Production flow AI management method and system based on time sequence production data flow

Abstract

The invention relates to a production process AI management method and system based on a time sequence production data stream, and the method comprises the steps: obtaining multi-source heterogeneous production data covering a whole production process, carrying out the preprocessing operation, and generating a standardized time sequence production data stream; analyzing the standardized time sequence production data flow, constructing an equipment health state feature vector and a production task coupling feature matrix, and further constructing a multi-dimensional production state feature map; inputting the multi-dimensional production state characteristic spectrum into a preset production process optimization model, performing global optimization reasoning, generating a preliminary scheduling adjustment instruction set, performing feasibility verification and conflict resolution, and generating a final scheduling adjustment instruction set; according to the scheme, intelligent management of the production process can be realized, potential bottlenecks and risk points in the production process are effectively identified, and the production efficiency and the resource utilization rate are improved.

Description

Technical Field

[0001] This application relates to the fields of industrial automation, big data and information technology, and in particular to an AI management method and system for production processes based on time-series production data flow. Background Technology

[0002] In modern manufacturing, production process management is a crucial link in ensuring efficient and stable production. With the continuous expansion of production scale and the increasing complexity of production processes, the types of equipment involved in the production process are numerous, and the data sources are diverse and varied in format, exhibiting a multi-source heterogeneous characteristic. This multi-source heterogeneous production data contains rich information such as equipment operating status, production execution progress, and material flow. Effective management and analysis of this data can provide important support for production decision-making.

[0003] To address the challenges of managing multi-source, heterogeneous production data, existing technologies typically employ data integration and simple data analysis methods. Data integration technology consolidates data from different sources into a unified platform, and performs preliminary cleaning and processing to eliminate noise and inconsistencies. Simultaneously, data analysis tools are used to mine and analyze the processed data, extracting valuable information to assist production managers in decision-making.

[0004] However, existing solutions have significant shortcomings in handling the global state analysis and optimal scheduling of complex production systems. Existing methods often only allow for partial and one-sided analysis of production data, making it difficult to comprehensively and accurately grasp the overall operational status of the production system. They fail to identify potential bottlenecks and risk points in the production process and struggle to achieve global optimal scheduling, leading to problems such as low production efficiency and resource waste. Summary of the Invention

[0005] The main purpose of this application is to provide a production process AI management method and system based on time-series production data flow, which can comprehensively and accurately analyze the global status of the production system, identify potential bottlenecks and risk points, and realize global optimization scheduling of the production process to improve production efficiency and resource utilization.

[0006] To achieve the above objectives, embodiments of the present invention provide an AI management method for production processes based on time-series production data streams, the method comprising the following steps:

[0007] Acquire multi-source heterogeneous production data covering the entire production process, including real-time collected equipment operating parameter data and production execution process data retrieved from the manufacturing execution system;

[0008] The multi-source heterogeneous production data is preprocessed to generate a standardized time-series production data stream;

[0009] By performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized time-series production data stream, a feature vector representing the health status of the equipment is constructed.

[0010] Based on the work order information and material flow records in the production execution process data of the standardized time-series production data stream, as well as the health status feature vector, a production task coupling feature matrix representing the correlation of production tasks is constructed.

[0011] The device health status feature vector is mapped to graph node attributes by a graph construction algorithm, and the production task coupling feature matrix is ​​mapped to the weight of graph edge to construct a multi-dimensional production status feature graph representing the global state of the production system. Nodes represent generating devices, and edges represent the logical relationships between production tasks.

[0012] The multidimensional production state feature map is input into a preset production process optimization model. The production process optimization model analyzes the topology and node attributes of the multidimensional production state feature map, identifies potential bottlenecks and risk points in the production process, performs global optimization reasoning, and generates a preliminary scheduling adjustment instruction set.

[0013] The feasibility of the preliminary scheduling adjustment instruction set is verified and conflicts are resolved. Combined with the preset production rule base and resource constraints, the preliminary scheduling adjustment instruction set is optimized and adjusted to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, so as to realize intelligent management of production process.

[0014] Accordingly, this application also provides a production process AI management system based on time-series production data flow, the system comprising:

[0015] The acquisition module is used to acquire multi-source heterogeneous production data covering the entire production process. The multi-source heterogeneous production data includes real-time collected equipment operating parameter data and production execution process data retrieved from the manufacturing execution system.

[0016] The preprocessing module is used to preprocess the multi-source heterogeneous production data to generate a standardized time-series production data stream.

[0017] The feature construction module is used to construct a feature vector representing the health status of the equipment by performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized time-series production data stream.

[0018] The matrix construction module is used to construct a production task coupling feature matrix that characterizes the correlation of production tasks based on the work order information and material flow records in the production execution process data in the standardized time-series production data stream, as well as the health status feature vector.

[0019] The graph generation module is used to map the device health status feature vector to graph node attributes through a graph construction algorithm, and to map the production task coupling feature matrix to the weights of graph edges, thereby constructing a multi-dimensional production status feature graph that represents the global state of the production system. In this graph, nodes represent generating devices, and edges represent the logical relationships between production tasks.

[0020] The optimization module is used to input the multidimensional production state feature map into a preset production process optimization model, analyze the topology and node attributes of the multidimensional production state feature map through the production process optimization model, identify potential bottlenecks and risk points in the production process, perform global optimization reasoning, and generate a preliminary scheduling adjustment instruction set.

[0021] The verification module is used to verify the feasibility and resolve conflicts of the preliminary scheduling adjustment instruction set. Combined with the preset production rule base and resource constraints, the preliminary scheduling adjustment instruction set is optimized and adjusted to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, so as to realize intelligent management of production process.

[0022] In summary, the technical solution of this application improves data quality and usability by acquiring multi-source heterogeneous production data covering the entire production process and performing preprocessing operations to generate a standardized time-series production data stream. Analyzing the standardized time-series production data stream allows for the construction of equipment health status feature vectors and production task coupling feature matrices, thereby creating a multi-dimensional production state feature map that comprehensively and accurately describes the global state of the production system. Inputting this multi-dimensional production state feature map into a pre-defined production process optimization model enables global optimization reasoning, generating a preliminary scheduling adjustment instruction set. Feasibility verification and conflict resolution are then performed to generate the final scheduling adjustment instruction set. This achieves intelligent management of the production process, effectively identifying potential bottlenecks and risks, and improving production efficiency and resource utilization. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1This is a schematic diagram of a scenario for the AI ​​management method for production processes based on time-series production data streams in an embodiment of this application;

[0025] Figure 2 A flowchart of a production process AI management method based on time-series production data stream is provided for embodiments of this application;

[0026] Figure 3 This is a schematic diagram illustrating the process of generating device health status feature vectors provided in the embodiments of this application.

[0027] Figure 4 This is a schematic diagram of the process for generating the coupling feature matrix provided in an embodiment of this application;

[0028] Figure 5 This is a schematic diagram illustrating the process of generating feature images provided in an embodiment of this application;

[0029] Figure 6 This is a schematic diagram illustrating the process of generating scheduling adjustment instructions provided in an embodiment of this application;

[0030] Figure 7 Another flowchart illustrating the generation of scheduling adjustment instructions provided in this application embodiment;

[0031] Figure 8 A schematic diagram of the structure of the production process AI management system based on time-series production data flow provided in this application embodiment;

[0032] Figure 9 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0033] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0034] This application provides a production process AI management method and system based on time-series production data flow, which will be described in detail below.

[0035] In this embodiment, the AI ​​(Artificial Intelligence) production process management method based on time-series production data stream is a comprehensive production management approach. It involves using AI and other technologies to collect, process, and analyze multi-source heterogeneous production data across the entire production process to comprehensively and accurately grasp the global state of the production system. Specifically, by performing timestamp alignment and data cleaning operations on the multi-source heterogeneous production data, a standardized time-series production data stream is generated. The equipment operating parameter data and production execution process data in the standardized time-series production data stream are analyzed to construct equipment health status feature vectors representing equipment health status and production task coupling feature matrices representing the correlation of production tasks, respectively. A graph construction algorithm maps the equipment health status feature vectors and production task coupling feature matrices to graph node attributes and edge weights, constructing a multi-dimensional production state feature graph. The multi-dimensional production state feature graph is input into a preset production process optimization model for global optimization reasoning, generating a preliminary scheduling adjustment instruction set. Feasibility verification and conflict resolution are then performed to generate the final scheduling adjustment instruction set, achieving intelligent management of the production process. The entire process is intelligent, comprehensive, and optimized, effectively improving production efficiency and resource utilization.

[0036] like Figure 1 As shown, a production process AI management method scenario based on time-series production data flow is provided. The production process AI management scenario based on time-series production data flow mainly includes production equipment, manufacturing execution system, data acquisition system, data analysis platform and production scheduling system; wherein the production equipment, manufacturing execution system, data acquisition system, data analysis platform and production scheduling system are connected through network.

[0037] Taking an automobile manufacturing plant as an example, this scenario involves a large number of production equipment, such as stamping presses, welding robots, painting equipment, and assembly lines. These devices work together to complete the automobile manufacturing process. The Manufacturing Execution System (MES) manages production plans, work order information, material flow, and other production execution process data. The Data Acquisition System (DAS) collects real-time operating parameter data of the production equipment, such as temperature, pressure, speed, and vibration.

