing, to solve the problems mentioned in the background art above: This invention proposes a method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis in 3C manufacturing, the method comprising: S1. Construct a three-level sensor network covering equipment, workstations and production lines, deploy sensor nodes on the equipment in the 3C manufacturing production line, generate sensor data, determine the topological relationship and physical constraints between each device based on the sensor data, and form a device relationship and constraint dataset. S2. Based on the device relationship and constraint dataset, perform spatiotemporal alignment processing on multi-device sensing data to generate spatiotemporally coupled sensing data; extract multimodal features from the spatiotemporally coupled sensing data to obtain multimodal feature data. S3. Input the multimodal feature data into the self-developed multimodal time-series graph neural network to model the temporal correlation between multiple devices and generate multi-machine time-series correlation model data; at the same time, use the causal discovery algorithm to analyze the multimodal feature data, mine the potential causal relationship of multi-machine collaborative failure, and obtain multi-machine collaborative failure causal relationship data. S4. Based on the multi-machine time-series correlation model data and the multi-machine collaborative fault causal relationship data, perform root cause localization processing for multi-machine collaborative faults to generate fault root cause localization data; based on the fault root cause localization data, generate real-time dynamic adjustment strategies for key process parameters to obtain process parameter adjustment strategy data. S5. Adjust key process parameters in real time according to the process parameter adjustment strategy data, and continuously collect new sensor data during the adjustment process; update and optimize the multi-machine time sequence correlation model data and the multi-machine collaborative fault causal relationship data to form dynamically optimized analysis model data.

[0008] This invention proposes a system for implementing the multi-machine collaborative scenario and spatiotemporally coupled sensing big data analysis method for 3C manufacturing as described above, the system comprising: Node Deployment Module: Constructs a three-level sensor network covering equipment, workstations, and production lines; deploys sensor nodes on equipment in the 3C manufacturing production line; generates sensor data; determines the topological relationships and physical constraints between devices based on the sensor data; and forms a dataset of device relationships and constraints. Feature extraction module: Based on the device relationship and constraint dataset, it performs spatiotemporal alignment processing on multi-device sensor data to generate spatiotemporally coupled sensor data; it then performs multimodal feature extraction on the spatiotemporally coupled sensor data to obtain multimodal feature data. Relationship mining module: Input multimodal feature data into a self-developed multimodal time series graph neural network to model the temporal correlation between multiple devices and generate multi-machine time series correlation model data; at the same time, use causal discovery algorithm to analyze multimodal feature data, mine potential causal relationships of multi-machine collaborative failures, and obtain multi-machine collaborative failure causal relationship data; Strategy generation module: Based on multi-machine time-series correlation model data and multi-machine collaborative fault causal relationship data, it performs root cause localization processing for multi-machine collaborative faults and generates fault root cause localization data; based on the fault root cause localization data, it generates real-time dynamic adjustment strategies for key process parameters and obtains process parameter adjustment strategy data. Update and optimization module: Real-time dynamic adjustment of key process parameters according to process parameter adjustment strategy data, and continuous collection of new sensor data during the adjustment process; Update and optimize multi-machine time sequence correlation model data and multi-machine collaborative fault causal relationship data to form dynamically optimized analysis model data.

[0009] The beneficial effects of this invention are as follows: By constructing a three-level sensor network covering equipment, workstations, and production lines, and employing a series of innovative analysis methods, the diagnostic accuracy of 3C manufacturing production lines is greatly improved. It can accurately pinpoint the root cause of multi-machine collaborative failures, and even complex problems such as camera malfunctions caused by the linkage between conveyor belts and robotic arms can be accurately identified. It reduces the lag in process adjustments, allowing for real-time dynamic adjustments to key process parameters such as torque and pressure based on real-time sensor data, promptly compensating for the impact of environmental and material differences. It enhances the production line's ability to respond to abnormal situations, achieving an intelligent closed loop from locating the root cause of collaborative anomalies to process self-healing. It reduces production delays and product defect rates caused by untimely fault diagnosis and lagging process adjustments. It avoids the failure to detect potential faults due to data silos and coarse diagnostic granularity, thus preventing large-scale production accidents. It not only improves the production efficiency and product quality of 3C manufacturing production lines but also reduces production and operating costs, providing a new generation of intelligent analysis and control paradigms for high-precision electronic manufacturing. Attached Figure Description

[0010] Figure 1 This is a diagram illustrating the steps of the method described in this invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0011] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0012] One embodiment of the present invention, such as Figure 1 As shown, a method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis in 3C manufacturing is proposed, the method comprising: S1. Construct a three-level sensor network covering equipment, workstations, and production lines. Deploy sensor nodes on the equipment in the 3C manufacturing production line. The equipment includes SMT placement machines, screw-fastening robots, AOI inspection instruments, and functional test benches. Generate sensor data, which includes equipment-level, workstation-level, and production line-level data. Determine the topological relationships and physical constraints between each piece of equipment based on the sensor data to form an equipment relationship and constraint dataset. S2. Based on the device relationship and constraint dataset, perform spatiotemporal alignment processing on multi-device sensing data to unify the data collected independently by different devices into the same spatiotemporal coordinate system and generate spatiotemporal coupled sensing data; perform multimodal feature extraction on the spatiotemporal coupled sensing data to obtain multimodal feature data, which includes multi-dimensional information such as vibration, temperature, and pressure. S3. Input the multimodal feature data into the self-developed multimodal temporal graph neural network (MT-GNN), and use the network to model the temporal correlation between multiple devices to generate multi-machine temporal correlation model data; at the same time, use the causal discovery algorithm to analyze the multimodal feature data, mine the potential causal relationship of multi-machine collaborative failure, and obtain multi-machine collaborative failure causal relationship data. S4. Based on the multi-machine time-series correlation model data and the multi-machine collaborative fault causal relationship data, perform multi-machine collaborative fault root cause localization processing to determine the key equipment and key factors causing the fault and generate fault root cause localization data; based on the fault root cause localization data, generate real-time dynamic adjustment strategies for key process parameters, including torque and pressure, and obtain process parameter adjustment strategy data. S5. Based on the process parameter adjustment strategy data, perform real-time dynamic adjustments to the key process parameters of the 3C manufacturing production line, and continuously collect new sensor data during the adjustment process; update and optimize the multi-machine timing correlation model data and multi-machine collaborative fault causal relationship data based on the new sensor data to form dynamically optimized analysis model data; continuously monitor and evaluate the production line operation status using the dynamically optimized analysis model data to generate production line operation status evaluation data; generate intelligent analysis and control commands for the 3C manufacturing production line based on the production line operation status evaluation data, realizing an intelligent closed loop from collaborative anomaly root cause localization to process self-healing, and obtaining intelligent analysis and control command data.

