Comprehensive processing method for dynamic production scheduling data of manufacturing resources

Abstract

The invention belongs to the technical field of multi-target global dynamic production scheduling, and discloses a manufacturing resource dynamic production scheduling data comprehensive processing method, which comprises the following steps: acquiring multi-source heterogeneous data, constructing a field mapping table, constructing an equipment fingerprint database in combination with historical equipment performance data, dynamically adjusting confidence coefficient weight, and further judging an equipment state. Based on an equipment state judgment result, integrating operator report data and multi-source heterogeneous data through a preset rule base trigger event, and generating an abnormal event flow with a context tag; based on the abnormal event type and the equipment state judgment result, a recommendation scheme is constructed through a production scheduling knowledge graph, and a production scheduling instruction set is generated; according to the production scheduling instruction set, generating a local optimization scheme and a global optimization strategy, and forming a final execution scheme; calculating a resource utilization rate and a bottleneck process index through the 3D virtual workshop model, and generating a difference analysis report; according to the method, global resource scheduling and closed-loop iteration are realized, and the production efficiency and the system robustness are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-objective global dynamic production scheduling. More specifically, the present invention relates to a comprehensive processing method for dynamic production scheduling data of manufacturing resources. Background Art

[0002] With the increasing demand for flexible production and intelligent production scheduling in the manufacturing industry, traditional manufacturing resource scheduling systems mostly rely on fixed rule engines and offline optimization algorithms, such as static scheduling models based on APS (Advanced Planning and Scheduling) and scheduling strategies driven by manual experience. However, when facing the fusion of multi-source heterogeneous data, dynamic fluctuations in equipment status, and sudden production disturbances, these methods generally have problems such as lagging response and rough optimization granularity. In recent years, the introduction of industrial Internet of Things (IIoT) and edge computing technologies has provided new ideas for real-time data collection and processing, but there are still significant shortcomings in multi-modal data collaboration, dynamic weight allocation, and global-local optimization collaboration in the existing technologies.

[0003] The existing technologies mainly focus on the application of single sensors or local optimization strategies, such as relying on fixed load thresholds or static path planning, and rarely integrate multi-source data (such as ERP orders, IoT sensor streams) with intelligent analysis; in addition, the determination of equipment status does not integrate multi-dimensional information (such as ERP order progress, manual feedback), resulting in one-sided results; the production scheduling optimization lacks support from global multi-objective algorithms and is difficult to balance resource utilization and delivery timeliness; the handling of abnormal events depends on fixed rules and fails to generate root cause tags by real-time associating order status, sensor alarms, and manual descriptions; at the same time, the knowledge base and equipment parameters rely on manual maintenance, lacking a closed-loop calibration mechanism based on execution data, and the virtual simulation and actual execution do not achieve dynamic linkage, further restricting the adaptive ability of the scheduling strategy; the collaboration between the edge layer and the cloud layer is insufficient, and there is a disconnection between the global strategy and local execution, resulting in low resource scheduling efficiency. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: A comprehensive processing method for dynamic production scheduling data of manufacturing resources, including:

[0005] S1: Obtain multi-source heterogeneous data, extract key fields, and construct a cross-system field mapping table; generate a device fingerprint vector and a device category benchmark template in combination with historical equipment performance data to construct a device fingerprint library; according to the device fingerprint library, calculate the sensor stability index in real time, dynamically adjust the confidence weight, and then determine the device status;

[0006] S2: Based on the device status determination result, align the time stamps of multi-source heterogeneous data, and trigger events through a preset rule library; integrate the data reported by operators and multi-source heterogeneous data to generate an abnormal event stream with context tags;

[0007] S3: Based on the abnormal event type and the equipment status determination result, call the preset micro-process, construct a recommended solution through the production scheduling knowledge graph, and then generate a production scheduling instruction set;

[0008] S4: According to the production scheduling instruction set, drive the equipment to execute the instructions through the edge agent to generate a local optimization solution locally; the cloud integrates the data of multiple factories, generates a global optimization strategy through the black-box multi-objective algorithm, and then forms a final execution plan; compare the execution results to generate a closed-loop analysis data set, and update the production scheduling knowledge graph and the preset rule base in reverse;

[0009] S5: Load the final execution plan into the 3D virtual workshop model, calculate the resource utilization rate and the bottleneck process index; then compare the final execution plan with the actual execution data, generate a difference analysis report, calibrate the equipment fingerprint library and the preset rule base in reverse, trigger the re-determination of the equipment status, and form an end-to-end closed-loop iteration.

[0010] Further, the generation method of the field mapping table includes:

[0011] The multi-source heterogeneous data includes ERP order data, MES equipment status data, IoT sensor stream data, and supply chain inventory data. The ERP order data includes order basic information, product details, status information, and association information. The MES equipment status data includes real-time operating status, performance indicators, theoretical production capacity, maintenance records, production progress, and alternative equipment lists. The IoT sensor stream data includes equipment health data, environmental data, event alarms, location and status. The supply chain inventory data includes inventory level, inventory changes, supply chain collaboration, and demand forecast;

[0012] Use natural language processing technology to parse the unstructured data in the multi-source heterogeneous data, extract the keyword fields; then establish a cross-system field mapping relationship based on the keyword fields to generate a field mapping table.

[0013] Further, the method for dynamically adjusting the confidence weight includes:

[0014] Obtain historical equipment performance data, including equipment operation efficiency indicators, sensor reference parameters, environmental and load data;

[0015] Extract features from the historical equipment performance data through the feature extraction method to obtain the hardware static features and dynamic behavior features of the equipment;

[0016] Use the importance of different fields in the field mapping table as the initial weights of the hardware static features and dynamic behavior features to construct an initial weight matrix, and update the weight matrix in real time according to the information entropy of the features through the entropy weight method;

[0017] Based on the weight matrix, the static hardware features and dynamic behavior features are weighted, and then the principal component analysis technique is used to reduce the dimension of all weighted features to form a device fingerprint vector;

[0018] Meanwhile, the Gaussian mixture model is used to perform unsupervised clustering on the features after dimensionality reduction to generate a device category benchmark template;

[0019] Through the federated learning framework, the device fingerprint vector and the device category benchmark template are distributedly and collaboratively optimized to obtain the optimized device fingerprint vector and device category benchmark template;

[0020] The optimized device fingerprint vector and device category benchmark template are integrated to obtain a device fingerprint database;

[0021] Based on different sensors of the device, the real-time data of the sensors are read, and the difference calculation is performed with the device category benchmark template of the corresponding category in the device fingerprint database to obtain the error value of the sensor relative to the benchmark;

[0022] The average value of the error values of the sensor in the past T moments is taken as the historical standard deviation, and then the ratio of the current error value to the historical standard deviation is taken as the error ratio; and the reciprocal of the sum of 1 and the error ratio is taken as the stability index of the sensor;

[0023] Set the error threshold and stability threshold, and based on the stability index and the preset error threshold, define the dynamic adjustment rule of the confidence weight of the sensor, including:

[0024] If the error value is less than or equal to the error threshold and the stability index is greater than or equal to the stability threshold, it is determined that the data collected by the sensor is stable;

[0025] If the error value is greater than the error threshold and the stability index is less than the stability threshold, it is determined that there are potential errors in the data collected by the sensor, and the confidence weight of the sensor data is reduced according to the preset adjustment ratio;

[0026] If the error value of the sensor is greater than the error threshold for N consecutive moments, it is determined that the sensor is abnormal, and the confidence weight of the sensor is reduced to the lowest according to the adjustment ratio;

[0027] According to the adjusted confidence weight of the sensor data, if the confidence weight is greater than the preset confidence weight threshold, it is marked as high-confidence sensor data; if the confidence weight is less than or equal to the preset confidence weight threshold, it is marked as low-confidence sensor data.