[0038] The data acquisition system transmits the collected equipment operating parameters to the data analysis platform. Simultaneously, the manufacturing execution system also transmits production execution process data to the data analysis platform. The data analysis platform then processes and analyzes the multi-source heterogeneous production data. First, it performs timestamp alignment and data cleaning operations on the multi-source heterogeneous production data to eliminate time discrepancies and noise interference between data sources, generating a standardized time-series production data stream. For example, when processing equipment operating parameter data, the different data acquisition frequencies of different devices may lead to timestamp discrepancies. Timestamp alignment unifies the data from different devices to the same time scale. For outliers in the data, such as abnormal temperature or pressure values ​​caused by sudden equipment failures, they are replaced according to preset outlier handling rules to ensure data accuracy.

[0039] Next, multi-scale time-frequency domain analysis is performed on the equipment operating parameter data in the standardized time-series production data stream to construct an equipment health status feature vector representing the equipment's health status. For example, by performing Fast Fourier Transform and Wavelet Transform on the equipment's vibration signals, feature values ​​in the frequency and time domains are extracted, and the equipment's health status score is calculated. The equipment health status scores within all time windows are then combined to form the equipment health status feature vector. Simultaneously, based on work order information and material flow records within the production execution process data in the standardized time-series production data stream, as well as the equipment health status feature vector, a production task coupling feature matrix representing the correlation of production tasks is constructed. For example, the logical dependencies between production tasks are determined based on work order information, the material supply relationships between production tasks are determined based on material flow records, and the coupling weights between production tasks are calculated by combining the equipment health status feature vector, generating the production task coupling feature matrix.

[0040] Then, a graph construction algorithm is used to map the equipment health status feature vectors to graph node attributes, and the production task coupling feature matrix to the weights of graph edges, thus constructing a multidimensional production state feature graph representing the global state of the production system. Nodes represent production equipment, and edges represent the logical relationships between production tasks. For example, in the multidimensional production state feature graph, the attributes of the stamping machine node include its equipment health status feature vector, and the weights of the edges between the stamping machine and the welding robot are determined by the production task coupling feature matrix, reflecting the degree of correlation between these two production tasks.

[0041] A multi-dimensional production state feature map is input into a pre-defined production process optimization model. The model analyzes the topology and node attributes of the map to identify potential bottlenecks and risks in the production process. Global optimization reasoning is then performed to generate a preliminary set of scheduling adjustment instructions. For example, by analyzing key nodes and critical paths in the map, bottlenecks in the production process are identified, such as a slow assembly line affecting the overall production schedule. Preliminary scheduling adjustment instructions are then generated for these bottlenecks, such as increasing the number of workers on the assembly line or adjusting the allocation of production tasks.

[0042] The production scheduling system verifies the feasibility and resolves conflicts in the initial set of scheduling adjustment instructions. Combining this with a pre-set production rule base and resource constraints, it optimizes and adjusts the initial set of instructions, generating a final set that includes targeted scheduling adjustment suggestions and resource reallocation plans, which is then sent to the manufacturing execution system. For example, when verifying the feasibility of scheduling adjustment instructions, it checks whether they meet production process requirements and whether resource conflicts exist. If it is found that increasing the number of workers on a certain assembly line would lead to insufficient manpower on other assembly lines, the instructions are adjusted according to resource constraints, and human resources are reallocated to achieve intelligent management of the production process.

[0043] refer to Figure 2 , Figure 2 This is a flowchart illustrating a production process AI management method based on time-series production data streams provided in this application embodiment. The execution subject of this method can be computer equipment, such as a server. The production process AI management method based on time-series production data streams provided in this application embodiment specifically includes:

[0044] S10: Acquire multi-source heterogeneous production data covering the entire production process, including real-time collected equipment operating parameter data and production execution process data retrieved from the manufacturing execution system.

[0045] In this embodiment, the entire production process refers to the entire production process from raw material input to final product output, encompassing all links and stages within the production system. It involves multiple aspects such as the operation of production equipment, the execution of production tasks, and the flow of materials. In the entire production process, raw materials first enter the production system and, after a series of processing and treatments, are gradually transformed into semi-finished products and final products.

[0046] In this embodiment, multi-source heterogeneous production data refers to production-related data from different data sources with different formats and characteristics. Equipment operating parameter data is key data reflecting the real-time operating status of production equipment, including parameters such as temperature, pressure, speed, and vibration. These parameters directly reflect the working status and performance of the equipment. For example, in an automobile manufacturing plant, the pressure parameters of a stamping press can reflect whether its stamping capacity is normal, and the temperature parameters of a welding robot can reflect its workload and heat dissipation. Production execution process data is detailed data recording the execution of production tasks, including work order information, material flow records, and production progress. Work order information clarifies the specific requirements and plans for production tasks, while material flow records the flow of raw materials and parts during the production process. For example, work order information may specify the production quantity, model, and delivery time of a certain batch of cars, while material flow records the flow of raw materials such as steel and tires from the warehouse to the production line.

[0047] In one embodiment, various sensors, such as temperature sensors, pressure sensors, and vibration sensors, can be installed on the production equipment to collect real-time equipment operating parameter data. These sensors can monitor the equipment's operating parameters in real time and transmit the data to a data acquisition system. The data acquisition system can use wireless transmission technology to send the collected data to a data analysis platform. For production execution process data retrieved from the Manufacturing Execution System (MES), interface technology can be used to connect with the MES to obtain production execution process data periodically or in real time. In an automotive manufacturing plant, data interaction can be achieved between the Enterprise Resource Planning (ERP) system and the MES to obtain work order information and material flow records.

[0048] S20: Perform preprocessing operations on the multi-source heterogeneous production data to generate a standardized time-series production data stream.

[0049] In one embodiment, the preprocessing operation may include timestamp alignment and data cleaning operations, that is, timestamp alignment and data cleaning operations may be performed on the multi-source heterogeneous production data to eliminate time deviations and noise interference between data sources and generate a standardized time-series production data stream.

[0050] In this embodiment, timestamp alignment refers to unifying data from different data sources onto the same time scale to eliminate time discrepancies between data. Because different devices may have different data acquisition frequencies, timestamps may differ. For example, some devices may acquire data every minute, while others may acquire data every five minutes, leading to inconsistencies in data timing. Data cleaning refers to processing noise, outliers, and missing values ​​in the data to improve its quality and usability. Noise refers to random interference signals mixed in with the data; outliers are data points that significantly deviate from the normal range; and missing values ​​are data missing at certain locations.

[0051] In this embodiment, by performing timestamp alignment and data cleaning operations on multi-source heterogeneous production data, time deviations and noise interference between data sources can be eliminated, generating a standardized time-series production data stream. The standardized time-series production data stream has a unified time format and data quality, facilitating subsequent data analysis and processing. Technically, this improves data quality and usability, providing a reliable data foundation for subsequent data analysis and modeling.

[0052] In one embodiment, step S20 can be implemented in the following manner, specifically including:

[0053] A1: Obtain the timestamp of each group of data in the multi-source heterogeneous production data, and calculate the maximum time difference between the timestamps.

[0054] In this embodiment, a timestamp refers to the point in time when data is recorded, used to identify the time when the data was generated. In multi-source heterogeneous production data, the data collection times of different data sources may differ, resulting in differences in the timestamps of the data. Calculating the maximum time difference between timestamps can reveal the degree of time deviation between data sources. For example, in a dataset containing operating parameter data and production execution process data from multiple devices, some devices collect data every minute, while others collect data every five minutes. This can lead to significant differences in the timestamps of different data. By calculating the maximum time difference, the upper limit of the time deviation can be determined, providing a basis for subsequent timestamp alignment operations.

[0055] In this embodiment, accurately obtaining the timestamp and calculating the maximum time difference helps determine whether timestamp alignment is necessary. If the maximum time difference is small, it indicates that the time deviation between data sources is not significant, and complex processing may not be required; if the maximum time difference is large, appropriate measures need to be taken for timestamp alignment.

[0056] In one embodiment, each group of data in the multi-source heterogeneous production data can be traversed to extract its timestamp information, and these timestamps are stored in an array. Then, the timestamps in the array are sorted, and the difference between the maximum value and the minimum value is calculated to obtain the maximum time difference. For example, the sorting function in a programming language is used to sort the timestamp array, and the maximum time difference is obtained through a simple subtraction operation.

[0057] A2: If the maximum time difference is greater than the preset time threshold, interpolation processing is performed on the multi-source heterogeneous production data to fill in the missing data in the time interval.

[0058] In the embodiments of the present application, the preset time threshold is a time limit set according to the actual production situation and data processing requirements. When the maximum time difference is greater than the preset time threshold, it indicates that the time deviation between data sources is relatively large, and there may be missing data in the time interval. Interpolation processing is a method of estimating unknown data points between known data points. By interpolation, the missing data in the time interval can be filled, making the data more continuous in the time dimension. For example, in the device operation parameter data, if there is no data record in a certain time period, the data value in this time period can be estimated according to the data points before and after through interpolation processing.

[0059] In the embodiments of the present application, performing interpolation processing on the multi-source heterogeneous production data can improve the integrity and continuity of the data. Interpolation processing can reduce the impact of time deviation on the data analysis results and improve the usability of the data.