[0013] The working principle and effects of the above technical solution are as follows: The three-level sensor network improves the comprehensiveness of the production line's sensing coverage and reduces blind spots in equipment collaborative status monitoring; spatiotemporal alignment processing improves the consistency of data from multiple devices and reduces analysis interference caused by data misalignment; multimodal feature extraction and MT-GNN modeling enhance the ability to capture the temporal correlation of multi-machine collaboration and improve the accuracy of collaborative status perception; causal discovery uncovers potential correlations in faults, reduces misjudgments in root cause localization, and avoids production losses caused by blindly adjusting processes; real-time dynamic parameter adjustment reduces the time spent on fault resolution and reduces production line interruption losses; dynamic model optimization enhances the environmental adaptability of the analysis method, avoids evaluation bias caused by model aging, and can both ensure the stable operation of multi-machine collaboration and improve the self-healing capability of the production line process, thereby improving overall production efficiency.

[0014] In one embodiment of the present invention, S1 includes: S11. Based on the equipment layout, workstation functional zoning and production line flow logic of the 3C manufacturing production line, design a three-level sensor network architecture for equipment, workstations and production line, determine the sensing accuracy, communication bandwidth and deployment priority of each level of sensor node, and generate a three-level sensor network deployment plan. S12. According to the deployment plan, in the key parts of the SMT placement machine, including the placement head assembly, the torque sensor module of the screw-locking robot, the image acquisition unit of the AOI detector, and the signal interface of the functional test bench, deploy multiple types of sensor nodes, including accelerometers, temperature sensors, pressure sensors, and high-definition vision sensors, to complete the protocol adaptation and communication debugging between the sensor nodes and the equipment control system and the production line cloud platform, and generate equipment-level real-time sensor data streams. S13. Perform station-level aggregation processing on the real-time sensor data streams of each device, associate the sensor data of different devices in the same station, such as the collaborative data of the screw-locking robot and the station positioning platform, remove data conflicts and invalid values ​​in the station, and generate station-level fused sensor data. S14. The fusion sensor data of each workstation is integrated at the production line level according to the production line flow sequence, and the associated data of the production line conveying equipment is supplemented. The associated data includes speed and position, forming a comprehensive production line-level sensor data covering the entire production line. S15. Based on device-level, workstation-level, and production line-level sensor data, the connection relationship, spatial location distribution, and motion interference boundary of each device are identified through device MAC address parsing and physical coordinate mapping algorithms, and the topological association rules between devices are determined. S16. Combining the 3C manufacturing process specifications and equipment hardware parameter limitations, the 3C manufacturing process specifications include a mounting accuracy of ±0.01mm and a locking torque of 5-8N・m; the equipment hardware parameter limitations include a maximum movement speed and a load limit. The physical constraints of each device are extracted, and the topology association rules are integrated with the physical constraints to generate a device relationship and constraint dataset.

[0015] The working principle and effects of the above technical solution are as follows: The three-level sensor network architecture improves the hierarchy and targeting of production line sensing coverage, reducing the omission of monitoring key equipment parts; Protocol adaptation and communication debugging improve the stability of data transmission and reduce the loss of sensor data due to connection failures; Workstation-level aggregation and production line-level integration enhance the correlation and integrity of data, reducing the analytical limitations caused by isolated data from single devices; Equipment topology association identification improves the accuracy of perception of relationships between devices, avoiding misjudgments of spatial location and connection relationships in collaborative analysis; The fusion of physical constraints improves the adaptability of the dataset, avoiding deviations caused by parameter mismatches in subsequent analysis, ensuring both the comprehensiveness and reliability of sensor data, and laying a solid data foundation for subsequent spatiotemporal alignment and collaborative modeling of multiple devices.

[0016] In one embodiment of the present invention, S11 includes: Collect equipment layout drawings, workstation function manuals and production line flow sequence tables of 3C manufacturing production lines, extract core information including equipment distribution coordinates, workstation operation range and process connection logic, and generate a production line basic information dataset. Based on the production line basic information dataset, we analyze the differentiated sensing requirements for single-machine operation monitoring, workstation collaborative linkage perception, and overall production line status control, clarify the hierarchical functional positioning of the three-level sensor network, and generate a list of network functional requirements. Based on the network functional requirements list, a three-level sensor network architecture framework is built, which includes equipment terminal sensing, workstation aggregation and transmission, and production line overall management. The coverage, data interaction protocols, and inter-level communication links of each level of the network are defined, and a preliminary network architecture is generated. Based on the initial architecture and combined with the precision requirements of key processes in 3C manufacturing, including SMT assembly and screw fastening, the sensing accuracy threshold of equipment-level sensing nodes and the communication bandwidth standard of workstation-level nodes are determined, and a core parameter configuration table for nodes is generated. Based on the priority of core processes in the production line and the impact of equipment failures, the deployment order of sensor nodes at all levels is sorted, and three deployment priority levels (high, medium, and low) are defined to generate a node deployment priority scheme. By integrating the initial network architecture, node core parameter configuration table, and deployment priority scheme, and supplementing the technical specifications and security requirements for deployment and implementation, a complete three-level sensor network deployment plan is generated.

[0017] The working principle and effects of the above technical solution are as follows: Systematic collection of basic production line information enhances the comprehensiveness of planning basis and reduces the blind spots in subsequent network design; differentiated sensing requirement analysis enhances the accuracy of network function positioning and avoids redundant or missing hierarchical functions; the construction of a three-level architecture framework improves the logic of network design and reduces conflicts in data interaction between levels; core parameter configuration aligns with the accuracy requirements of key processes, improving the adaptability of sensor nodes and avoiding insufficient monitoring accuracy or bandwidth waste; deployment priority ranking improves implementation efficiency and reduces the lag in monitoring core process equipment; the solution integrates and supplements technical specifications and safety requirements, enhancing the integrity of the plan and avoiding technical vulnerabilities and security risks during deployment. This ensures a high degree of compatibility between the sensor network and the production line and provides clear and reliable execution guidance for subsequent node deployment.