[0028] Furthermore, the determination method of the device state includes:

[0029] Map the real-time data of the sensor and multi-source heterogeneous data into standardized fields according to the field mapping table, and classify them into sensor data, manual data, and ERP order data according to the data type;

[0030] Use the confidence weight of the sensor data as the weight of the sensor data, and assign fixed weights to the manual data and ERP order data;

[0031] Perform weighted summation on all standardized fields of the sensor data, manual data, and ERP order data respectively to obtain the sensor comprehensive value, manual comprehensive value, and ERP comprehensive value; and then horizontally splice them to form a comprehensive state vector;

[0032] According to the comprehensive state vector, define the sensor comprehensive value threshold, ERP comprehensive value threshold, and manual comprehensive value threshold respectively, and define the device status determination rule as:

[0033] If the sensor comprehensive value is greater than or equal to the sensor comprehensive value threshold, and the ERP comprehensive value is greater than or equal to the ERP comprehensive value threshold, and the manual comprehensive value is greater than or equal to the manual comprehensive value threshold, then the device status is determined to be normal operation;

[0034] If any one of the sensor comprehensive value, manual comprehensive value, and ERP comprehensive value is less than the corresponding threshold, then the device status is determined to be partially faulty;

[0035] If the sensor comprehensive value is lower than the sensor comprehensive value threshold for N consecutive moments, and the ERP comprehensive value is less than the ERP comprehensive value threshold, and the manual comprehensive value is less than the manual comprehensive value threshold, then the device status is determined to be shutdown for maintenance.

[0036] Furthermore, the generation method of the abnormal event stream includes:

[0037] Based on the device status determination result, use the determined device status as the status label, align it with the timestamp of the multi-source heterogeneous data to the same time axis, and then add the current device status determination result as a label to the multi-source heterogeneous data at each moment;

[0038] Build a preset rule library, monitor the device status determination result and multi-source heterogeneous data in real time, and generate a trigger event when the conditions in the preset rule library are met;

[0039] Based on the generated trigger event, use the order status, MES process progress, and IoT sensor anomalies in the ERP order data as multi-source data labels, and then combine them with the description reported by the operator in real time and the corresponding device status determination result to form a root cause label as the context label;

[0040] Integrate the context label, the type, priority, and original data source of the trigger event with the corresponding timestamp into an abnormal event stream.

[0041] Furthermore, the construction method of the preset rule library includes:

[0042] Based on industry standards and historical equipment performance data, using the equipment status determination result as the basis for dynamic rule adjustment, extract the threshold rule basic parameters from the multi-source heterogeneous data after aligning industry standards, historical equipment performance data, and timestamps, including static rule parameters and dynamic rule parameters;

[0043] Define the rules of the preset rule library to include static rules and dynamic rules. Define static rules according to static rule parameters and dynamic rules according to dynamic rule parameters;

[0044] Static rules represent fixed thresholds preset based on industry standards and historical equipment performance data, including process timeout threshold, order delay threshold, and inventory safety threshold; Dynamic rules represent defining conditions and adjustment logics according to the equipment status determination result, including dynamic adjustment of equipment OEE, sensor alarm linkage, and inventory linkage.

[0045] Furthermore, the generation method of the production scheduling instruction set includes:

[0046] According to the equipment status determination result and the abnormal event stream, match the event types in the abnormal event stream with the preset micro-process library to generate a preliminary adjustment plan;

[0047] According to the event type of the equipment, obtain the average repair time of similar faults, the performance records of alternative equipment, and the associated solutions;

[0048] Extract the equipment, orders, processes, and alternative equipment in the multi-source heterogeneous data after timestamp alignment as entities. According to the preliminary adjustment plan, extract the relationships between each entity in the abnormal event stream and the preset rule library, and add the average repair time of similar faults, the performance records of alternative equipment, and the associated solutions as attributes into the relationships, and then construct a production scheduling knowledge graph;

[0049] Use the inference engine to query the historical repair time of similar faults and the performance records of alternative equipment through the production scheduling knowledge graph according to the event type and context tags, and construct a recommended plan; Generate a production scheduling instruction set based on the recommended plan.

[0050] Furthermore, the formation method of the final execution plan includes:

[0051] Synchronize the production scheduling instruction set to the workshop-level edge agent, drive the equipment control layer and the production management system to execute the production scheduling instructions. During the instruction execution process, real-time collect the real-time operation status and production progress of the equipment through IoT sensors to form an execution feedback stream; and immediately generate a local optimization plan by the edge agent based on the equipment health status and the list of alternative equipment;

[0052] After the instruction execution is completed, the final execution result of the instruction is obtained by summarizing the execution feedback stream.

[0053] Compare the execution result with the conditions in the preset rule library, the device status determination result, the device status data, and the order data to identify the execution deviation value.

[0054] Collect the execution feedback stream and the abnormal event stream of multiple factories through the cloud server to form a global resource view.

[0055] Obtain the global optimization strategy through the black box multi-objective algorithm based on the global resource view and the historical deviation value of the device.

[0056] Integrate the global optimization strategy of the cloud server with the local optimization plan of the edge agent to form the final execution plan.

[0057] Associate the deviation value with the original scheduling instruction in the scheduling instruction set and the context label of the abnormal event stream to form a closed-loop analysis data set; furthermore, update the scheduling knowledge graph and the preset rule library based on the closed-loop analysis data set.

[0058] Furthermore, the resource utilization rate represents the measurement of the usage efficiency of device resources and material resources, including device utilization rate, material turnover rate, and labor utilization rate.

[0059] The bottleneck process index represents identifying the key bottleneck links in the production process, including the critical path delay time and the throughput of the bottleneck device.

[0060] Load the final execution plan into the 3D virtual workshop model, and obtain the longest dependency path of the order according to the simulation result of the 3D virtual workshop model through the Gantt chart. Take the sum of the process delays on the longest dependency path of the order as the critical path delay time.

[0061] Set the standard production capacity threshold of the theoretical production capacity of the device. If the actual production capacity of the device is less than the standard production capacity threshold, the device is determined to be a bottleneck device, and the ratio of the actual production capacity of the bottleneck device to the theoretical production capacity is taken as the throughput of the bottleneck device.

[0062] Furthermore, the generation method of the difference analysis report includes:

[0063] According to the final execution plan, execute the production scheduling plan, monitor the actual execution data in real time, compare the final execution plan with the actual execution data, and generate a difference analysis report, including the difference analysis of resource utilization rate and the difference analysis of bottleneck process index.