[0060] In one embodiment, linear interpolation can be used to perform interpolation processing on the multi-source heterogeneous production data. Linear interpolation is a simple and commonly used interpolation method, which assumes that the data changes linearly between two adjacent known data points. The specific steps are as follows: First, determine the time point to be interpolated; then, find the two known data points before and after this time point; finally, calculate the interpolated data at this time point according to the linear relationship. For example, given that the temperature values of a certain device at times t1 and t2 are T1 and T2 respectively, and the temperature value T at time t (t1 < t < t2) needs to be calculated, the linear interpolation formula T = T1 + (T2 - T1) * (t - t1) / (t2 - t1) can be used for calculation.

[0061] A3: Detect the outliers in the multi-source heterogeneous production data, and replace the outliers according to the preset outlier processing rules to generate the initially cleaned production data.

[0062] In this embodiment, outliers refer to data points that significantly deviate from the normal range. In multi-source heterogeneous production data, outliers may be caused by equipment failure, sensor errors, data transmission errors, etc. Detecting and processing outliers can improve data quality and reliability. The preset outlier handling rules are formulated based on actual production conditions and data analysis requirements to determine how to handle detected outliers. For example, the rules may specify replacing outliers with the average of adjacent data points, the median of historical data, etc.

[0063] In one embodiment, a statistical method can be used to detect outliers. For example, the mean and standard deviation of the data are calculated, and data points that deviate from the mean by more than a certain multiple of the standard deviation are identified as outliers. Assuming the data follows a normal distribution, data points that deviate from the mean by more than three times the standard deviation are typically considered outliers. Then, according to preset outlier handling rules, the detected outliers are replaced with appropriate values. For example, outliers are replaced with the average of adjacent data points. In practice, each data point in the dataset can be iterated over, its difference from the mean can be calculated, and it can be determined whether it exceeds a preset multiple of the standard deviation; if it does, it is replaced.

[0064] A4: Perform sliding window filtering on the pre-cleaned production data to further remove noise interference and generate a standardized time-series production data stream.

[0065] In this embodiment, sliding window filtering is a commonly used signal processing method for smoothing data and removing noise. It works by sliding a fixed-size window across the data sequence, performing operations such as averaging or weighted averaging on the data within the window to obtain a new data sequence. Even after initial cleaning, some subtle noise interference may still exist in the production data. Sliding window filtering can further remove this noise, making the data smoother and more stable. For example, in equipment operating parameter data, small fluctuations may occur due to sensor measurement errors, and sliding window filtering can smooth out these fluctuations.

[0066] In one embodiment, a simple moving average method can be used for sliding window filtering. The simple moving average method involves calculating the arithmetic mean of the data within a sliding window. The specific steps are as follows: First, determine the size of the sliding window; then, slide the window sequentially across the data sequence, calculating the average value of the data within the window each time, and using this average value as the data value at the center of the window. For example, if a window size of 5 is selected, for the data sequence [x1,x2,x3,x4,x5,x6,...], the first window contains the data [x1,x2,x3,x4,x5], and its average is calculated as the new value of x3; then the window is moved one position to the right, containing the data [x2,x3,x4,x5,x6], and its average is calculated as the new value of x4, and so on.

[0067] S30: By performing multi-scale time-frequency domain analysis on the equipment operation parameter data in the standardized time-series production data stream, a feature vector representing the health status of the equipment is constructed.

[0068] In this embodiment, multi-scale time-frequency domain analysis is a method for analyzing signals, capable of simultaneously decomposing and analyzing signals in both the time and frequency domains. Through multi-scale time-frequency domain analysis, equipment operating parameter data can be decomposed into signals with different frequency components and time scales, extracting feature information reflecting the equipment's health status. The equipment health status feature vector is a multi-dimensional vector, with each dimension representing a feature of the equipment's health status. For example, the equipment's vibration signal can be decomposed into components of different frequencies, and the amplitude and phase of each frequency component can serve as a feature of the equipment's health status.

[0069] In this embodiment, by performing multi-scale time-frequency domain analysis on equipment operating parameter data in a standardized time-series production data stream, a device health status feature vector is constructed, which can accurately assess the health status of the equipment. The device health status feature vector comprehensively reflects various operating states and performance indicators of the equipment, providing an important basis for equipment maintenance and management. From a technical perspective, it enables timely detection of potential equipment faults and anomalies, allowing for proactive maintenance measures, reducing equipment downtime, and improving production efficiency.

[0070] In one embodiment, Fast Fourier Transform (FFT) and Wavelet Transform (WT) can be used to perform multi-scale time-frequency domain analysis on equipment operating parameter data. Fast Fourier Transform converts time-domain signals into frequency-domain signals, extracting the frequency components. Wavelet Transform analyzes signals at different time scales, providing both time and frequency information. For example, performing a Fast Fourier Transform on the vibration signal of equipment yields its spectrum; analyzing the peak frequency and amplitude in the spectrum can determine whether abnormal vibration exists. Performing a Wavelet Transform on the vibration signal extracts wavelet coefficients at different scales; based on the changes in these wavelet coefficients, further analysis of equipment fault characteristics can be performed. These features are then combined to form an equipment health status feature vector.

[0071] S40: Based on the work order information and material flow records in the production execution process data of the standardized time-series production data stream, as well as the health status feature vector, construct a production task coupling feature matrix that characterizes the correlation of production tasks.

[0072] In this embodiment, the work order information clearly defines the specific requirements and plans for the production task, including the name, quantity, and delivery time of the production task. The material flow record documents the flow of raw materials and components during the production process, reflecting the material dependencies between production tasks. The equipment health status feature vector reflects the health status of equipment related to the production task. The production task coupling feature matrix is ​​a two-dimensional matrix, where elements represent the degree of association between production tasks. For example, an element in a row and column of the matrix represents the coupling weight between two production tasks, reflecting their logical dependency and material supply relationship.

[0073] In this embodiment, a production task coupling feature matrix is ​​constructed based on work order information, material flow records, and equipment health status feature vectors, which can comprehensively and accurately describe the correlation between production tasks.

[0074] In one embodiment, the logical dependencies between production tasks can be determined based on work order information. For example, a production task may require the completion of another production task before it can begin; this logical dependency can be represented by a logical matrix. The material supply relationships between production tasks are determined based on material flow records. For example, a production task may require components from another production task as raw materials; this material supply relationship can be represented by a material supply matrix. The coupling weights between production tasks are calculated by combining equipment health status feature vectors. For example, when the health status of equipment on which a production task depends is poor, its coupling weight with other production tasks will decrease accordingly. These weights are then combined to form a production task coupling feature matrix.

[0075] S50: The device health status feature vector is mapped to graph node attributes through a graph construction algorithm, and the production task coupling feature matrix is ​​mapped to the weight of graph edges to construct a multi-dimensional production status feature graph representing the global state of the production system. Nodes represent generating devices, and edges represent the logical relationships between production tasks.

[0076] In this embodiment, the graph construction algorithm is a method for converting data into a graph structure, capable of mapping equipment health status feature vectors and production task coupling feature matrices into node attributes and edge weights of the graph. A multidimensional production status feature graph is a graph structure composed of nodes and edges. Nodes represent production equipment, and their attributes include equipment health status feature vectors, reflecting the equipment's health status and performance. Edges represent the logical relationships between production tasks, and their weights are determined by the production task coupling feature matrix, reflecting the degree of association between production tasks. For example, in a multidimensional production status feature graph, the attributes of a stamping machine node include its equipment health status feature vector, and the weight of the edge between the stamping machine and the welding robot reflects the material supply and logical dependency relationship between these two production tasks.

[0077] In this embodiment, by constructing a multi-dimensional production state feature map, the global state of the production system can be described intuitively and comprehensively. From a technical perspective, the multi-dimensional production state feature map provides a visual tool for analyzing and optimizing production processes, helping to identify potential bottlenecks and risk points in the production system, and enabling global optimization and scheduling of the production process.

[0078] In one embodiment, a graph embedding algorithm can be used to map the device health status feature vector and the production task coupling feature matrix into node attributes and edge weights of the graph. First, the definitions of nodes and edges in the graph are determined, with production devices as nodes and the logical relationships between production tasks as edges. Then, the device health status feature vector is used as the attribute of each node, and each node is initialized. The production task coupling feature matrix is ​​used as the weight of each edge, and each edge is initialized. Finally, an initial multidimensional production status feature graph is generated based on the initial node set and initial edge set. The initial multidimensional production status feature graph is then optimized in terms of topology to eliminate redundant nodes and edges, generating the final multidimensional production status feature graph. For example, if the weight of an edge between some nodes is found to be very small, indicating a low degree of correlation between the two production tasks, these edges can be deleted to simplify the graph structure.

[0079] S60: Input the multidimensional production state feature map into the preset production process optimization model, analyze the topology and node attributes of the multidimensional production state feature map through the production process optimization model, identify potential bottlenecks and risk points in the production process, perform global optimization reasoning, and generate a preliminary scheduling adjustment instruction set.

[0080] In this embodiment, the production process optimization model is a machine learning or deep learning-based model capable of analyzing and processing multi-dimensional production state feature graphs. Topology refers to the connection method of nodes and edges in the multi-dimensional production state feature graph, reflecting the logical relationships and dependencies between production tasks. Node attributes refer to the equipment health status feature vectors contained in the nodes, reflecting the health status and performance of the equipment. Potential bottlenecks refer to links in the production process that may limit production efficiency, such as insufficient production capacity of a certain piece of equipment or excessively long waiting times for a certain production task. Risk points refer to problems and faults that may occur during the production process, such as equipment failure or raw material shortages. The preliminary scheduling adjustment instruction set is a set of instructions used to adjust the scheduling of production tasks, including adjusting the priority of production tasks, allocating equipment, and supplying materials.