[0018] In one embodiment of the present invention, S2 includes: S21. Extract the clock synchronization reference signal and physical coordinate origin parameters of each device from the device relationship and constraint dataset, construct a globally unified spatiotemporal coordinate system, and define the mapping function and error threshold for data spatiotemporal alignment. S22. Based on the global spatiotemporal coordinate system, timestamp calibration and spatial coordinate transformation are performed on the sensor data at the device level, workstation level, and production line level to generate a preliminary spatiotemporal aligned dataset. The timestamp calibration uses the NTP network time protocol to eliminate clock deviation. The spatial coordinate transformation unifies the position reference through homogeneous coordinate transformation. S23. Perform coupling verification on the initial spatiotemporal aligned dataset. By calculating the spatiotemporal correlation coefficient of data from different devices, such as the Pearson correlation coefficient, abnormal data with spatiotemporal correlation below the threshold are removed, and high-precision spatiotemporal coupled sensing data is generated. S24. Construct a multimodal feature extraction framework. For vibration signals in spatiotemporally coupled sensing data, use time-domain analysis to extract time-domain features, including peak value, kurtosis, and impulse factor. Use frequency-domain analysis to extract fundamental frequency-domain features, including frequency and harmonic components. For statistical features, including temperature and pressure data, extract mean, variance, and rate of change. For visual data, use CNN to extract spatial features, including image texture and contour. S25. Perform feature filtering and weighted fusion on the above-mentioned multi-dimensional basic features to generate multimodal feature data. The feature filtering is based on mutual information entropy to remove redundant features. The weighted fusion adopts an attention mechanism to allocate feature importance weights. The multimodal feature data includes multi-dimensional key information such as vibration, temperature, pressure, and vision.

[0019] The working principle and effects of the above technical solution are as follows: the construction of a globally unified spatiotemporal coordinate system improves the consistency of data from multiple devices and reduces misalignment interference caused by differences in spatiotemporal references between different devices; timestamp calibration and spatial coordinate transformation reduce the impact of clock deviation and inconsistent position references, improving the accuracy of data alignment; coupling verification enhances the reliability of spatiotemporally coupled sensing data by eliminating low-correlation abnormal data and reduces the interference of invalid data on subsequent analysis; the multimodal feature extraction framework covers multiple types of signals and data, improving the comprehensiveness of feature information and avoiding the limitations of single-dimensional features; feature filtering and weighted fusion reduce the interference of redundant features and improve the prominence of key features; the entire process can ensure the spatiotemporal uniformity and accuracy of sensing data, and extract core features that meet the analysis requirements, providing high-quality data support for subsequent multi-machine correlation modeling and fault causal mining.

[0020] In one embodiment of the present invention, step S3 includes: S31. Perform time-series segmentation processing on the multimodal feature data, divide the sub-time-series samples according to the single process cycle of 3C products, for example, the mounting process is 20s / piece, and use multiple methods to perform data augmentation, including time flipping and Gaussian noise injection, to generate a time-series feature dataset suitable for model training. S32. Input the time series feature dataset into the self-developed multimodal temporal graph neural network (MT-GNN). The long-term and short-term time dependencies of the data are captured through the temporal attention layer in the model. The graph convolutional layer models the spatial relationship topology between devices. The spatiotemporal dual-dimensional features are fused to complete the multi-device collaborative temporal relationship modeling and generate a preliminary multi-machine temporal relationship model. S33. The performance of the preliminary multi-machine time series correlation model is verified using test set data. The model's metrics are calculated, including the time series correlation prediction accuracy and correlation delay error. The hyperparameters of the model are optimized using a grid search algorithm, including the learning rate and hidden layer dimension, to generate high-precision multi-machine time series correlation model data. S34. For the fault sample subset in multimodal feature data, a causal discovery algorithm based on structural equation model is adopted to construct a causal relationship graph between features and identify potential paths and key intermediate variables of fault propagation. S35. Perform a significance test on the causal relationship graph. The significance test includes using a permutation test to verify the reliability of the causal relationship, eliminating false association paths, clarifying the triggering source, transmission link and impact range of the multi-machine collaborative failure, and generating causal relationship data of the multi-machine collaborative failure.

[0021] The working principle and effects of the above technical solution are as follows: Temporal segmentation processed according to a single process cycle improves the fit between the samples and actual production; data augmentation reduces the interference of insufficient or unevenly distributed samples on model training; the fusion of temporal attention and graph convolutional layers in MT-GNN enhances the ability to capture spatiotemporal dual-dimensional features and improves the accuracy of multi-device collaborative temporal correlation modeling; model performance verification and hyperparameter optimization reduce prediction errors and improve the reliability of the temporal correlation model; causal discovery algorithms uncover potential fault correlations and reduce the blindness of multi-machine collaborative fault analysis; significance testing eliminates false correlation paths, avoids misjudgment of fault propagation paths and trigger sources, and clarifies the fault transmission logic; the entire process can ensure high accuracy of multi-machine correlation modeling and accurately locate fault causal relationships, providing solid model and data support for subsequent fault root cause localization and process adjustment.

[0022] In one embodiment of the present invention, step S31 includes: Extract the standard operating cycle parameters of each core process in 3C manufacturing, and combine them with the actual production cycle record of the production line to generate a single process cycle parameter comparison table. The core processes include, for example, mounting, screw fastening, and inspection. Based on the single-process cycle parameter reference table, the time-series segmentation window length and sliding step size are set, the multimodal feature data is segmented by window, and sub-time-series samples are divided according to the process type to generate an initial sub-time-series sample set. The initial sub-time series sample set is validated, and abnormal samples with insufficient length or missing data rate exceeding the threshold are removed. Slightly missing samples are filled in using linear interpolation to generate a standardized sub-time series sample set. Analyze the data distribution characteristics of the standardized sub-time series sample set, determine the applicable scenarios and parameter thresholds for time flipping and Gaussian noise injection, the parameter thresholds include Gaussian noise intensity and the proportion of flipped time series, and formulate implementation details for multiple data augmentation methods; Based on the implementation details, data augmentation operations are performed on scarce process samples and fault-related samples in the standardized sub-time series sample set to generate an augmented sub-time series sample set. The standardized sub-time series sample set and the enhanced sub-time series sample set are merged, and the training subset and validation subset are divided according to a preset ratio. Sample labels are added to generate a time series feature dataset adapted for model training. The sample labels include process type and data source equipment.