[0064] The technical effects and advantages of a comprehensive processing method for manufacturing resource dynamic scheduling data according to the present invention:

[0065] Through the field mapping table and the dynamic confidence weight mechanism, the present invention integrates multi-source heterogeneous data and improves the reliability of sensor data. Secondly, the multi-dimensional weighted analysis method is used to comprehensively determine the device state based on multi-source data, significantly improving the determination accuracy. Then, based on the edge-cloud collaborative architecture, a global optimization strategy is generated through the black-box multi-objective algorithm to achieve dynamic coordination of cross-factory resources. Next, multi-source data is integrated to generate an abnormal event stream with root cause tags, shortening the fault response time. Further, the knowledge base and device parameters are closed-loop calibrated through the difference analysis report, replacing manual maintenance. At the same time, the 3D virtual workshop is linked with the actual execution data to dynamically optimize the production scheduling strategy. Finally, the real-time parameter synchronization between the edge and the cloud ensures the coordination of global and local executions. In summary, this solution breaks through the bottleneck of traditional systems, realizes global resource scheduling, dynamic adaptability, and closed-loop iteration, significantly improving production efficiency and system robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a schematic diagram of a comprehensive processing method for dynamic production scheduling data of manufacturing resources according to the present invention;

[0067] Figure 2 It is a schematic diagram of the edge-cloud collaborative optimization process of a comprehensive processing method for dynamic production scheduling data of manufacturing resources according to the present invention;

[0068] Figure 3 It is a schematic diagram of a comprehensive processing system for dynamic production scheduling data of manufacturing resources according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] Embodiment 1

[0071] Please refer to Figure 1 and Figure 2 As shown, a comprehensive processing method for dynamic production scheduling data of manufacturing resources in this embodiment includes:

[0072] S1: Obtain multi-source heterogeneous data, extract key fields, and construct a cross-system field mapping table; generate a device fingerprint vector and a device category benchmark template in combination with historical device performance data to construct a device fingerprint library; calculate the sensor stability index in real time according to the device fingerprint library, dynamically adjust the confidence weight, and then determine the device state;

[0073] S2: Based on the device status determination result, align the timestamps of multi-source heterogeneous data, and trigger events through a pre-set rule library; integrate the data reported by operators and multi-source heterogeneous data to generate an abnormal event stream with context tags;

[0074] S3: Based on the abnormal event type and the device status determination result, call a pre-set micro-process, construct a recommended solution through a production scheduling knowledge graph, and then generate a production scheduling instruction set;

[0075] S4: According to the production scheduling instruction set, drive the device to execute the instructions through an edge agent to generate a local optimization solution locally; the cloud integrates the data of multiple factories, generates a global optimization strategy through a black-box multi-objective algorithm, and then forms a final execution plan; compare the execution results to generate a closed-loop analysis data set, and update the production scheduling knowledge graph and the pre-set rule library in reverse;

[0076] S5: Load the final execution plan into the 3D virtual workshop model, calculate the resource utilization rate and bottleneck process indicators; then compare the final execution plan with the actual execution data, generate a difference analysis report, calibrate the device fingerprint library and the pre-set rule library in reverse, trigger the re-determination of the device status, and form an end-to-end closed-loop iteration;

[0077] Multi-source heterogeneous data includes ERP order data, MES device status data, IoT sensor stream data, and supply chain inventory data. ERP order data (focusing on the full life cycle management of orders, accessed through REST API or direct database connection) includes order basic information (order number, customer / supplier information, order placement time, delivery date, amount, priority), product details (material code, quantity, specification, unit price, delivery requirements), status information (order status (such as to be produced, shipped, completed), approval process, exception flag), and association information (association relationships with production plans, procurement applications, and inventory). MES device status data (reflecting the real-time operation and efficiency of on-site production equipment, transmitted through the OPC UA protocol) includes real-time operation status (whether the device is running / shutdown, current task, fault code), performance indicators (production speed, yield rate, energy consumption, OEE (Overall Equipment Effectiveness)), theoretical production capacity, maintenance records (latest maintenance time, repair history, spare part usage), production progress (current process progress, remaining time, quantity produced), and alternative equipment list (i.e., the MES device status data of the equipment used to replace the original equipment to continue production when the equipment in production has problems). IoT sensor stream data (providing real-time perception of the physical layer of devices and the environment, real-time stream data from each sensor, transmitted through the MQTT message queue) includes device health data (such as real-time physical parameters like temperature, vibration, current, pressure), environmental data (workshop temperature and humidity, air quality, noise level), event alarms (sensor anomalies (such as temperature exceeding the standard), real-time events triggered by device shutdown), location and status (device location, material conveyor belt status, robot movement trajectory). Supply chain inventory data (monitoring material flow and supply-demand balance, synchronized through database interfaces or files (CSV / Excel)) includes inventory levels (current inventory quantity, safety inventory threshold, warehouse location), inventory changes (inbound / outbound records, transfer history, expiration warnings), supply chain collaboration (supplier arrival time, in-transit inventory, shortage risk prediction), and demand forecasting (inventory demand analysis based on historical orders, replenishment suggestions);

[0078] Use natural language processing technology (NLP) to parse unstructured data in multi-source heterogeneous data (such as text data in order basic information and product details, such as automatically extracting the work order numbers of ERP and device IDs of MES through regular expressions or natural language processing technology), and extract key fields (for example, identifying "work order number: 2025XXXX" from the work order description. The key fields can be automatically identified according to different actual situations through field automatic identification technology, and different key fields can be distinguished);

[0079] Furthermore, establish cross-system field mapping relationships based on key fields (which can be achieved through predefined rules or machine learning techniques such as pattern matching), and generate a field mapping table (such as an Excel or database table, like the ERP work order number). Process Sensor group, ERP "Order Priority" Scheduling system "Urgency label", MES "Equipment model" "Sensor number"), and this data mapping table supports dynamic updates and conflict resolution;

[0080] Obtain historical equipment performance data, including equipment operation efficiency indicators (Overall Equipment Effectiveness (OEE), which reflects the availability, performance rate, and good product rate of the equipment and is used to evaluate the equipment health status; Mean Time Between Failures (MTBF), which records the duration of continuous failure-free operation of the equipment and is used to predict the maintenance cycle, obtained from the MES system or equipment logs), sensor reference parameters (spindle speed error, temperature / vibration threshold, long-term recorded and stored in the PostgreSQL time series database by ltT sensors), and environmental and load data (environmental temperature and humidity, equipment load rate (such as CPU usage, memory occupancy), periodically collected by the SCADA system);

[0081] Extract features from the historical equipment performance data through feature extraction methods (such as statistical methods (calculating mean, variance, quantiles), dimensionality reduction techniques (reducing data dimensions through Orthogonal Projection Canonical Correlation Analysis (ORPCA)), or machine learning methods (such as using Convolutional Neural Networks (CNN) to automatically extract the time series features of sensor data)) to obtain the static hardware features of the equipment (including immutable attributes such as MAC address, CPU model, memory capacity, etc. For example, obtain the hash value of the MAC address by parsing the equipment log or system API, and generate a unique hardware identifier in combination with the equipment model database (such as Intel ARK)) and dynamic behavior features (extracted through time series data analysis, such as communication protocol periodicity (Modbus response time), network traffic pattern (packet interval statistics), sensor signal stability (variance of temperature / vibration sensors), etc. For example, use FFT to analyze the spindle speed spectrum energy distribution to identify the equipment degradation trend);