[0081] In this embodiment, a multi-dimensional production state feature map is input into a preset production process optimization model. This leverages the model's powerful analysis and reasoning capabilities to identify potential bottlenecks and risks in the production process and perform global optimization reasoning. Technically, the generated preliminary scheduling adjustment instruction set can provide specific suggestions and solutions for optimizing the production process, helping to improve production efficiency and resource utilization.

[0082] In one embodiment, the production process optimization model can employ a Graph Neural Network (GNN) model. First, the topological structure and node attribute information of a multi-dimensional production state feature map are acquired and analyzed. Through the message passing mechanism of the GNN, nodes exchange information, extracting key nodes and critical paths from the map. Key nodes are nodes that play a crucial role in the production process, such as the node corresponding to a bottleneck device; the critical path is the longest path in the production process, determining the length of the entire production cycle. Then, based on the key nodes and critical paths, a preset bottleneck identification algorithm identifies potential bottlenecks in the production process. For example, the load and utilization rate of each node are calculated; when the load of a node exceeds a certain threshold, it is identified as a potential bottleneck. For each potential bottleneck, a preset risk assessment algorithm calculates its risk value, which can comprehensively consider factors such as equipment failure rate and material supply stability. Bottlenecks are prioritized according to their risk values, with higher-risk bottlenecks being addressed first. Finally, based on the priority ranking of the bottlenecks, a preliminary scheduling adjustment instruction set is generated through a global optimization inference algorithm. For example, when an assembly line is identified as a potential bottleneck, instructions can be generated to increase the number of workers on that assembly line or adjust the allocation of production tasks.

[0083] S70: Perform feasibility verification and conflict resolution on the preliminary scheduling adjustment instruction set. Combine the preset production rule base and resource constraints to optimize and adjust the preliminary scheduling adjustment instruction set, and generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, so as to realize AI management of production process based on time-series production data flow.

[0084] In this embodiment, feasibility verification refers to checking whether the preliminary scheduling adjustment instruction set conforms to the actual production situation and relevant rules. For example, whether the instructions meet production process requirements, whether they exceed equipment production capacity, and whether they meet personnel workload. Conflict resolution refers to resolving potential conflicts and contradictions in the preliminary scheduling adjustment instruction set. For example, conflicts in the allocation of the same resource by different instructions, logical conflicts between instructions, etc. The production rule base is a database storing various rules and constraints in the production process, including production process rules, quality standards, safety regulations, etc. Resource constraints refer to limitations on available resources in the production process, such as the number of equipment, the number of personnel, and the supply of raw materials. Targeted scheduling adjustment suggestions are specific adjustment suggestions for the scheduling of production tasks based on the actual production situation and optimization goals, such as adjusting the order of production tasks and changing equipment usage plans. The resource reallocation scheme is a plan for the redistribution of resources in the production process, including equipment allocation, personnel deployment, and adjustments to the supply of raw materials.

[0085] In this embodiment, the preliminary scheduling adjustment instruction set undergoes feasibility verification and conflict resolution, and is optimized and adjusted in conjunction with the production rule base and resource constraints. This ensures that the final scheduling adjustment instruction set is feasible and effective. From a technical perspective, the generated final scheduling adjustment instruction set enables intelligent management of the production process, improving production efficiency, reducing costs, and mitigating risks.

[0086] In one embodiment, each instruction in the preliminary scheduling adjustment instruction set is first acquired, and each instruction is verified against a preset production rule base to ensure compliance with production rules. If an instruction does not comply with production rules, its priority is evaluated, and the instruction is modified or deleted based on the evaluation results. For example, an instruction requiring equipment to operate beyond its rated power clearly violates production rules and requires modification or deletion. Next, resource occupancy analysis is performed on the modified scheduling adjustment instruction set to determine if resource conflicts exist. If resource conflicts exist, the instruction order in the scheduling adjustment instruction set is rearranged according to resource constraints, and dynamic resource reallocation is performed based on a resource allocation algorithm until the resource conflict is eliminated. For example, two instructions requiring the same equipment to perform different production tasks simultaneously create a resource conflict, necessitating a readjustment of the instruction order or allocation of other equipment. Finally, the adjusted instruction set is further optimized by combining the production rule base and resource constraints to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes.

[0087] In one embodiment, the feasibility verification and conflict resolution of the preliminary scheduling adjustment instruction set in step S70 can be achieved in the following way:

[0088] B1: Obtain each instruction in the preliminary scheduling adjustment instruction set, and verify each instruction against the production rules according to the preset production rule library.

[0089] In this embodiment, the initial scheduling adjustment instruction set is a set of instructions generated by the production process optimization model for adjusting production task scheduling. The pre-set production rule base contains various rules and constraints that must be followed during production, such as production process rules, quality standards, and safety regulations. Verifying whether each instruction conforms to the production rules is a crucial step in ensuring the scheduling adjustment instructions are practically operable. For example, the production rule base may stipulate that a certain piece of equipment cannot perform a certain operation within a specific time period, or that a certain production task must be completed under specific environmental conditions.

[0090] In one embodiment, instruction verification can be implemented programmatically. First, a preliminary set of scheduling adjustment instructions is stored in a data structure, such as a list or array. Then, each instruction in the instruction set is traversed, and matched and judged according to a preset production rule base. Assume the production rule base is stored in the form of a database, and each rule has clearly defined conditions and constraints. For each instruction, the database is queried to check whether it meets the rule's conditions. If it does not meet the conditions, the instruction is marked as not conforming to the production rules.

[0091] B2: If the instruction does not comply with the production rules, the instruction will be given priority evaluation, and the instruction will be modified or deleted according to the priority evaluation result.

[0092] In this embodiment, when an instruction does not conform to production rules, it needs to be prioritized. Priority assessment determines the priority of an instruction based on factors such as its importance and urgency to production goals. Modifying or deleting instructions based on the priority assessment results aims to achieve production goals as much as possible while ensuring production rules are followed. For example, an instruction may require the early completion of a production task, but this does not conform to production process rules. By assessing its priority, if the task has a minor impact on the overall production schedule, the instruction can be considered for deletion; if the impact is significant, the instruction can be modified, such as adjusting the task's timeline.

[0093] In this embodiment, instructions that do not conform to production rules are prioritized and processed, which optimizes the scheduling adjustment scheme while adhering to production rules. From a technical perspective, this improves the rationality and effectiveness of the scheduling adjustment instruction set.

[0094] In one embodiment, several evaluation indicators can be set for each instruction, such as the degree of improvement in production efficiency, the impact on product quality, and the degree of optimization of resource utilization. A priority score is calculated for each instruction based on these indicators. Instructions with low priority scores can be directly deleted; instructions with high priority scores can be modified to conform to production rules through further matching and negotiation with the production rule base. For example, parameters such as task time and resource allocation in the instruction can be modified to meet the requirements of the production rules.

[0095] B3: Perform resource usage analysis on the modified scheduling adjustment instruction set to determine whether there are resource conflicts; if resource conflicts exist, rearrange the instruction order in the scheduling adjustment instruction set according to resource constraints, and perform dynamic resource reallocation based on resource allocation algorithm until the resource conflicts are eliminated.

[0096] In this embodiment, resource occupancy analysis evaluates the modified scheduling adjustment instruction set to check whether excessive occupancy or competition for the same resource occurs when these instructions are executed. Resource conflict refers to multiple instructions simultaneously requesting the use of the same resource, resulting in the resource being unable to meet the needs of all instructions. Resource constraints refer to limitations on available resources during the production process, such as the number of equipment, personnel, and raw material supply. Reordering the instruction sequence and dynamically reallocating resources based on resource constraints aims to rationally utilize resources and eliminate resource conflicts.

[0097] In this embodiment, resource occupancy analysis and resource conflict handling ensure the smooth execution of scheduling and adjustment instruction sets even with limited resources. Technically, this improves resource utilization and avoids resource waste and production interruptions.

[0098] In one embodiment, a resource model can be established to store information and status of various resources. For the modified scheduling adjustment instruction set, the execution process of the instructions is simulated, calculating the resource requirements and occupancy of each instruction. If resource conflicts are detected, the instruction order is adjusted according to resource constraints. For example, instructions with lower critical resource requirements are prioritized for execution. Simultaneously, resources are dynamically reallocated based on resource allocation algorithms, such as greedy algorithms or dynamic programming algorithms. During execution, resource conflicts are continuously checked for resolution until all resource conflicts are resolved.

[0099] In one embodiment, reference Figure 3 Step S30 may include steps S31-S34, which will be described in detail below:

[0100] S31: Obtain the equipment operation parameter data in the standardized time-series production data stream, and perform segmentation processing on the equipment operation parameter data to obtain multiple subsets within time windows.

[0101] In this embodiment, the standardized time-series production data stream is data with a unified time format and high quality, processed through timestamp alignment and data cleaning. Equipment operating parameter data is an important component of this data stream, reflecting the real-time operating status of the equipment. Segmenting the equipment operating parameter data involves dividing it into multiple fixed-length time windows in chronological order, with the data within each time window constituting a subset. For example, dividing the equipment's temperature data into 10-minute time windows yields multiple subsets containing 10-minute temperature data points.