[0023] The working principle and effects of the above technical solution are as follows: combining core process cycle parameters with actual production rhythm improves the accuracy of segmentation criteria and reduces the blindness of time-series segmentation; windowing and segmentation of samples by process enhances the fit between samples and actual production, avoiding the problem of samples being out of sync with processes; validity verification removes abnormal samples and fills in missing data, improving sample quality and reducing the interference of invalid data on model training; data augmentation prioritizes the processing of scarce and faulty samples, balancing the sample distribution and avoiding model training bias caused by sample imbalance; sample fusion and label addition improve the adaptability of the dataset, ensuring the reasonable division of training and validation data; the entire process can not only ensure the integrity and reliability of the time-series feature dataset, but also provide accurately adapted input data for subsequent MT-GNN model training, improving the efficiency and effectiveness of model training.

[0024] In one embodiment of the present invention, S32 includes: The time series feature dataset is processed for model input adaptation and converted into a tensor format that can be recognized by the self-developed multimodal temporal graph neural network (MT-GNN). Simultaneously, data dimension normalization and sample label alignment are completed to generate MT-GNN adapted time series feature data. The MT-GNN is adapted to input the temporal feature data into the model's temporal attention layer, the attention weight parameters are initialized, the association weights of data at different temporal step sizes are calculated through the self-attention mechanism, key temporal information is filtered, the long-term and short-term time dependencies of the data are captured, and a set of temporal dependency feature vectors is generated. The device topology association rules in the device relationship and constraint dataset are retrieved to construct the device spatial association graph structure. This graph structure is then input into the graph convolutional layer of the MT-GNN along with the temporal dependency feature vector set. The spatial association features between devices are extracted through graph convolution operations to generate a spatial association feature matrix. Through the adaptive spatiotemporal fusion module, the temporally dependent feature vector set and the spatially related feature matrix are matched in dimension and weighted. The weighted summation algorithm is used to complete the deep fusion of spatiotemporal dual-dimensional features and generate spatiotemporal fusion feature data. The spatiotemporal fusion feature data is input into the fully connected layer and output layer of MT-GNN. The model parameters are optimized through the backpropagation algorithm to complete the modeling and training of multi-device collaborative temporal correlation relationship and generate a preliminary multi-machine temporal correlation model.

[0025] The working principle and effects of the above technical solution are as follows: Data adaptation processing realizes format conversion and normalization, which improves the adaptability of temporal feature data and MT-GNN and reduces training obstacles caused by format incompatibility or dimensional differences; the temporal attention layer filters key information through association weights, enhances the ability to capture long-term and short-term time dependencies, and reduces the interference of irrelevant temporal data; combining equipment topology rules to construct equipment spatial association graphs makes spatial association feature extraction more in line with the actual production line, improves the accuracy of spatial features, and avoids modeling deviations that are detached from the actual equipment layout; adaptive spatiotemporal fusion realizes deep integration of dual-dimensional features, improves the integrity of feature information, and reduces the analytical limitations caused by single-dimensional features; backpropagation optimizes model parameters, improves the accuracy of multi-device collaborative temporal association modeling, and the generated preliminary model provides a reliable foundation for subsequent performance optimization, which can not only ensure the smooth and efficient modeling process, but also make the model better fit the actual temporal association characteristics of multi-machine collaboration.

[0026] In one embodiment of the present invention, step S4 includes: S41. Integrate multi-machine time-series correlation model data and multi-machine collaborative fault causal relationship data to construct a three-layer fault root cause localization inference model of features, correlation and causation, and define the confidence threshold and inference rules for root cause determination. S42. The spatiotemporal coupled sensor data collected in real time from the production line under fault conditions is extracted and input into the inference model. By tracing back to the starting point of the causal relationship path and combining the dynamic prediction results of the temporal relationship model, the key equipment and core influencing factors that cause multi-machine collaborative failure are identified, and preliminary fault root cause localization data is generated. The key equipment is, for example, a screw-locking robot; the core influencing factors are, for example, torque fluctuation and temperature drift. S43. Cross-validate the preliminary root cause localization data by combining equipment historical fault records and process standard documents, and correct the root cause judgment bias. The correction of the root cause judgment bias includes, for example, distinguishing between equipment faults and collaborative interaction faults, and generating accurate fault root cause localization data. S44. Based on accurate fault root cause location data, analyze the quantitative relationship between key influencing factors and process quality indicators, including product mounting pass rate and mounting yield. Establish an objective function for adjusting process parameters, with the objective function for adjusting process parameters having the dual objectives of the lowest fault rate and the highest production efficiency. S45. An improved particle swarm optimization algorithm is used to solve the objective function, generate the real-time adjustment range, adjustment step size and adjustment sequence of key process parameters, and obtain process parameter adjustment strategy data. The key process parameters include clamping torque, mounting speed and detection threshold.

[0027] The working principle and effects of the above technical solution are as follows: The three-layer fault root cause localization reasoning model integrates temporal correlation and causal relationship data, improving the comprehensiveness of root cause localization and reducing judgment bias caused by single data support; the reverse tracing of causal paths combined with temporal prediction results improves the accuracy of locking key equipment and core influencing factors, reducing misjudgment of multi-machine collaborative fault root causes; cross-validation combined with historical records and process standards corrects root cause judgment bias, enhances the reliability of root cause data, and avoids confusion between equipment faults and collaborative interaction faults; the analysis of the quantitative relationship between key factors and quality indicators establishes a dual objective function, balancing fault control and production efficiency, and avoiding losses caused by one-sided process adjustments; the improved particle swarm optimization algorithm solves and generates specific adjustment parameters, improving the operability of process adjustment strategies and reducing the blindness of parameter adjustments; the entire process can quickly and accurately locate fault root causes and formulate scientific and reasonable process adjustment plans, improving production line fault handling efficiency and reducing production interruption losses.