[0082] Assign weights to different features to reflect their importance for equipment status assessment:

[0083] Take the field importance recorded in the field mapping table as the initial weights of the hardware static features and dynamic behavior features, and construct an initial weight matrix (allocate initial weights according to the sensitivity and stability of the features to the device state. For example, the weight of the hardware static feature is 0.6 (high stability), and the weight of the dynamic behavior feature is 0.4 (high sensitivity)), and update the weight matrix in real time according to the information entropy of the features by the entropy weight method (for example, if the variance of a certain sensor data increases significantly (abnormal fluctuation), then reduce its confidence weight and increase the weights of other stable sensors);

[0084] Based on the weight matrix, weight the hardware static features and dynamic behavior features, and then use the principal component analysis technology to perform dimensionality reduction processing on all weighted features to form a device fingerprint vector;

[0085] At the same time, use the Gaussian mixture model (GMM) to perform unsupervised clustering on the dimensionality-reduced features to obtain the clustering centers, and use them as the benchmark templates for the same type of devices to generate device category benchmark templates; for example, input the historical performance data (OEE, MTBF) of the same type of devices, and generate the clustering centers through iterative EM algorithm;

[0086] Through the federated learning framework, perform distributed collaborative optimization on the device fingerprint vector and the device category benchmark template, and then obtain the optimized device fingerprint vector and device category benchmark template;

[0087] Integrate the optimized device fingerprint vector and the device category benchmark template to obtain a device fingerprint library;

[0088] Based on different sensors of the device, read the real-time data of the sensors, and perform difference calculation with the device category benchmark template of the corresponding category in the device fingerprint library to obtain the error value of the sensor relative to the benchmark;

[0089] Take the average value of the error values of the sensor in the past T moments as the historical standard deviation, and then take the ratio of the current error value to the historical standard deviation as the error ratio; and take the reciprocal of the sum of 1 and the error ratio as the stability index of the sensor;

[0090]

[0091] Among them, when the error value is much larger than the historical standard deviation: the ratio increases → the denominator approaches infinity → the stability index approaches 0, indicating that the stability of the sensor data is extremely poor;

[0092] When the error value is close to 0: the ratio is close to 0 → the denominator is 1 → the stability index is equal to 0, indicating that the stability of the sensor data is extremely high;

[0093] It should be noted that in the stability index formula, the 1 in the denominator can ensure that the formula is meaningful when the error is 0, avoid division-by-zero errors, and limit the stability index within the range of [0,1]. It can also increase balance and robustness. Through the stability index, it can be quickly determined whether the sensor data is abnormal due to environmental changes, hardware failures, or drifts. Combined with the subsequent dynamic calibration rules, the confidence weight of unstable sensors can be reduced without manual intervention.

[0094] (Example: If the current error of a temperature sensor is 5% and its historical standard deviation is 2%, then the stability index is 0.286, indicating poor data stability)

[0095] Set the error threshold and stability threshold (set by relevant personnel in the industry based on experience). Based on the stability index and the preset error threshold, define the dynamic adjustment rules for the confidence weight of the sensor, including:

[0096] If the error value is less than or equal to the error threshold and the stability index is greater than or equal to the stability threshold, it is determined that the data collected by the sensor is stable.

[0097] If the error value is greater than the error threshold and the stability index is less than the stability threshold, it is determined that there are potential errors in the data collected by the sensor, and the confidence weight of the sensor data is reduced according to the preset adjustment ratio.

[0098] If the error value of the sensor is greater than the error threshold for N consecutive moments, it is determined that the sensor is abnormal, and the confidence weight of the sensor is reduced to the lowest according to the adjustment ratio.

[0099] According to the adjusted confidence weight of the sensor data, if the confidence weight is greater than the preset confidence weight threshold, it is marked as high-confidence sensor data; if the confidence weight is less than or equal to the preset confidence weight threshold, it is marked as low-confidence sensor data.

[0100] Among them, the error threshold, stability threshold, and confidence weight threshold can be dynamically updated through the 3σ principle, that is, the threshold = reference value ± 3 × historical standard deviation; the historical standard deviation can be calculated from the error values, stability indices, and confidence weights in the past T moments.

[0101] Map the real-time data and multi-source heterogeneous data of the sensor to standardized fields according to the field mapping table, and classify them into sensor data, manual data, and ERP order data according to the data type.

[0102] Use the confidence weight of sensor data as the weight of sensor data, and assign fixed weights to manual data and ERP order data (e.g., manual data category 0.5 (since manual data is filled in manually and can be of medium reliability), ERP order data category 0.7 (since ERP data is structured data with high real-time performance and high reliability));

[0103] Perform weighted summation on all standardized fields of sensor data, manual data, and ERP order data respectively to obtain the sensor comprehensive value, manual comprehensive value, and ERP comprehensive value; then horizontally splice them to form a comprehensive status vector;

[0104] According to the comprehensive status vector, define the sensor comprehensive value threshold, ERP comprehensive value threshold, and manual comprehensive value threshold respectively, and define the device status determination rule as:

[0105] If the sensor comprehensive value is greater than or equal to the sensor comprehensive value threshold, and the ERP comprehensive value is greater than or equal to the ERP comprehensive value threshold, and the manual comprehensive value is greater than or equal to the manual comprehensive value threshold, then the device status is determined to be normal operation;

[0106] If any one of the sensor comprehensive value, manual comprehensive value, and ERP comprehensive value is less than the corresponding threshold, then the device status is determined to be partially faulty;

[0107] If the sensor comprehensive value is lower than the sensor comprehensive value threshold for N consecutive moments, and the ERP comprehensive value is less than the ERP comprehensive value threshold, and the manual comprehensive value is less than the manual comprehensive value threshold, then the device status is determined to be shutdown for maintenance;

[0108] Based on the device status determination result, use the determined device status as the status label, align it with the timestamp of the multi-source heterogeneous data to the same time axis, and then add the current device status determination result as a label to the multi-source heterogeneous data at each moment;

[0109] Build a preset rule library, monitor the device status determination result and multi-source heterogeneous data in real time, and generate a trigger event when the conditions in the preset rule library are met;

[0110] Exemplary trigger conditions:

[0111] Static rule: Order completion rate < 80% (ERP data) → Trigger the "order delay" event;

[0112] Dynamic rule: Equipment OEE < 80% (MES data) → Adjust the threshold and trigger the "equipment performance anomaly" event;

[0113] Based on the generated trigger events, the order status in the ERP order data (such as "awaiting production"), the MES process progress (such as "exceeding the remaining time"), and the IoT sensor anomalies (such as "exceeding the temperature") are used as multi-source data tags (for example, order C is delayed → the tag includes "process B progress is delayed", "equipment A is overloaded"), and then combined with the data reported by the operator in real time and the corresponding equipment status determination results to form root cause tags, which are used as context tags (for example, the operator reports "equipment A is jammed" → the tag adds "equipment A operation is abnormal");

[0114] An exemplary priority adjustment logic:

[0115] Dynamically adjust the logic according to the rule library: such as equipment status impact, sensor alarm impact, and conflict between manual marking and rules;

[0116] Such as equipment status impact:

[0117] If the equipment status is "partially faulty" or "shutdown", the priority of the associated event is automatically raised to "high";