[0102] In one embodiment, the length of the time window can be determined, and the data can then be segmented according to that length. For example, fixed-length data segments can be sequentially extracted from the dataset and stored as sub-datasets. Assuming that the device operating parameter data is stored in an array, and the time window length is 10 data points, 10 data points are extracted at a time in a loop to form a sub-dataset until the entire dataset has been traversed.

[0103] S32: Perform Fast Fourier Transform and Wavelet Transform on the subset of data within each time window to extract the feature values ​​in the frequency domain and time domain, respectively.

[0104] In this embodiment, the Fast Fourier Transform (FFT) is an efficient algorithm for calculating the Discrete Fourier Transform (DFT), used to convert time-domain signals into frequency-domain signals. Through the Fast Fourier Transform, a subset of data within each time window can be transformed from the time domain to the frequency domain, extracting the frequency components and amplitude information of the signal. For example, for the vibration signal of a device, after the Fast Fourier Transform, vibration amplitudes of different frequency components can be obtained, and these amplitudes can be used as eigenvalues ​​in the frequency domain. Wavelet Transform is a time-frequency analysis method that can simultaneously decompose and analyze signals in both the time and frequency domains. Through the wavelet transform, feature information of the signal can be extracted at different time scales and frequency ranges, obtaining eigenvalues ​​in the time domain. For example, the wavelet transform can decompose a signal into wavelet coefficients of different scales, which reflect the local characteristics of the signal at different times and frequencies.

[0105] In one embodiment, specialized signal processing libraries can be used to implement Fast Fourier Transform (FFT) and Wavelet Transform. For example, in Python, the `fft` function from the NumPy library can be used for FFT, and the `PyWavelets` library can be used for Wavelet Transform. Specifically, a subset of the data within each time window is taken as input, the corresponding function is called to perform the transformation, and then the desired feature values ​​are extracted from the transformation results. For FFT, the amplitudes of different frequency components are extracted; for Wavelet Transform, wavelet coefficients at different scales are extracted.

[0106] S33: Calculate the device health status score within each time window based on the characteristic values ​​in the frequency domain and time domain.

[0107] In this embodiment, the equipment health status score is a comprehensive indicator that assesses the health status of the equipment within each time window. The score is calculated based on characteristic values ​​in the frequency and time domains, taking into account multiple factors such as the periodicity, stability, and sudden changes in equipment operation. For example, an abnormal increase in the amplitude of certain frequency components in the frequency domain may indicate a potential equipment malfunction, while a sudden change in the signal in the time domain may reflect a sudden anomaly. By comprehensively analyzing these characteristic values, a quantitative score is obtained, which can intuitively reflect the health status of the equipment within that time window.

[0108] In this embodiment, calculating the device health status score within each time window helps to promptly identify potential problems with the device. By comparing and analyzing the scores from different time windows, the changing trends of the device health status can be monitored.

[0109] In one embodiment, a weighted summation method can be used to calculate the equipment health status score. First, a weight is assigned to each feature value in the frequency and time domains, with the weight determined based on the degree of influence of the feature value on the equipment health status. For example, for the frequency domain feature values ​​of equipment vibration signals, the amplitude weight of specific frequency components related to equipment faults can be set higher. Then, each feature value is multiplied by its corresponding weight, and all results are summed to obtain the equipment health status score. The specific formula can be expressed as: Score = Frequency domain feature value 1 * weight 1 + Frequency domain feature value 2 * weight 2 + ... + Time domain feature value 1 * weight n + ...

[0110] S34: Combine the device health status scores within all time windows to form a device health status feature vector.

[0111] In this embodiment, the device health status feature vector is a multi-dimensional vector, with each dimension corresponding to a device health status score within a time window. Combining the scores from all time windows comprehensively reflects the changes in the device's health status over a period of time. For example, the device may be in different working states at different times, and the trend of these states can be clearly seen through the device health status feature vector.

[0112] In one embodiment, the device health status scores calculated within each time window can be arranged chronologically to form a one-dimensional array, which serves as the device health status feature vector. For example, assuming there are 10 time windows, each calculating a device health status score, storing these 10 scores sequentially in an array forms the device health status feature vector. In practical applications, arrays or vector data structures from programming languages ​​can be used to store these scores.

[0113] In one embodiment, reference Figure 4 Step S40 may include steps S41-S46, which will be described in detail below:

[0114] S41: Obtain the production execution process data from the standardized time-series production data stream, and extract work order information and material flow records from the production execution process data.

[0115] In this embodiment, the standardized time-series production data stream is processed data with a unified time format and high quality. Production execution process data is a crucial component of this data stream, recording the execution status of production tasks. Work order information includes specific requirements, schedules, and delivery times for production tasks, clearly defining the goals and constraints of the production tasks. Material flow records detail the movement of raw materials and components during the production process, including information such as the source, destination, and quantity of materials. For example, work order information specifies the production quantity and quality standards for a particular batch of products, while material flow records document the transportation of steel from the warehouse to the production line.

[0116] In this embodiment, acquiring production execution process data and extracting work order information and material flow records are the foundation for constructing a production task coupling feature matrix. Accurate work order information and material flow records can reflect the logical relationships and material dependencies between production tasks.

[0117] S42: Determine the logical dependencies between production tasks based on the work order information, and assign a dependency weight value to each pair of production tasks based on the task path length, wherein the dependency weight value is inversely proportional to the path length.

[0118] In this embodiment, the work order information clearly defines the sequence and logical relationship of production tasks. For example, in automobile manufacturing, the stamping task must be completed before the welding task, which constitutes a logical dependency. The task path length refers to the number of intermediate tasks traversed from the start of one production task to the end of another. The dependency weight value is used to measure the strength of the dependency relationship between two production tasks. The dependency weight value is inversely proportional to the path length, meaning that the shorter the path length, the stronger the dependency relationship between the two production tasks, and the higher the assigned dependency weight value.

[0119] In this embodiment, by determining logical dependencies and assigning dependency weight values, the degree of correlation between production tasks can be quantified. From a technical perspective, this helps in subsequently constructing an accurate production task coupling feature matrix, better reflecting the actual relationships between production tasks.

[0120] In one embodiment, the logical dependencies between production tasks can be represented by constructing a directed graph. Production tasks are treated as nodes, and logical dependencies as directed edges. The shortest path length between nodes is calculated as the task path length. Assuming the shortest path from task A to task B involves 2 intermediate tasks, and the shortest path from task A to task C involves 4 intermediate tasks, then the dependency weight between task A and task B will be higher than the dependency weight between task A and task C. The dependency weight can be calculated using a preset formula, such as dependency weight = 1 / (1 + task path length).

[0121] S43: Determine the material supply relationship between each production task based on the material flow record, and assign a supply weight value to each pair of production tasks based on the material inventory saturation.

[0122] In this embodiment, the material flow record details the movement of raw materials and components between production tasks. Analyzing these records allows for the determination of the material supply relationship between production tasks. Material inventory saturation refers to the ratio of current material inventory to maximum inventory, reflecting the material reserve status. The supply weight value measures the tightness of the material supply relationship between two production tasks; a higher material inventory saturation and a larger supply weight value indicate a more stable material supply between the two production tasks.

[0123] In this embodiment, supply weight values ​​are allocated based on material supply relationships and material inventory saturation, which takes into account the impact of material factors on the correlation of production tasks. From a technical perspective, this makes the production task coupling feature matrix more comprehensively reflect the actual correlation between production tasks.

[0124] In one embodiment, material flow records can be analyzed to determine which production tasks have material supply relationships. Then, the material inventory system is queried to obtain the inventory saturation of each material. For production task pairs with material supply relationships, supply weight values ​​are assigned based on the material inventory saturation. For example, when one production task requires a component from another production task, and the component's inventory saturation is high, a higher supply weight value is assigned. The supply weight value can be expressed as a linear function of the material inventory saturation, such as: Supply weight value = Material inventory saturation * Preset coefficient.

[0125] S44: The dependency weight value and the supply weight value are weighted and summed to obtain the basic coupling weight.

[0126] In this embodiment, the dependency weight value reflects the logical dependency relationship between production tasks, and the supply weight value reflects the material supply relationship between production tasks. Weighting and summing these two values ​​comprehensively considers both logical and material factors to obtain a basic coupling weight that fully reflects the correlation between production tasks. Through weighted summation, the proportions of logical and material factors in the basic coupling weight can be adjusted according to actual circumstances.

[0127] In one embodiment, a weight coefficient can be assigned to both the dependency weight value and the supply weight value. Assume the weight coefficient for the dependency weight value is 'a', and the weight coefficient for the supply weight value is 'b', with a + b = 1. The basic coupling weight = dependency weight value * a + supply weight value * b. The values ​​of a and b can be adjusted based on actual production conditions to highlight the importance of logical or material factors.

[0128] S45: Calculate the health score decay factor of the equipment associated with the production task based on the equipment health status feature vector.

[0129] In this embodiment, the equipment health status feature vector includes the equipment's health status score within different time windows. The health status of equipment associated with production tasks directly affects the execution of those tasks. The health score decay factor measures the degree of influence of equipment health status on the correlation between production tasks; the lower the equipment health score, the smaller the health score decay factor, indicating that poor equipment health status reduces the correlation strength between production tasks.