[0028] In one embodiment of the present invention, S42 includes: Preprocess the spatiotemporal coupled sensor data collected in real time from the production line under fault conditions, remove data noise, fill in missing values ​​and complete format standardization to generate standardized sensor data under fault conditions. The multimodal feature extraction framework of S2 is reused to perform targeted feature extraction on standardized sensor data of fault conditions, retaining core features with high correlation to faults, including vibration, temperature and pressure, and generating fault condition-adaptive feature data. Input the fault condition adaptation feature data into the three-layer fault root cause localization reasoning model, activate the causal relationship path tracing module in the model, load the preset causal relationship graph structure, and generate path tracing initialization parameters. Based on the initialization parameters, the reverse tracing algorithm is executed. Starting from the terminal node corresponding to the fault characteristics, the algorithm traces back to the starting point of the path layer by layer along the causal relationship path, filters out potential root cause related nodes, and generates a set of candidate root cause nodes. The dynamic prediction results of the multi-machine time series correlation model are retrieved, the equipment coordination time series features during the fault occurrence period are extracted, and the time series and causal correlation are matched with the candidate root cause node set to calculate the matching confidence. Based on the preset confidence threshold, the matching results are filtered and false root cause nodes with confidence levels below the threshold are eliminated, the key equipment and core influencing factors that cause multi-machine collaborative failure are identified, and preliminary root cause location data is generated.

[0029] The working principle and effects of the above technical solution are as follows: Preprocessing of fault condition sensor data removes noise and fills in missing values, improving data purity and reducing interference from impurities in subsequent analysis; reusing a mature multimodal feature extraction framework to extract core features enhances the correlation between features and faults, reducing redundant interference from irrelevant features; activating the causal path tracing module and loading a preset graph structure makes the tracing process more directional, reducing inefficiency caused by blind tracing; reverse tracing back from the fault terminal node layer by layer improves the accuracy of potential root cause node screening and avoids the omission of key root cause nodes; combining time series and causal correlation matching with dynamic prediction results enhances the reliability of root cause determination and reduces bias caused by single-dimensional analysis; confidence threshold screening removes false nodes, improving the accuracy of preliminary root cause location data and avoiding the ineffective costs of subsequent root cause verification. This approach can quickly identify key fault equipment and core factors, and lay a solid foundation for accurate root cause determination.

[0030] In one embodiment of the present invention, S45 includes: Extract the equipment hardware limitations and process specification requirements of key process parameters, determine the value boundaries of each parameter, and generate a set of parameter constraints; based on the set of parameter constraints, initialize the core parameters of the improved particle swarm optimization algorithm, which includes particle population size, learning factor, maximum number of iterations, and inertia weight decay coefficient, and generate algorithm initialization configuration data; The established dual objective function and parameter constraint set are input into the initialized improved particle swarm optimization algorithm to start the iterative optimization process. The optimal fitness value of each generation of particles is calculated through the particle position and velocity update formula, and the iterative optimization process dataset is generated. Set an iteration convergence judgment threshold. When the fluctuation of the optimal fitness value is less than the threshold for multiple consecutive generations, stop the iteration and extract the global optimal solution. Map the real-time adjustment range of each key process parameter and generate a parameter adjustment range dataset. By combining the real-time operating cycle time of the production line and the response delay characteristics of the equipment, the parameter adjustment range is subdivided, the difference between adjacent adjustment nodes is calculated as the adjustment step size, and parameter adjustment step size data is generated. Based on the multi-machine collaborative timing association rules, the order and time nodes for adjusting each key process parameter are planned to avoid parameter adjustment conflicts and generate a parameter adjustment timing scheme. The parameter adjustment interval dataset, adjustment step size data and adjustment timing scheme are integrated to supplement the execution conditions and safety warning thresholds for parameter adjustment and generate complete process parameter adjustment strategy data. The working principle and effects of the above technical solution are as follows: Extracting the parameter constraint set clarifies the value boundaries, preventing parameter adjustments from exceeding equipment hardware and process specification limitations, and improving the compliance of adjustments; initializing the core parameters of the algorithm ensures the stable start of iterative optimization, reducing confusion and deviation in the optimization process; iterative optimization combines a dual objective function to balance the needs of minimizing the failure rate and maximizing production efficiency, improving the adaptability of the optimal solution and avoiding production losses caused by one-sided optimization; precise mapping of adjustment intervals reduces the problem of parameter range ambiguity, and step size calculation closely matches the production line cycle time and equipment response delay, making parameter adjustments smoother and reducing equipment operation fluctuations; timing planning is based on multi-machine collaborative timing rules to avoid parameter adjustment conflicts and improve the coordination of multi-parameter collaborative adjustments; strategy fusion supplements execution conditions and safety warning thresholds, enhancing the integrity and security of the solution and reducing risks during execution; the entire process can generate accurately adapted process adjustment parameters and ensure smooth and controllable adjustment implementation, improving the efficiency and reliability of production line process self-healing.