[0118] Example: The "progress deviation" event reported by the operator (original priority "medium") has its priority raised to "high" due to the equipment status being "partially faulty";

[0119] Sensor alarm impact:

[0120] If the IoT sensor detects that the temperature / vibration exceeds the standard → the event priority is raised to "high";

[0121] When there is a conflict between manual marking and rules:

[0122] Give priority to the priority dynamically adjusted by the rule library (for example, when the rule library determines it as "high", it overrides the operator's "medium" level marking);

[0123] Integrate the context tags, the type of trigger event, the priority, and the original data source of the trigger event with the corresponding timestamp into an abnormal event stream;

[0124] According to industry standards (such as equipment safety operation specifications, inventory safety thresholds) and historical equipment performance data (such as MTBF, sensor reference parameters), use the equipment status determination results as the basis for dynamically adjusting rules, and extract the threshold rule basic parameters from the multi-source heterogeneous data after aligning the industry standards, historical equipment performance data, and timestamp, including static rule parameters and dynamic rule parameters;

[0125] Exemplarily, device safety operation specifications (such as device OEE ≥ 85% being normal) and inventory safety thresholds (such as the inventory level needs to be maintained above 60% of the safety inventory) can be extracted from industry standards, and the historical average time between failures of the device, sensor reference parameters (such as the spindle vibration threshold ≤ 120 Hz), and the historical average value of the overall equipment effectiveness can be extracted from historical device performance data;

[0126] Then, based on these extracted data, static rule parameters are constructed, such as the theoretical standard duration of the process (calculated based on historical production capacity data in the MES system), temperature safety threshold (such as "spindle temperature ≤ 45°C" being normal in the historical sensor data);

[0127] Dynamic rule parameters can be extracted from multi-source heterogeneous data. For example, the current OEE value, real-time temperature value of IoT sensors, and the proportion of safety inventory in inventory (from the supply chain system) are extracted from MES data;

[0128] The threshold rules are defined as static rules and dynamic rules. Static rules represent fixed thresholds preset based on industry standards and historical device performance data, including the process timeout threshold (statistical standard duration of a certain process from historical MES data (such as the average time-consuming 2 hours), combined with safety redundancy (such as 120%) → static threshold: timeout threshold = 2 × 120% = 2.4 hours), order delay threshold (such as stipulating that "when the order completion rate is lower than 80%, a delay warning is triggered" → static threshold: order delay threshold = 80%), and inventory safety threshold (such as calculating 30% of the safety inventory as the out-of-stock risk threshold based on historical inventory data → static threshold: inventory safety threshold = current inventory < safety inventory × 30%). Dynamic rules represent the definition of conditions and adjustment logic based on the device status determination result, including device OEE dynamic adjustment, sensor alarm linkage, and inventory linkage; The device status determination result is corresponded one by one with the dynamic rule parameters. For example, partial failure → process timeout threshold, order delay threshold → lower threshold (stricter), shutdown → inventory safety threshold, order delay threshold → higher threshold (looser), sensor alarm → event priority → elevated to "high";

[0129] Exemplary dynamic rule setting:

[0130] Device OEE dynamic adjustment: The condition is that the device status is determined to be partially faulty, and the action is to adjust the process timeout threshold from 120% to 110%. The rule logic is that if the device status = partially faulty, then the timeout threshold = standard duration × 1.1; if the device status ≠ partially faulty, then the timeout threshold = standard duration × 1.2;

[0131] Inventory linkage: The condition is that the supply chain inventory safety threshold < 30%, and the action is to increase the order delay threshold from 80% to 90%. The rule logic is that if the inventory safety ratio < 30%, then the order delay threshold = 90%; otherwise, the order delay threshold = 80%.

[0132] Sensor alarm linkage: The condition is that the IoT sensor detects that the temperature exceeds the standard (such as > 45°C), and the action is to increase the priority of the associated event from "medium" to "high". The rule logic is that if the sensor temperature > 45°C, then the event priority = "high".

[0133] According to the device status determination result and the abnormal event flow, match the event types in the abnormal event flow with the preset micro - process library to generate a preliminary adjustment plan.

[0134] It should be noted that the micro - process matches the preset rules according to the event type. For example, for the "equipment shutdown" event, it triggers the "equipment switching process", and generates a process transfer plan by querying available resources (such as the load rate of alternative equipment and inventory status) to match the equipment with a load rate ≤ 70%; for the "order delay" event, it triggers the "urgent order insertion process", automatically raises the order priority and allocates resources.

[0135] According to the event type of the device, obtain the average repair time of similar faults, the performance records of alternative equipment, and the associated solutions.

[0136] Extract the device (status, load rate), order (priority, deadline), process (dependency relationship), and alternative equipment (such as the list of standby equipment) in the multi - source heterogeneous data after timestamp alignment as entities. According to the preliminary adjustment plan, extract the relationships between each entity in the abnormal event flow and the preset rule library (such as, equipment failure → order delay (causal relationship), equipment A → compatible process B (ability association), alternative equipment C → load rate ≤ 70% (constraint condition)), and add the average repair time of similar faults, the performance records of alternative equipment, and the associated solutions as attributes into the relationships, and then construct a production scheduling knowledge graph.

[0137] Use the inference engine to query the historical repair time of similar faults and the performance records of alternative equipment through the production scheduling knowledge graph according to the event type and context tags, and construct a recommended plan; generate a production scheduling instruction set based on the recommended plan.

[0138] The knowledge graph filters historical data based on event tags, such as obtaining the average repair time of similar faults and the maintenance records of alternative equipment, so as to recommend the optimal plan (such as "equipment B undertakes the process" or "adjust the order delivery time") that meets the resource availability (such as the load rate of alternative equipment ≤ 70%) and the historical repair success rate.

[0139] Then, the rule engine is used to verify whether the recommended solution meets the constraint conditions. The rule engine checks parameters such as the load rate of alternative devices, repair time, order deadline, and occupancy status of target devices according to preset rules (such as device load balancing degree ≥ 70%, repair completed within the time window, and no resource conflicts). If the conditions are met, the solution passes; if there are conflicts (such as device load exceeding the limit), a micro - process is triggered to regenerate the solution. Furthermore, a production scheduling instruction set that meets the constraints is finally generated (such as "Device B undertakes Process B", "Order C's priority is increased"). The verified instructions are integrated with the context tags (such as "Device A fails"), priority (such as "high"), and timestamp of the abnormal event stream to form an executable production scheduling instruction set, which includes specific operation instructions, effective time, context tags, and resource allocation details.

[0140] The production scheduling instruction set is synchronized to the workshop - level edge agent (such as local PLC, edge server) to drive the device control layer (such as PLC, SCADA system) and production management system (such as ERP, MES) to execute the production scheduling instructions. During the instruction execution process, the real - time operating status and production progress of the devices are collected in real - time through IoT sensors to form an execution feedback stream; and based on the device health status and alternative device list, the edge agent immediately generates a local optimization solution.

[0141] Exemplarily, if Device B suddenly fails during execution, the edge agent automatically invokes the "emergency switch process" to match the standby Device C with a load rate ≤ 70%.

[0142] After the instruction execution is completed, the final execution result of the instruction is obtained by summarizing the execution feedback stream.