[0130] In this embodiment, calculating the health score decay factor incorporates equipment health factors into the analysis of production task correlation. Technically, this makes the production task coupling feature matrix more consistent with actual production conditions, improving the accuracy of production process management.

[0131] In one embodiment, the health status level of an equipment can be determined based on a health score in the equipment health status feature vector. For example, the health score can be divided into several intervals, each interval corresponding to a health status level. For equipment associated with production tasks, a health score decay factor is calculated based on its health status level. A decay function, such as health score decay factor = 1 - (1 - health score / maximum health score)^preset exponent, can be used to calculate the decay factor.

[0132] S46: Multiply the basic coupling weights by the health score decay factor to generate the final production task coupling feature matrix.

[0133] In this embodiment, the basic coupling weight comprehensively considers the logical dependencies and material supply relationships between production tasks, while the health score decay factor considers the impact of equipment health status on the correlation of production tasks. Multiplying the two allows equipment health factors to be incorporated into the logical and material-based production task correlation analysis, generating the final production task coupling feature matrix. Each element in the matrix represents the coupling strength between a pair of production tasks.

[0134] In one embodiment, the basic coupling weights can be stored in a matrix, and the health score decay factor can be stored in a vector. For each element in the matrix, it is multiplied by the corresponding health score decay factor to obtain the final production task coupling feature matrix.

[0135] In one embodiment, reference Figure 5 Step S50 may include steps S51-S55, which will be described in detail below:

[0136] S51: Obtain the equipment health status feature vector and the production task coupling feature matrix, and determine the mapping rules of the two in the graph respectively.

[0137] In this embodiment, the equipment health status feature vector is a multi-dimensional vector representing the equipment health status, and the production task coupling feature matrix is ​​a matrix describing the degree of correlation between production tasks. The mapping rule refers to how to map the information in the equipment health status feature vector and the production task coupling feature matrix to the node attributes and edge weights of the graph. For example, for the equipment health status feature vector, each element in the vector can be mapped to an attribute value of a graph node; for the production task coupling feature matrix, each element in the matrix can be mapped to the weight value of a graph edge.

[0138] In this embodiment, determining the mapping rules is a crucial step in constructing a multidimensional production state feature map. Accurate mapping rules ensure that the map accurately reflects the global state of the production system.

[0139] In one embodiment, mapping rules can be determined based on the graph's data structure and requirements. For the device health status feature vector, the elements in the vector can be mapped sequentially to the attribute fields of the graph nodes. For the production task coupling feature matrix, the rows and columns of the matrix can be mapped to nodes in the graph, with the element values ​​in the matrix serving as the weights of the edges connecting the two nodes. This mapping relationship can be implemented using dictionaries or mapping tables in a programming language.

[0140] S52: Use the device health status feature vector as a graph node attribute, initialize each node, and generate an initial node set.

[0141] In this embodiment, the device health status feature vector is used as the graph node attribute, meaning that each node represents a production device, and the node's attributes contain the device's health status information. Initialization processing refers to assigning initial attribute values ​​and states to each node. Generating an initial node set is the first step in constructing the graph, laying the foundation for adding edges and building a complete graph.

[0142] In this embodiment, by mapping the device health status feature vector to node attributes and performing initialization processing, the device's health status can be intuitively displayed in the graph. From a technical perspective, this facilitates convenient monitoring and analysis of the device.

[0143] In one embodiment, a graph database or graph processing framework can be used to initialize nodes. First, an empty set of nodes is created. Then, the device health status feature vectors are traversed, creating a node for each vector and using the elements in the vectors as the node's attribute values. For example, when using the graph database Neo4j, nodes can be created using the Cypher language, and the node attribute values ​​can be set.

[0144] S53: Use the production task coupling feature matrix as the weight of the graph edge, initialize each edge, and generate an initial edge set.

[0145] In this embodiment, the elements in the production task coupling feature matrix represent the degree of association between production tasks. Using these elements as weights for graph edges reflects the logical relationships and dependency strength between production tasks. Initialization processing refers to assigning initial weight values ​​and states to each edge. Generating an initial edge set is a crucial step in constructing the graph, connecting the relationships between nodes through edges.

[0146] In this embodiment, by mapping the production task coupling feature matrix to edge weights and performing initialization processing, the relationships between production tasks can be clearly displayed in the graph. Technically, this helps to identify potential bottlenecks and risks in the production system.

[0147] In one embodiment, the connection relationships between nodes in the graph are determined based on the row and column information of the production task coupling feature matrix. For each element in the matrix, an edge connecting the corresponding node is created, and the element value is used as the edge weight. For example, when using the graph processing framework NetworkX, edges can be created and their weights set using Python code.

[0148] S54: Based on the initial set of nodes and the initial set of edges, generate an initial multidimensional production state feature map.

[0149] In this embodiment, the initial node set contains nodes representing production equipment and their attribute information, and the initial edge set contains edges connecting nodes and their weight information. Combining these two sets generates an initial multidimensional production state feature graph. This graph graphically displays the structure and state of the production system, with nodes representing production equipment and edges representing the logical relationships between production tasks.

[0150] In one embodiment, a graph database or graph processing framework can be used to merge the initial set of nodes and the initial set of edges to generate an initial multidimensional production state feature graph. For example, in a graph database, the information of nodes and edges can be imported into the database to form a complete graph structure.

[0151] S55: Perform topology optimization on the initial multidimensional production state feature map to eliminate redundant nodes and edges, and generate the final multidimensional production state feature map.

[0152] In this embodiment, the initial multidimensional production state feature graph may contain some redundant nodes and edges. These redundant elements increase the complexity of the graph and affect the efficiency of analysis and processing. Topology optimization refers to filtering and adjusting the nodes and edges of the graph to remove unnecessary elements. Eliminating redundant nodes and edges can make the graph more concise and clear, highlighting key information.

[0153] In this embodiment, topology optimization of the initial multidimensional production state feature map can improve the quality and usability of the map. From a technical perspective, it facilitates in-depth analysis and optimized scheduling of the production system.

[0154] In one embodiment, graph algorithms can be used for topology optimization. For example, the importance of nodes can be determined by calculating metrics such as degree centrality and betweenness centrality. Nodes with low degree centrality and minimal impact on the graph structure can be considered for deletion. Edges with small weights and minimal impact on production task association can also be deleted. After multiple rounds of filtering and adjustment, the final multidimensional production state feature graph is generated.

[0155] In one embodiment, reference Figure 6 Step S60 may include steps S61-S64, which will be described in detail below:

[0156] S61: Obtain the topological structure information and node attribute information of the multidimensional production state feature map, and parse it to extract the key nodes and key paths in the map.

[0157] In this embodiment, the topological structure information of the multidimensional production state feature map describes the connection relationships between nodes, reflecting the logical dependencies and process sequence between production tasks. Node attribute information includes key information such as the health status of equipment. Parsing the topological structure and node attribute information is to gain a deeper understanding of the operating status of the production system. Critical nodes refer to nodes that play an important role in the production process, such as the node corresponding to bottleneck equipment; the critical path refers to the longest path in the production process, which determines the length of the entire production cycle.

[0158] In one embodiment, graph algorithms can be used to analyze a multidimensional production state feature graph. For example, depth-first search (DFS) or breadth-first search (BFS) algorithms can be used to traverse the graph and obtain topological information. Node attribute information can be extracted directly from the nodes. Key nodes are identified by calculating metrics such as degree and betweenness centrality. The longest path algorithm, such as the Floyd-Warshall algorithm, is used to find the critical path.

[0159] S62: Based on the key nodes and key paths, identify potential bottlenecks in the production process using a preset bottleneck identification algorithm.

[0160] In this embodiment, the preset bottleneck identification algorithm is formulated based on actual production conditions and experience, and is used to determine whether there are potential bottlenecks on critical nodes and critical paths. Potential bottlenecks refer to nodes or links that may limit production efficiency and cause delays in the production process. For example, on the critical path, if the processing capacity of a certain device is insufficient, or the waiting time of a certain production task is too long, it may become a potential bottleneck.

[0161] In this embodiment, potential bottlenecks are identified based on critical nodes and critical paths, which can focus on key parts of the production process and accurately identify factors that may affect production efficiency.

[0162] In one embodiment, the preset bottleneck identification algorithm can comprehensively consider multiple factors. For example, it can calculate the load rate of key nodes, and when the load rate of a node exceeds a certain threshold, the node is determined to be a potential bottleneck. For production tasks on the critical path, their time consumption can be analyzed. If the execution time of a task is too long and has a significant impact on the overall production cycle, the corresponding node is considered a potential bottleneck. In practice, a query statement can be written in the graph database to calculate the load rate and time consumption based on the node's attributes and the edge weights to determine the bottleneck.

[0163] S63: For each potential bottleneck, calculate its risk value using a preset risk assessment algorithm, and prioritize the bottlenecks based on the risk value.

[0164] In this embodiment, the preset risk assessment algorithm is a method for quantifying the degree of risk that potential bottlenecks may pose. The risk value is a quantitative indicator that comprehensively considers multiple factors, such as the probability of equipment failure, the impact of the failure on production schedule, and repair costs. Prioritizing bottlenecks based on their risk values ​​helps determine which bottlenecks require immediate attention, allowing for the rational allocation of resources for resolution.