[0031] In one embodiment of the present invention, step S5 includes: S51. Convert the process parameter adjustment strategy data into a control instruction format that the equipment can execute, such as PLC ladder diagram instructions or robot motion control code, and send it to the corresponding key equipment controller in real time via industrial Ethernet to complete the dynamic adjustment of process parameters. S52. During the parameter adjustment process, the equipment operating parameter data, process data and product quality inspection data are continuously collected through a three-level sensor network to generate a real-time sensor dataset after adjustment. S53. Reuse the standardized processing flow of S1 and the spatiotemporal alignment and feature extraction flow of S2 on the adjusted real-time sensing dataset to generate new multimodal feature data as incremental data for model updates. S54. Employ online incremental learning algorithms, such as incremental gradient descent, to integrate new multimodal feature data into the multi-machine time-series correlation model and the multi-machine collaborative fault causal relationship model, update the model weight parameters, eliminate model drift, and generate dynamically optimized analysis model data. S55. Based on the dynamic optimization analysis model data, the core indicators of the production line are monitored in real time. The core indicators include equipment collaboration efficiency, process parameter stability, and product quality pass rate, generating multi-dimensional production line operation status monitoring data. A production line operation status evaluation index system is set up. The production line operation status evaluation index system includes three primary indicators: equipment status, collaboration efficiency, and quality level, and eight secondary indicators. The analytic hierarchy process is used to determine the index weights, and the monitoring data is quantitatively scored to generate production line operation status evaluation data. S56. Based on the evaluation data results, if the index score is higher than the qualified threshold, a stable control instruction to maintain the current process parameters is generated; if there is a potential abnormal risk (score close to the threshold), an early warning control instruction for early intervention is generated; if a new fault characteristic appears (score below the threshold), a fault handling control instruction for rapid response is generated; through the above process, an intelligent closed loop of anomaly detection, root cause localization, parameter adjustment, model update, status evaluation and instruction output is formed, obtaining intelligent analysis and control instruction data covering the entire scenario, realizing adaptive optimization and process self-healing of multi-machine collaboration in 3C manufacturing production lines.

[0032] The working principle and effects of the above technical solution are as follows: Control command format conversion and real-time issuance improve the timeliness and accuracy of process parameter adjustments, reducing the lag and errors of manual operation; continuous collection of multiple types of data ensures the comprehensiveness of incremental data, providing reliable support for model updates; reuse of mature data processing processes reduces redundant development and improves data processing efficiency; online incremental learning updates model parameters, eliminates model drift, enhances the environmental adaptability of the analysis model, and avoids evaluation bias caused by model aging; multi-dimensional monitoring and quantitative scoring improve the accuracy of production line status assessment and reduce the omission of potential abnormal risks; hierarchical control commands enable early intervention and rapid fault response, preventing small faults from escalating into serious production problems; intelligent closed-loop throughout the entire process enables adaptive optimization and process self-healing of multi-machine collaboration on the production line, improving the stability and continuity of production line operation, reducing production interruption losses, ensuring production quality, and improving overall production efficiency.

[0033] One embodiment of the present invention, such as Figure 2 As shown, a system for implementing the multi-machine collaborative scenario and spatiotemporally coupled sensing big data analysis method for 3C manufacturing as described above is provided. The system includes: Node Deployment Module: Constructs a three-level sensor network covering equipment, workstations, and production lines. Deploys sensor nodes on equipment in the 3C manufacturing production line, including SMT placement machines, screw-fastening robots, AOI inspection instruments, and functional testing benches. Generates sensor data, including equipment-level, workstation-level, and production line-level data. Determines the topological relationships and physical constraints between each device based on the sensor data, forming a device relationship and constraint dataset. Feature extraction module: Based on the device relationship and constraint dataset, it performs spatiotemporal alignment processing on multi-device sensing data, unifying the data collected independently by different devices into the same spatiotemporal coordinate system to generate spatiotemporally coupled sensing data; it performs multimodal feature extraction on the spatiotemporally coupled sensing data to obtain multimodal feature data, which includes multi-dimensional information such as vibration, temperature, and pressure; Relationship mining module: Input multimodal feature data into the self-developed multimodal temporal graph neural network (MT-GNN), use the network to model the temporal correlation between multiple devices, and generate multi-machine temporal correlation model data; at the same time, use the causal discovery algorithm to analyze the multimodal feature data, mine the potential causal relationship of multi-machine collaborative failure, and obtain multi-machine collaborative failure causal relationship data; Strategy generation module: Based on multi-machine time-series correlation model data and multi-machine collaborative fault causal relationship data, it performs root cause localization processing of multi-machine collaborative faults, identifies key equipment and key factors causing the fault, and generates fault root cause localization data; based on the fault root cause localization data, it generates real-time dynamic adjustment strategies for key process parameters, including torque and pressure, and obtains process parameter adjustment strategy data. The update and optimization module dynamically adjusts key process parameters of the 3C manufacturing production line in real time according to the process parameter adjustment strategy data, and continuously collects new sensor data during the adjustment process; it updates and optimizes the multi-machine time sequence correlation model data and multi-machine collaborative fault causal relationship data based on the new sensor data, forming dynamically optimized analysis model data; it continuously monitors and evaluates the production line operation status using the dynamically optimized analysis model data, generating production line operation status evaluation data; and it generates intelligent analysis and control commands for the 3C manufacturing production line based on the production line operation status evaluation data, realizing an intelligent closed loop from collaborative anomaly root cause localization to process self-healing, and obtaining intelligent analysis and control command data.

[0034] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for multi-machine collaboration scene and spatiotemporal coupling sensing big data analysis for 3C manufacturing, characterized in that, The method includes: S1. Construct a three-level sensor network covering equipment, workstations and production lines, deploy sensor nodes on the equipment in the 3C manufacturing production line, generate sensor data, determine the topological relationship and physical constraints between each device based on the sensor data, and form a device relationship and constraint dataset. S2. Based on the device relationship and constraint dataset, perform spatiotemporal alignment processing on multi-device sensing data to generate spatiotemporally coupled sensing data; extract multimodal features from the spatiotemporally coupled sensing data to obtain multimodal feature data. S3. Input the multimodal feature data into the self-developed multimodal time-series graph neural network to model the temporal correlation between multiple devices and generate multi-machine time-series correlation model data; at the same time, use the causal discovery algorithm to analyze the multimodal feature data, mine the potential causal relationship of multi-machine collaborative failure, and obtain multi-machine collaborative failure causal relationship data. S4. Based on the multi-machine time-series correlation model data and the multi-machine collaborative fault causal relationship data, perform root cause localization processing for multi-machine collaborative faults to generate fault root cause localization data; based on the fault root cause localization data, generate real-time dynamic adjustment strategies for key process parameters to obtain process parameter adjustment strategy data. S5. Adjust key process parameters in real time according to the process parameter adjustment strategy data, and continuously collect new sensor data during the adjustment process; update and optimize the multi-machine time sequence correlation model data and the multi-machine collaborative fault causal relationship data to form dynamically optimized analysis model data.