[0143] The execution result is compared with the conditions in the preset rule library, the device status determination result, device status data, and order data to identify the execution deviation value.

[0144] It should be noted that comparing the execution result is actually comparing the execution result with the expected target. For example, the device load rate threshold in the preset rule library (such as "≤ 70%"), order deadline, resource allocation priority. Example: The rule library stores that "when the device load rate ≥ 70%, new tasks are prohibited from being allocated", and this rule is a quantitative expression of the expected target; the device status determination result (such as "normal", "fault") also implies the expected resource allocation target (such as "faulty devices need to be switched to alternative devices"); device status data: includes the load rate threshold when the device is operating normally (such as "design load rate ≤ 80%"); order data: includes preset targets such as the order deadline and priority.

[0145] The deviation value is calculated by comparing the actual execution data (real-time data from the execution feedback stream, such as the actual device load rate of 72% and an order delay of 1 hour) with the expected target data (such as the expected load rate ≤ 70% and the order deadline of 16:00). As for the calculation process, for numerical data, the difference between the actual value and the target value is directly calculated (such as a 2% overload of the load rate and an order delay of 1 hour). For logical data, it is judged whether it meets the constraint conditions according to the rule engine (such as "load rate overload → trigger deviation");

[0146] Collect the execution feedback stream and abnormal event stream of multiple factories through the cloud server to form a global resource view;

[0147] Through black-box multi-objective algorithms (such as genetic algorithms and reinforcement learning), based on the global resource view and the historical deviation values of the devices, obtain global optimization strategies (such as "transfer order C to factory B" and "adjust the maintenance plan of device D");

[0148] Integrate the global optimization strategy of the cloud server with the local optimization plan of the edge agent to form the final execution plan (such as a local device B failure triggers the edge agent to switch to device C (local optimization), and the cloud coordinates the transfer of some orders to factory B according to the production capacity of multiple factories (global optimization));

[0149] Associate the deviation value with the original scheduling instructions in the scheduling instruction set and the context tags of the abnormal event stream to form a closed-loop analysis data set; furthermore, update the scheduling knowledge graph and the preset rule library based on the closed-loop analysis data set;

[0150] Resource utilization rate represents the measurement of the usage efficiency of device resources and material resources; it includes device utilization rate, material turnover rate, and labor utilization rate;

[0151] The calculation method of the resource utilization rate:

[0152] Obtain the device operation duration, order completion quantity, and material consumption through IoT sensors and the MES system;

[0153] Then obtain it through the device utilization rate calculation formula and the material turnover rate calculation formula;

[0154] Exemplarily, the utilization rate of device A = (actual operation time of device A / (24 hours × number of shifts - fault downtime)) × 100%;

[0155] Material turnover rate = (total amount of materials consumed during a certain period / average inventory);

[0156] The bottleneck process index represents identifying the key bottleneck links in the production process, including the critical path delay time (total delivery delay caused by bottlenecks) and the throughput of bottleneck devices (the ratio of the actual production capacity to the theoretical production capacity of bottleneck devices);

[0157] Load the final implementation plan into the 3D virtual workshop model. According to the simulation results of the 3D virtual workshop model through the Gantt chart (or topological sorting), obtain the longest dependency path of the order, and take the total process delay on the longest dependency path of the order as the critical path delay time;

[0158] Set the standard production capacity threshold of the equipment theoretical production capacity (for example, set the standard production capacity threshold to 85% of the theoretical production capacity). If the actual production capacity of the equipment is less than the standard production capacity threshold (i.e., less than 85% of the theoretical production capacity), then determine the equipment as a bottleneck equipment, and take the ratio of the actual production capacity to the theoretical production capacity of the bottleneck equipment as the bottleneck equipment throughput;

[0159] According to the final implementation plan, execute the production scheduling plan, monitor the actual execution data in real time, compare the final implementation plan with the actual execution data, and generate a difference analysis report, including the difference analysis of resource utilization rate and the difference analysis of bottleneck process indicators;

[0160] Exemplarily, the difference analysis of resource utilization rate is as follows:

[0161] Equipment utilization rate deviation:

[0162] Planned value: Equipment A utilization rate 70% → Actual value: 85% (exceeding the limit by 15%);

[0163] Trace through the context label: Associated with the event label of "transfer of process to Equipment A due to Equipment B failure", triggering overload;

[0164] Material turnover rate deviation:

[0165] Planned value: 3.5 times / month → Actual value: 2.8 times / month (decrease by 20%).

[0166] Trace through the context label: Associated with the "material shortage event" label, resulting in the idle rate of Equipment C rising to 30%;

[0167] The difference analysis of bottleneck process indicators is as follows:

[0168] Critical path delay time:

[0169] Planned value: The critical path delay of Order C ≤ 2 hours → Actual value: Delay of 4.5 hours;

[0170] Trace through the context label: The actual production capacity of bottleneck equipment D is 70% of the theoretical production capacity (bottleneck equipment throughput 0.7), lower than the threshold n% = 80%, resulting in the accumulation of critical path process delays;

[0171] Bottleneck equipment throughput:

[0172] Actual production capacity of equipment D: 70% of the theoretical production capacity → marked as a bottleneck equipment, throughput ratio 0.7;

[0173] Based on the results of the difference analysis, reverse calibrate the weights of the preset rule library and the parameters of the equipment fingerprint library;

[0174] In an exemplary manner, based on the results of the difference analysis, dynamically adjust the system parameters through the following mechanism:

[0175] Adjust the weights of the preset rule library to improve the priority of equipment load balancing and adjust the weights for material shortage response;

[0176] Improve the priority of equipment load balancing:

[0177] If the number of times the equipment utilization rate exceeds the limit exceeds the threshold (e.g., equipment A exceeds the limit ≥ 3 times / week), then increase the weight of "load balancing priority": original weight: 0.6 → new weight: 0.8 (updated through the rule engine API);

[0178] Execution effect: In subsequent production scheduling, preferentially allocate low-load equipment (e.g., equipment C with a load rate of 60%) to avoid overloading equipment A;

[0179] Adjust the weights for material shortage response:

[0180] If the material turnover rate continuously falls below the target (e.g., continuously < 3.0 times / month for 2 weeks), then increase the weight of "urgent transfer priority":

[0181] Original weight: 0.4 → new weight: 0.6, trigger automatic application for cross-factory material transfer;

[0182] Update the parameters of the equipment fingerprint library to correct the parameters of the bottleneck equipment and the warning threshold for high-load equipment;

[0183] Correct the parameters of the bottleneck equipment:

[0184] Since the actual production capacity of equipment D is lower than 80% of the theoretical value, update the parameters of the equipment fingerprint library:

[0185] Recommended priority: reduced from 80% to 60%;

[0186] Allowed load range: adjusted from "≤ 90%" to "≤ 80%" (to avoid overloading);

[0187] Execution effect: In subsequent production scheduling, reduce the allocation weight of equipment D and preferentially select equipment E with stable performance;

[0188] Warning threshold for high-load equipment:

[0189] Since equipment A frequently exceeds the limit, update the "warning threshold" in the fingerprint library:

[0190] Original threshold: 85% → New threshold: 80%, triggering an earlier load balancing strategy;

[0191] The bottleneck mitigation rule is optimized into a critical path delay compensation mechanism;

[0192] If the critical path delay of an order > 20% of the planned value (e.g., order C is delayed by 4.5 hours > 2 hours × 1.2 of the planned value), then: Automatically increase the priority weight of this order (e.g., from 0.5 → 0.7) and preferentially allocate resources;

[0193] Calibrate the "critical path priority" parameter in the rule library and extend the maintenance cycle inspection frequency of device D;

[0194] Synchronize the calibrated rule weights (e.g., load balancing weight 0.8) and device fingerprint parameters (e.g., recommended priority of device D 60%) to the production scheduling knowledge graph and rule engine through the API.