[0165] In one embodiment, the preset risk assessment algorithm can employ a weighted summation method. A weight is assigned to each factor affecting risk, such as weight 'a' for equipment failure probability, weight 'b' for impact on production schedule, and weight 'c' for repair cost. For each potential bottleneck, the value of each factor is calculated separately, and then the risk value is calculated according to the formula: Risk Value = Equipment Failure Probability * a + Impact on Production Schedule * b + Repair Cost * c. Finally, the bottlenecks are sorted in descending order based on their risk values, with bottlenecks having higher risk values ​​listed first.

[0166] S64: Based on the priority ranking results of the bottleneck points, a preliminary scheduling adjustment instruction set is generated through a global optimization inference algorithm.

[0167] In this embodiment, the global optimization inference algorithm is an algorithm that comprehensively considers the overall situation of the production system to achieve optimal scheduling. Based on the priority ranking of bottleneck points, the algorithm can formulate a series of instructions to adjust the scheduling of production tasks, including adjusting task priorities, reallocating resources, and optimizing production processes. The initial set of scheduling adjustment instructions is a set of instructions designed to solve potential bottleneck problems and improve production efficiency.

[0168] In this embodiment, a preliminary scheduling adjustment instruction set is generated based on the priority ranking result, which can specifically address bottleneck problems in the production process. From a technical perspective, this can optimize production scheduling and improve the overall performance of the production system.

[0169] In one embodiment, the global optimization reasoning algorithm can employ a heuristic search algorithm, such as a genetic algorithm or simulated annealing algorithm. The algorithm takes the priority ranking of bottleneck points as input and combines it with constraints of the production system, such as equipment availability and material supply, to search for the optimal scheduling scheme. For example, using a genetic algorithm, the scheduling scheme for production tasks is encoded as chromosomes, and through selection, crossover, mutation, and other operations, iterative optimization is continuously performed to ultimately generate a preliminary set of scheduling adjustment instructions.

[0170] In one embodiment, reference Figure 7In step S70, the preliminary scheduling adjustment instruction set is optimized and adjusted based on the preset production rule base and resource constraints to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes. This may include steps S71-S75, which will be described in detail below:

[0171] S71: Obtain the set of scheduling adjustment instructions after feasibility verification and conflict resolution, and group them to generate multiple scheduling adjustment subsets.

[0172] In this embodiment, the scheduling adjustment instruction set, after feasibility verification and conflict resolution, has excluded instructions that do not conform to production rules and have resource conflicts. Grouping processing involves classifying these instructions according to certain rules, such as grouping them by production task type, involved equipment, execution time, etc. The generated multiple scheduling adjustment subsets can be more conveniently used for subsequent analysis and processing.

[0173] In this embodiment, grouping the scheduling adjustment instruction set helps improve the efficiency and relevance of instruction processing. From a technical perspective, it allows for a clearer understanding of the characteristics and requirements of different types of instructions.

[0174] In one embodiment, instructions can be grouped according to the type of production task involved. For example, instructions related to stamping tasks can be grouped into one group, and instructions related to welding tasks into another. Conditional statements in programming languages ​​can be used to iterate through the scheduling and adjustment instruction set, allocating them to different subsets based on the task type information in the instructions.

[0175] S72: For each scheduling adjustment subset, determine its corresponding resource requirements and execution priority by combining the preset production rule base and resource constraints.

[0176] In this embodiment, a pre-defined production rule base specifies the rules that must be followed during the production process, and resource constraints limit the quantity and usage of available resources. For each scheduling adjustment subset, these rules and conditions are combined to determine the required resources, such as equipment, personnel, and raw materials, as well as the execution priority. The execution priority can be determined based on factors such as the urgency of the task and its impact on the overall production schedule.

[0177] In one embodiment, for each scheduling adjustment subset, the required quantity and type of resources can be calculated based on the task requirements in the instruction and the production rule base. For example, for a scheduling adjustment subset related to an assembly task, the required number of assembly equipment, workers, and parts is determined based on the task's output and process requirements. Simultaneously, the execution priority is determined based on the task's delivery time and its impact on subsequent production tasks. Priority evaluation algorithms, such as the Analytic Hierarchy Process (AHP), can be used to comprehensively consider multiple factors to determine priorities.

[0178] S73: Based on the resource requirements and execution priorities, allocate corresponding resources to each scheduling adjustment subset using a resource allocation algorithm.

[0179] In this embodiment, the resource allocation algorithm is a method that rationally allocates available resources to various scheduling adjustment subsets based on resource demand and execution priority. This resource allocation algorithm ensures optimal resource utilization and avoids resource waste and shortages. For example, it prioritizes allocating resources to scheduling adjustment subsets with high execution priority, while also considering resource balance and coordination.

[0180] From a technical perspective, it can improve the overall performance and efficiency of the production system.

[0181] In one embodiment, a greedy algorithm can be used for resource allocation. First, the scheduling adjustment subsets are sorted according to execution priority, with higher-priority subsets listed first. Then, resources are allocated to each subset sequentially until all resources are allocated or the needs of the subsets are met. During the allocation process, it is checked whether the resources meet the requirements of the production rule base and resource constraints. If not, the allocation scheme is adjusted.

[0182] S74: Integrate the scheduling adjustment subset after resource allocation to generate a preliminary resource reallocation scheme.

[0183] In this embodiment, the integration process involves merging and organizing the various scheduling adjustment subsets after resource allocation to form a complete resource reallocation scheme. During integration, it is necessary to check the coordination and consistency between the subsets to ensure that resource allocation does not lead to new conflicts or problems. The initial resource reallocation scheme includes the resource allocation details and execution order for each scheduling adjustment subset.

[0184] In this embodiment, the integrated processing of the scheduling adjustment subset can form a unified resource reallocation scheme. From a technical perspective, this can provide a specific operational plan for optimizing the scheduling of production processes.

[0185] In one embodiment, a data structure can be used to store and integrate information about scheduling adjustment subsets. For example, a list or dictionary can be used to record the resource allocation and execution order of each subset. During the integration process, it is checked whether there are resource conflicts or contradictory execution orders between subsets, and adjustments are made if so.

[0186] S75: Optimize and adjust the preliminary resource reallocation scheme to ensure that it meets the balance requirements of production efficiency and resource utilization, and generate the final scheduling adjustment instruction set.

[0187] In this embodiment, optimization involves further analysis and improvement of the initial resource reallocation scheme to achieve the best balance between production efficiency and resource utilization. Production efficiency requires completing the most production tasks in the shortest time, while resource utilization requires the rational use of resources and avoidance of waste. Through optimization, resource allocation and task scheduling are fine-tuned to make the scheme more in line with actual production needs.

[0188] In this embodiment, optimizing and adjusting the initial resource reallocation scheme can improve the overall performance of the production system. From a technical perspective, the generated final scheduling and adjustment instruction set can achieve intelligent management of the production process, improving production efficiency and resource utilization.

[0189] In one embodiment, an optimization algorithm can be used to optimize and adjust the initial resource reallocation scheme. For example, a linear programming algorithm can be used, with production efficiency and resource utilization as objective functions, and production rule base and resource constraints as constraints, to solve for the optimal resource allocation and task scheduling scheme. Based on the optimization results, the initial scheme is adjusted to generate the final scheduling adjustment instruction set.

[0190] In one embodiment, the training process of the production process optimization model of this application may include:

[0191] C1: Obtain historical production data, which includes historical equipment operating parameter data and historical production execution process data.

[0192] In this embodiment, historical production data refers to data generated during the operation of the production system over a past period, serving as a crucial foundation for training the production process optimization model. Historical equipment operating parameter data records the operating status of the equipment at different times, such as temperature, pressure, and speed; historical production execution process data includes work order information, material flow records, production progress, and other information. This data reflects the actual operation of the production system and the execution process of production tasks.

[0193] In this embodiment, acquiring historical production data provides rich samples for model training. From a technical perspective, this helps the model learn the operational patterns and rules of the production system.

[0194] In one embodiment, historical production data can be extracted from the production system's database. For example, historical data can be retrieved from a table storing device operating parameter data, and corresponding data can be retrieved from a table storing production execution process data, using database query statements. A time range can be set, such as retrieving historical data from the past year, to ensure the timeliness and representativeness of the data.

[0195] C2: Perform timestamp alignment and data cleaning operations on the historical production data to generate a standardized historical time-series production data stream.

[0196] In this embodiment, the timestamp alignment operation unifies historical production data from different data sources onto the same time scale, eliminating time discrepancies. The data cleaning operation removes noise, outliers, and missing values ​​from the data, improving data quality. Standardized historical time-series production data streams, after processing, have a unified format and higher quality, making them more suitable for model training.

[0197] In this embodiment, timestamp alignment and data cleaning of historical production data improve data availability and accuracy. Technically, this helps the model learn the characteristics and patterns of the production system more accurately.

[0198] In one embodiment, timestamp alignment can be achieved using interpolation or resampling. For data cleaning, statistical methods can be used to detect outliers, such as calculating the mean and standard deviation of the data, and identifying data points that deviate from the mean by more than a certain multiple of the standard deviation as outliers and replacing them. For missing values, interpolation or imputation based on the statistical characteristics of historical data can be used.

[0199] C3: Based on the standardized historical time-series production data stream, extract the historical equipment health status feature vector and the historical production task coupling feature matrix.