2. The method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis for 3C manufacturing as described in claim 1, characterized in that, S1 includes: S11. Based on the equipment layout, workstation functional zoning and production line flow logic of the 3C manufacturing production line, design a three-level sensor network architecture for equipment, workstations and production line, determine the sensing accuracy, communication bandwidth and deployment priority of each level of sensor node, and generate a three-level sensor network deployment plan. S12. According to the deployment plan, deploy multiple types of sensor nodes in key parts of the SMT placement machine, complete the protocol adaptation and communication debugging between the sensor nodes and the equipment control system and production line cloud platform, and generate equipment-level real-time sensor data streams. S13. Perform station-level aggregation processing on the real-time sensor data streams of each device to generate station-level fused sensor data. S14. Integrate the fused sensor data of each workstation according to the production line flow sequence, supplement the associated data of the production line conveying equipment, and form a comprehensive production line-level sensor data covering the entire production line. S15. Based on device-level, workstation-level, and production line-level sensor data, the connection relationship, spatial location distribution, and motion interference boundary of each device are identified through device MAC address parsing and physical coordinate mapping algorithms, and the topological association rules between devices are determined. S16. Combining 3C manufacturing process specifications and equipment hardware parameter limitations, extract the physical constraints of each device, integrate the topology association rules with the physical constraints, and generate a device relationship and constraint dataset.

3. The method for multi-machine collaborative scenarios and spatiotemporal coupled sensing big data analysis for 3C manufacturing according to claim 2, characterized in that, S11 includes: Collect equipment layout drawings, workstation function manuals, and production line flow sequence tables of 3C manufacturing production lines, extract core information, and generate a production line basic information dataset; Based on the production line basic information dataset, we analyze the differentiated sensing requirements for single-machine operation monitoring, workstation collaborative linkage perception, and overall production line status control, clarify the hierarchical functional positioning of the three-level sensor network, and generate a list of network functional requirements. Based on the network functional requirements list, a three-level sensor network architecture framework is built, which includes equipment terminal sensing, workstation aggregation and transmission, and production line overall management. The coverage, data interaction protocols, and inter-level communication links of each level of the network are defined, and a preliminary network architecture is generated. Based on the initial architecture and combined with the precision requirements of key processes in 3C manufacturing, the sensing accuracy threshold of equipment-level sensing nodes and the communication bandwidth standard of workstation-level nodes are determined, and a core parameter configuration table for nodes is generated. Based on the priority of core processes in the production line and the impact of equipment failures, the deployment order of sensor nodes at all levels is sorted, and three deployment priority levels (high, medium, and low) are defined to generate a node deployment priority scheme. By integrating the initial network architecture, node core parameter configuration table, and deployment priority scheme, and supplementing the technical specifications and security requirements for deployment and implementation, a complete three-level sensor network deployment plan is generated.

4. The method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis for 3C manufacturing as described in claim 1, characterized in that, S2 includes: S21. Extract the clock synchronization reference signal and physical coordinate origin parameters of each device from the device relationship and constraint dataset, construct a globally unified spatiotemporal coordinate system, and define the mapping function and error threshold for data spatiotemporal alignment. S22. Based on the global spatiotemporal coordinate system, perform timestamp calibration and spatial coordinate transformation on device-level, workstation-level, and production line-level sensor data to generate a preliminary spatiotemporal aligned dataset. S23. Perform coupling verification on the preliminary spatiotemporal aligned dataset. By calculating the spatiotemporal correlation coefficient of data from different devices, remove abnormal data with spatiotemporal correlation below the threshold and generate high-precision spatiotemporal coupled sensing data. S24. Construct a multimodal feature extraction framework. For vibration signals in spatiotemporally coupled sensing data, use time-domain analysis to extract time-domain features, use frequency-domain analysis to extract fundamental frequency-domain features, and use CNN to extract spatial features for statistical features and visual data. S25. Perform feature filtering and weighted fusion on the above multi-dimensional basic features to generate multimodal feature data.

5. The method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis for 3C manufacturing according to claim 1, characterized in that, The S3 includes: S31. Perform time-series segmentation processing on the multimodal feature data, divide the sub-time-series samples according to the single process cycle of 3C products, use multiple methods to perform data augmentation, and generate a time-series feature dataset suitable for model training. S32. Input the time series feature dataset into the self-developed multimodal temporal graph neural network. Capture the long-term and short-term time dependencies of the data through the temporal attention layer in the model. Model the spatial association topology between devices through the graph convolutional layer. Integrate spatiotemporal dual-dimensional features to complete the multi-device collaborative temporal association modeling and generate a preliminary multi-machine temporal association model. S33. The performance of the preliminary multi-machine time series correlation model is verified using test set data, the model's metrics are calculated, the hyperparameters of the model are optimized using a grid search algorithm, and high-precision multi-machine time series correlation model data is generated. S34. For the fault sample subset in multimodal feature data, a causal discovery algorithm based on structural equation model is adopted to construct a causal relationship graph between features and identify potential paths and key intermediate variables of fault propagation. S35. Perform a significance test on the causal relationship graph, eliminate false association paths, and generate causal relationship data for multi-machine collaborative failures.

6. The method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis for 3C manufacturing according to claim 1, characterized in that, The S4 includes: S41. Integrate multi-machine time-series correlation model data and multi-machine collaborative fault causal relationship data to construct a three-layer fault root cause localization inference model of features, correlation and causation, and define the confidence threshold and inference rules for root cause determination. S42. The spatiotemporal coupled sensor data collected in real time from the production line under fault conditions is extracted and input into the inference model. By tracing back to the starting point of the causal relationship path and combining the dynamic prediction results of the time-series correlation model, the key equipment and core influencing factors that cause multi-machine collaborative failure are identified, and preliminary fault root cause localization data is generated. S43. Cross-validate the preliminary root cause localization data by combining equipment historical fault records and process standard documents, correct the root cause judgment deviation, and generate accurate fault root cause localization data. S44. Based on accurate fault root cause location data, analyze the quantitative relationship between key influencing factors and process quality indicators, and establish an objective function for adjusting process parameters. S45. An improved particle swarm optimization algorithm is used to solve the objective function, generate the real-time adjustment range, adjustment step size and adjustment sequence of key process parameters, and obtain process parameter adjustment strategy data.