[0195] For example: When generating a new production scheduling instruction set, automatically reduce the allocation probability of device D and preferentially select device E with a higher priority in the fingerprint library;

[0196] Load the calibrated parameters back into the 3D virtual workshop model, resimulate the production scheduling plan, and verify the adjustment effect:

[0197] For example, the load rate of device A drops from 85% to 78%, and the critical path delay of device D drops from 4.5 hours to 3.2 hours;

[0198] Store the association relationship between the event tags (such as "device B failure", "material shortage") in the difference analysis and the calibrated parameters in the knowledge graph to provide historical reference for subsequent similar scenarios.

[0199] Example 2

[0200] Please refer to Figure 3 As shown, for the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. Provide a comprehensive processing system for dynamic production scheduling data of manufacturing resources, including:

[0201] Multi-source data integration and calibration module: Used to obtain multi-source heterogeneous data, extract key fields, and construct a cross-system field mapping table; Generate device fingerprint vectors and device category benchmark templates in combination with historical device performance data, and construct a device fingerprint library; According to the device fingerprint library, calculate the sensor stability index in real time, dynamically adjust the confidence weight, and then determine the device state;

[0202] Multi-dimensional device state determination module: Based on the device state determination result, align the timestamps of multi-source heterogeneous data, and trigger events through a preset rule library; Integrate the data reported by operators and multi-source heterogeneous data to generate an abnormal event stream with context tags;

[0203] Edge-Cloud Collaborative Optimization Unit: Based on the abnormal event type and the determination result of the device status, call the preset micro process, construct a recommendation plan through the production scheduling knowledge graph, and then generate a production scheduling instruction set;

[0204] Abnormal Root Cause Analysis and Closed-Loop Calibration Module: According to the production scheduling instruction set, drive the device to execute the instruction through the edge agent to generate a local optimization plan locally; the cloud integrates the data of multiple factories, generates a global optimization strategy through the black-box multi-objective algorithm, and then forms a final execution plan; compare the execution results to generate a closed-loop analysis data set, and update the production scheduling knowledge graph and the preset rule base in reverse;

[0205] Virtual Simulation and Execution Linkage Unit: Load the final execution plan into the 3D virtual workshop model, calculate the resource utilization rate and the bottleneck process index; then compare the final execution plan with the actual execution data, generate a difference analysis report, calibrate the device fingerprint library and the preset rule base in reverse, trigger the re-determination of the device status, and form an end-to-end closed-loop iteration.

[0206] Embodiment 3

[0207] This embodiment publicly provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided method for comprehensively processing manufacturing resource dynamic production scheduling data.

[0208] Since the electronic device introduced in this embodiment is the electronic device used to implement a method for comprehensively processing manufacturing resource dynamic production scheduling data in an embodiment of the present application, based on the method for comprehensively processing manufacturing resource dynamic production scheduling data introduced in an embodiment of the present application, those skilled in the art can understand the specific implementation manner and various change forms of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiment of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the method for comprehensively processing manufacturing resource dynamic production scheduling data in an embodiment of the present application, it falls within the scope of protection of the present application.

[0209] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain a formula that is closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0210] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A comprehensive processing method for dynamic scheduling data of manufacturing resources, characterized in that: include: S1: Acquire multi-source heterogeneous data, extract key fields, and build a cross-system field mapping table; Combine historical device performance data to generate device fingerprint vectors and device category benchmark templates to build a device fingerprint library; According to the device fingerprint library, the sensor stability index is calculated in real time, and the confidence weight is adjusted dynamically to determine the device status; S2: Based on the equipment status determination results, align the timestamps of multi-source heterogeneous data and trigger events through the preset rule base; integrate the operator's reported data with multi-source heterogeneous data to generate an abnormal event stream with context labels; S3: Based on the abnormal event type and equipment status determination results, call the preset micro-process, build a recommendation plan through the production scheduling knowledge graph, and then generate a production scheduling instruction set; S4: Based on the production scheduling instruction set, the edge agent drives the device to execute the instruction and generates a local optimization plan; The cloud integrates data from multiple factories, generates a global optimization strategy through a black box multi-objective algorithm, and then forms a final execution plan; compares the execution results to generate a closed-loop analysis data set, and reversely updates the production scheduling knowledge graph and preset rule base; S5: Load the final execution plan into the 3D virtual workshop model, calculate resource utilization and bottleneck process indicators; then compare the final execution plan with the actual execution data, generate a difference analysis report, reversely calibrate the equipment fingerprint library and the preset rule library, trigger the re-determination of the equipment status, and form an end-to-end closed-loop iteration.

2. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 1, characterized in that: The field mapping table is generated in the following manner: Multi-source heterogeneous data includes ERP order data, MES equipment status data, IoT sensor stream data and supply chain inventory data. ERP order data includes basic order information, product details, status information and related information. MES equipment status data includes real-time operation status, performance indicators, theoretical production capacity, maintenance records, production progress and alternative equipment list. IoT sensor stream data includes equipment health data, environmental data, event alarms, location and status. Supply chain inventory data includes inventory levels, inventory changes, supply chain collaboration and demand forecasts. Use natural language processing technology to parse unstructured data in multi-source heterogeneous data and extract key fields; then establish cross-system field mapping relationships based on the key fields and generate a field mapping table.

3. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 2, characterized in that: The method of dynamically adjusting the confidence weight includes: Obtain historical equipment performance data, including equipment operating efficiency indicators, sensor baseline parameters, and environmental and load data; The feature extraction method is used to extract the features of historical equipment performance data to obtain the hardware static features and dynamic behavior features of the equipment; The importance of different fields in the field mapping table is used as the initial weights of the hardware static features and dynamic behavior features to construct an initial weight matrix, and the weight matrix is ​​updated in real time according to the information entropy of the features through the entropy weight method; Based on the weight matrix, the hardware static features and dynamic behavior features are weighted, and then the principal component analysis technology is used to reduce the dimension of all weighted features to form a device fingerprint vector; At the same time, the Gaussian mixture model is used to perform unsupervised clustering on the reduced-dimensional features to generate a benchmark template for device categories; The device fingerprint vector and the device category benchmark template are distributedly optimized through the federated learning framework, thereby obtaining the optimized device fingerprint vector and device category benchmark template; Integrate the optimized device fingerprint vector and the device category benchmark template to obtain a device fingerprint library; Based on different sensors of the device, read the real-time data of the sensor, and perform difference calculation with the device category benchmark template of the corresponding category in the device fingerprint library to obtain the error value of the sensor relative to the benchmark; The mean error value of the sensor at time T in the past is taken as the historical standard deviation, and then the ratio of the current error value to the historical standard deviation is taken as the error ratio; and the reciprocal of the sum of 1 and the error ratio is taken as the stability index of the sensor; Set the error threshold and stability threshold, and define the dynamic adjustment rules of the sensor's confidence weight based on the stability index and the preset error threshold, including: If the error value is less than or equal to the error threshold and the stability index is greater than or equal to the stability threshold, the data collected by the sensor is determined to be stable; If the error value is greater than the error threshold and the stability index is less than the stability threshold, it is determined that the data collected by the sensor has potential errors, and the confidence weight of the sensor data is reduced according to a preset adjustment ratio; If the error value of the sensor is greater than the error threshold for N consecutive moments, the sensor is judged to be abnormal, and the confidence weight of the sensor is reduced to the minimum according to the adjustment ratio; According to the adjusted confidence weight of the sensor data, if the confidence weight is greater than the preset confidence weight threshold, it is marked as high-confidence sensor data; if the confidence weight is less than or equal to the preset confidence weight threshold, it is marked as low-confidence sensor data.

4. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 3, characterized in that: The device status determination method includes: According to the field mapping table, the real-time data of sensors and multi-source heterogeneous data are mapped into standardized fields, and classified into sensor data, manual data and ERP order data according to data type; The confidence weight of sensor data is used as the weight of sensor data, and fixed weights are assigned to manual data and ERP order data; All standardized fields of sensor data, manual data and ERP order data are weighted and summed respectively to obtain sensor comprehensive value, manual comprehensive value and ERP comprehensive value respectively; and then horizontally spliced ​​to form a comprehensive state vector; According to the comprehensive state vector, the sensor comprehensive value threshold, ERP comprehensive value threshold and manual comprehensive value threshold are defined respectively, and the equipment state judgment rule is defined as: If the sensor comprehensive value is greater than or equal to the sensor comprehensive value threshold, the ERP comprehensive value is greater than or equal to the ERP comprehensive value threshold, and the manual comprehensive value is greater than or equal to the manual comprehensive value threshold, the device status is determined to be normal operation; If any of the sensor comprehensive value, manual comprehensive value and ERP comprehensive value is less than the corresponding threshold, the equipment status is judged as partial failure; If the sensor comprehensive value is lower than the sensor comprehensive value threshold for N consecutive moments, and the ERP comprehensive value is less than the ERP comprehensive value threshold, and the manual comprehensive value is less than the manual comprehensive value threshold, the equipment status is determined to be shutdown for maintenance.

5. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 4, characterized in that: The generation method of the abnormal event stream includes: Based on the device status determination result, the determined device status is used as a status label, and the timestamps of the multi-source heterogeneous data are aligned to the same time axis, and then the current device status determination result is added as a label to the multi-source heterogeneous data at each moment; Build a preset rule base to monitor the equipment status judgment results and multi-source heterogeneous data in real time. When the conditions in the preset rule base are met, a trigger event is generated. Based on the generated trigger events, the order status, MES process progress and IoT sensor anomalies in the ERP order data are used as multi-source data tags, which are then combined with the description reported in real time by the operator and the corresponding equipment status determination results to form a root cause tag as a context tag; The context tags, the type and priority of the triggering event, the original data source of the triggering event and the corresponding timestamp are integrated into the abnormal event stream.

6. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 5, characterized in that: The construction method of the preset rule base includes: According to industry standards and historical equipment performance data, the equipment status determination results are used as the basis for dynamic rule adjustment. The basic parameters of threshold rules are extracted from industry standards, historical equipment performance data and multi-source heterogeneous data after timestamp alignment, including static rule parameters and dynamic rule parameters. The rules of the preset rule base include static rules and dynamic rules. Static rules are defined according to static rule parameters, and dynamic rules are defined according to dynamic rule parameters. Static rules represent fixed thresholds preset based on industry standards and historical equipment performance data, including process timeout thresholds, order delay thresholds, and inventory safety thresholds; dynamic rules represent the definition of conditions and adjustment logic based on the results of equipment status judgment, including dynamic adjustment of equipment OEE, sensor alarm linkage, and inventory linkage.

7. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 6, characterized in that: The generation method of the production scheduling instruction set includes: According to the equipment status determination results and the abnormal event flow, the event type in the abnormal event flow is matched with the preset micro-process library to generate a preliminary adjustment plan; Based on the event type of the equipment, obtain the mean repair time for similar faults, performance records of alternative equipment, and associated solutions; Extract equipment, orders, processes and alternative equipment from multi-source heterogeneous data after timestamp alignment as entities. According to the preliminary adjustment plan, extract the relationship between each entity in the abnormal event flow and the preset rule base, and add the average repair time of similar faults, performance records of alternative equipment and related solutions as attributes into the relationship, thereby building a production scheduling knowledge graph. Use the inference engine to query the historical repair time of similar faults and the performance records of alternative equipment through the production scheduling knowledge graph according to event types and context labels, and build a recommendation plan; generate a production scheduling instruction set based on the recommendation plan.

8. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 7, characterized in that: The final implementation plan is formed in the following ways: The production scheduling instruction set is synchronized to the workshop-level edge agent to drive the equipment control layer and the production management system to execute the production scheduling instructions. During the execution of the instructions, the real-time operation status and production progress of the equipment are collected in real time through IoT sensors to form an execution feedback stream; and the local optimization plan is generated instantly through the edge agent based on the equipment health status and the list of alternative equipment; After the instruction is executed, the final execution result of the instruction is obtained by summarizing the execution feedback flow; Compare the execution results with the conditions in the preset rule base, the equipment status determination results and equipment status data, and the order data to identify the deviation value of the execution; Collect execution feedback flows and abnormal event flows from multiple factories through cloud servers to form a global resource view; The global optimization strategy is obtained based on the global resource view and the historical deviation value of the device through the black box multi-objective algorithm; Integrate the global optimization strategy of the cloud server with the local optimization solution of the edge agent to form the final execution plan; The deviation value is associated with the original scheduling instructions in the scheduling instruction set and the context label of the abnormal event stream to form a closed-loop analysis data set; then the scheduling knowledge graph and preset rule base are updated based on the closed-loop analysis data set.

9. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 8, characterized in that: The resource utilization rate refers to the efficiency of using equipment resources and material resources, including equipment utilization rate, material turnover rate and manpower utilization rate; Bottleneck process indicators indicate the identification of key bottleneck links in the production process, including critical path delay time and bottleneck equipment throughput; The final execution plan is loaded into the 3D virtual workshop model. The longest dependent path of the order is obtained according to the simulation results of the 3D virtual workshop model through the Gantt chart. The sum of the process delays on the longest dependent path of the order is taken as the critical path delay time. A standard capacity threshold is set for the theoretical capacity of the equipment. If the actual capacity of the equipment is less than the standard capacity threshold, the equipment is determined to be a bottleneck equipment, and the ratio of the actual capacity of the bottleneck equipment to the theoretical capacity is taken as the bottleneck equipment throughput.

10. A comprehensive processing method for dynamic scheduling data of manufacturing resources according to claim 9, characterized in that: The generation method of the difference analysis report includes: According to the final execution plan, execute the production schedule, monitor the actual execution data in real time, compare the final execution plan with the actual execution data, and generate a difference analysis report, including resource utilization difference analysis and bottleneck process indicator difference analysis.

Images