7. The method for multi-machine collaborative scenarios and spatiotemporal coupled sensing big data analysis for 3C manufacturing according to claim 6, characterized in that, S42 includes: Preprocess the spatiotemporal coupled sensor data collected in real time from the production line under fault conditions, remove data noise, fill in missing values ​​and complete format standardization to generate standardized sensor data under fault conditions. The multimodal feature extraction framework of S2 is reused to perform targeted feature extraction on standardized sensor data of fault conditions, retain core features with high correlation to faults, and generate fault condition-adaptive feature data. Input the fault condition adaptation feature data into the three-layer fault root cause localization reasoning model, activate the causal relationship path tracing module in the model, load the preset causal relationship graph structure, and generate path tracing initialization parameters. Based on the initialization parameters, the reverse tracing algorithm is executed. Starting from the terminal node corresponding to the fault characteristics, the algorithm traces back to the starting point of the path layer by layer along the causal relationship path, filters out potential root cause related nodes, and generates a set of candidate root cause nodes. The dynamic prediction results of the multi-machine time series correlation model are retrieved, the equipment coordination time series features during the fault occurrence period are extracted, and the time series and causal correlation are matched with the candidate root cause node set to calculate the matching confidence. Based on the preset confidence threshold, the matching results are filtered and false root cause nodes with confidence levels below the threshold are eliminated, the key equipment and core influencing factors that cause multi-machine collaborative failure are identified, and preliminary root cause location data is generated.

8. The method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis for 3C manufacturing according to claim 6, characterized in that, The S45 includes: Extract the equipment hardware limitations and process specification requirements of key process parameters, determine the value boundaries of each parameter, and generate a set of parameter constraints; based on the set of parameter constraints, initialize the core parameters of the improved particle swarm optimization algorithm and generate algorithm initialization configuration data; The established dual objective function and parameter constraint set are input into the initialized improved particle swarm optimization algorithm to start the iterative optimization process. The optimal fitness value of each generation of particles is calculated through the particle position and velocity update formula, and the iterative optimization process dataset is generated. Set an iteration convergence judgment threshold. When the fluctuation of the optimal fitness value is less than the threshold for multiple consecutive generations, stop the iteration and extract the global optimal solution. Map the real-time adjustment range of each key process parameter and generate a parameter adjustment range dataset. By combining the real-time operating cycle time of the production line and the response delay characteristics of the equipment, the parameter adjustment range is subdivided, the difference between adjacent adjustment nodes is calculated as the adjustment step size, and parameter adjustment step size data is generated. Based on the multi-machine collaborative timing association rules, the order and time nodes for adjusting each key process parameter are planned to avoid parameter adjustment conflicts and generate a parameter adjustment timing scheme. The parameter adjustment interval dataset, adjustment step size data and adjustment timing scheme are integrated to supplement the execution conditions and safety warning thresholds for parameter adjustment and generate complete process parameter adjustment strategy data.

9. The method for multi-machine collaborative scenarios and spatiotemporally coupled sensing big data analysis for 3C manufacturing according to claim 1, characterized in that, The S5 includes: S51. Convert the process parameter adjustment strategy data into a control command format that can be executed by the equipment, and send it to the corresponding key equipment controller in real time via industrial Ethernet to complete the dynamic adjustment of process parameters; S52. During the parameter adjustment process, the equipment operating parameter data, process data and product quality inspection data are continuously collected through a three-level sensor network to generate a real-time sensor dataset after adjustment. S53. Reuse the standardized processing flow of S1 and the spatiotemporal alignment and feature extraction flow of S2 on the adjusted real-time sensing dataset to generate new multimodal feature data as incremental data for model updates. S54. An online incremental learning algorithm is adopted to integrate new multimodal feature data into the multi-machine time series correlation model and the multi-machine collaborative fault causal relationship model, update the model weight parameters, eliminate model drift, and generate dynamically optimized analysis model data. S55. Based on the dynamic optimization analysis model data, monitor the core indicators of the production line in real time and generate multi-dimensional production line operation status monitoring data; set up a production line operation status evaluation index system, use the analytic hierarchy process to determine the index weights, quantify and score the monitoring data, and generate production line operation status evaluation data. S56. Based on the evaluation data results, if the index score is higher than the qualified threshold, a stable control instruction to maintain the current process parameters is generated; if there is a potential abnormal risk, an early warning control instruction for early intervention is generated; if new fault characteristics appear, a fault handling control instruction for rapid response is generated; through the above process, an intelligent closed loop is formed to obtain intelligent analysis and control instruction data covering the entire scenario.

10. A system for implementing the multi-machine collaborative scenario and spatiotemporally coupled sensing big data analysis method for 3C manufacturing as described in claim 1, characterized in that, The system includes: Node Deployment Module: Constructs a three-level sensor network covering equipment, workstations, and production lines; deploys sensor nodes on equipment in the 3C manufacturing production line; generates sensor data; determines the topological relationships and physical constraints between devices based on the sensor data; and forms a dataset of device relationships and constraints. Feature extraction module: Based on the device relationship and constraint dataset, it performs spatiotemporal alignment processing on multi-device sensor data to generate spatiotemporally coupled sensor data; it then performs multimodal feature extraction on the spatiotemporally coupled sensor data to obtain multimodal feature data. Relationship mining module: Input multimodal feature data into a self-developed multimodal time series graph neural network to model the temporal correlation between multiple devices and generate multi-machine time series correlation model data; at the same time, use causal discovery algorithm to analyze multimodal feature data, mine potential causal relationships of multi-machine collaborative failures, and obtain multi-machine collaborative failure causal relationship data; Strategy generation module: Based on multi-machine time-series correlation model data and multi-machine collaborative fault causal relationship data, it performs root cause localization processing for multi-machine collaborative faults and generates fault root cause localization data; based on the fault root cause localization data, it generates real-time dynamic adjustment strategies for key process parameters and obtains process parameter adjustment strategy data. Update and optimization module: Real-time dynamic adjustment of key process parameters according to process parameter adjustment strategy data, and continuous collection of new sensor data during the adjustment process; Update and optimize multi-machine time sequence correlation model data and multi-machine collaborative fault causal relationship data to form dynamically optimized analysis model data.

Images