[0200] In this embodiment, the historical equipment health status feature vector is extracted based on historical equipment operating parameter data, reflecting the equipment's health status within a historical time period. The historical production task coupling feature matrix is ​​constructed based on historical production execution process data, describing the degree of correlation between production tasks historically. Extracting these feature vectors and matrices allows historical production data to be transformed into feature representations that the model can process.

[0201] In this embodiment, extracting the feature vector of historical equipment health status and the coupled feature matrix of historical production tasks can provide targeted feature data for model training. Technically, this helps the model learn the key features and relationships of the production system.

[0202] In one embodiment, extracting the historical equipment health status feature vector can employ a similar feature extraction method described above, performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized historical time-series production data stream. Extracting the historical production task coupling feature matrix can be done using a similar method described above, constructed based on work order information and material flow records within the historical production execution process data.

[0203] C4: Input the historical equipment health status feature vector and the historical production task coupling feature matrix into the initial production process optimization model, and train the model using a reinforcement learning algorithm with the unit time production capacity as the reward function.

[0204] In this embodiment, the initial production process optimization model is a model to be trained, used to optimize and schedule the production process. Reinforcement learning is an algorithm that allows an agent to interact with its environment, continuously try different actions, and learn based on reward signals. Unit-time productivity refers to the number of products the production system can complete per unit of time; as a reward function, it guides the model to learn in a direction that improves production efficiency.

[0205] In this embodiment, a reinforcement learning algorithm is used to train the model with the output per unit time as the reward function. This enables the model to find the optimal production scheduling strategy through continuous trial and learning. From a technical perspective, this helps improve the model's optimization scheduling capability.

[0206] In one embodiment, reinforcement learning algorithms such as Deep Q-Networks (DQN) can be used for model training. The historical equipment health status feature vector and the historical production task coupling feature matrix are used as the environmental state input to the model. The model outputs actions (scheduling adjustment instructions) based on the state. A reward value is calculated based on the unit time productivity obtained after the action is executed. The model parameters are updated using the reward value, and training is iteratively continued.

[0207] C5: Based on the reward value generated during training, the initial production process optimization model is iteratively optimized until the reward value converges to above a preset threshold, thereby generating the final production process optimization model.

[0208] In this embodiment, the reward value reflects the effect of the actions taken by the model in each training iteration, i.e., the degree of improvement in production efficiency. Iterative optimization involves continuously adjusting the model's parameters based on the reward value to continuously improve the model's performance. The preset threshold is a reward value limit set according to the actual production situation and model training requirements. When the reward value converges to above the preset threshold, it indicates that the model has learned a better production scheduling strategy.

[0209] In one embodiment, during each training iteration, the gradient of the model parameters is calculated based on the reward value, and the model parameters are updated using an optimization algorithm (such as stochastic gradient descent). This process is repeated until the reward value stabilizes above a preset threshold multiple times consecutively. At this point, the model training is considered to have converged, and the final optimized production process model is generated.

[0210] Accordingly, to better implement the above methods, this application also provides a production process AI management system based on time-series production data flow. For example... Figure 8 As shown, the production process AI management system 80 based on time-series production data flow includes:

[0211] The acquisition module 801 is used to acquire multi-source heterogeneous production data covering the entire production process. The multi-source heterogeneous production data includes real-time collected equipment operating parameter data and production execution process data retrieved from the manufacturing execution system.

[0212] The preprocessing module 802 is used to perform preprocessing operations on the multi-source heterogeneous production data to generate a standardized time-series production data stream.

[0213] The feature construction module 803 is used to construct a feature vector representing the health status of the equipment by performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized time-series production data stream.

[0214] The matrix construction module 804 is used to construct a production task coupling feature matrix that characterizes the correlation of production tasks based on the work order information and material flow records in the production execution process data in the standardized time-series production data stream, as well as the health status feature vector.

[0215] The graph generation module 805 is used to map the device health status feature vector to graph node attributes through a graph construction algorithm, and to map the production task coupling feature matrix to the weight of graph edges, thereby constructing a multi-dimensional production status feature graph that represents the global state of the production system. In this graph, nodes represent generating devices, and edges represent the logical relationships between production tasks.

[0216] The optimization module 806 is used to input the multi-dimensional production state feature map into a preset production process optimization model, analyze the topology and node attributes of the multi-dimensional production state feature map through the production process optimization model, identify potential bottlenecks and risk points in the production process, perform global optimization reasoning, and generate a preliminary scheduling adjustment instruction set.

[0217] The verification module 807 is used to verify the feasibility and resolve conflicts of the preliminary scheduling adjustment instruction set. Combined with the preset production rule base and resource constraints, the preliminary scheduling adjustment instruction set is optimized and adjusted to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, so as to realize intelligent management of the production process.

[0218] The implementation details of each module are provided in the preceding method embodiments and will not be repeated here. The technical effects achieved by each module and device are described in the foregoing method embodiments.

[0219] like Figure 9 As shown, this application embodiment also provides a computer device 90, which includes a processor 901 and a memory 902, wherein the memory 902 stores a computer program, and when the computer program is executed by the processor 901, the processor 901 performs the steps of any of the methods described above.

[0220] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or otherwise.

[0221] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application are still within the scope of this application.

Claims

1. A production process AI management method based on time-series production data flow, characterized in that, The method includes: Acquire multi-source heterogeneous production data covering the entire production process, including real-time collected equipment operating parameter data and production execution process data retrieved from the manufacturing execution system; The multi-source heterogeneous production data is preprocessed to generate a standardized time-series production data stream; By performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized time-series production data stream, a feature vector representing the health status of the equipment is constructed. Based on the work order information and material flow records in the production execution process data of the standardized time-series production data stream, as well as the health status feature vector, a production task coupling feature matrix representing the correlation of production tasks is constructed. The device health status feature vector is mapped to graph node attributes by a graph construction algorithm, and the production task coupling feature matrix is ​​mapped to the weight of graph edge to construct a multi-dimensional production status feature graph representing the global state of the production system. Nodes represent generating devices, and edges represent the logical relationships between production tasks. The multidimensional production state feature map is input into a preset production process optimization model. The production process optimization model analyzes the topology and node attributes of the multidimensional production state feature map, identifies potential bottlenecks and risk points in the production process, performs global optimization reasoning, and generates a preliminary scheduling adjustment instruction set. The feasibility of the preliminary scheduling adjustment instruction set is verified and conflicts are resolved. Combined with the preset production rule base and resource constraints, the preliminary scheduling adjustment instruction set is optimized and adjusted to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, so as to realize intelligent management of production process.

2. The method according to claim 1, characterized in that, Preprocessing the multi-source heterogeneous production data to generate a standardized time-series production data stream includes: Obtain the timestamp of each group of data in the multi-source heterogeneous production data, and calculate the maximum time difference between the timestamps; If the maximum time difference is greater than a preset time threshold, then the multi-source heterogeneous production data is interpolated to fill in the missing data in the time interval. Outliers in the multi-source heterogeneous production data are detected, and the outliers are replaced according to preset outlier processing rules to generate pre-cleaned production data. The pre-cleaned production data is subjected to sliding window filtering to further remove noise interference and generate a standardized time-series production data stream.

3. The method according to claim 1 or 2, characterized in that, By performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized time-series production data stream, a feature vector characterizing the equipment health status is constructed, including: Obtain equipment operating parameter data from the standardized time-series production data stream, and segment the equipment operating parameter data to obtain multiple subsets within a time window; Perform Fast Fourier Transform and Wavelet Transform on the subset of data within each time window to extract feature values ​​in the frequency domain and time domain, respectively. Based on the characteristic values ​​in the frequency domain and time domain, calculate the device health status score within each time window; The device health status scores within all time windows are combined to form a device health status feature vector.

4. The method according to claim 3, characterized in that, Based on the work order information and material flow records within the production execution process data in the standardized time-series production data stream, and the health status feature vector, a production task coupling feature matrix characterizing the correlation of production tasks is constructed, including: Obtain production execution process data from the standardized time-series production data stream, and extract work order information and material flow records from the production execution process data; The logical dependencies between production tasks are determined based on the work order information, and a dependency weight value is assigned to each pair of production tasks based on the task path length, wherein the dependency weight value is inversely proportional to the path length. The material supply relationship between each production task is determined based on the material flow records, and a supply weight value is assigned to each pair of production tasks based on the material inventory saturation. The basic coupling weight is obtained by weighted summing the dependency weight value and the supply weight value. Based on the equipment health status feature vector, calculate the health score decay factor of the equipment associated with the production task; The basic coupling weights are multiplied by the health score decay factor to generate the final production task coupling feature matrix.

5. The method according to claim 4, characterized in that, The feasibility of the preliminary scheduling adjustment instruction set is verified and conflicts are resolved, including: Obtain each instruction from the preliminary scheduling adjustment instruction set, and verify each instruction against the production rules according to the preset production rule library; If the instruction does not comply with the production rules, the instruction will be given a priority evaluation, and the instruction will be modified or deleted based on the priority evaluation result. Resource usage analysis is performed on the modified scheduling adjustment instruction set to determine whether there are resource conflicts. If resource conflicts exist, the instruction order in the scheduling adjustment instruction set is rearranged according to resource constraints, and dynamic resource reallocation is performed based on resource allocation algorithm until the resource conflicts are eliminated.

6. The method according to claim 5, characterized in that, The device health status feature vector is mapped to graph node attributes using a graph construction algorithm, and the production task coupling feature matrix is ​​mapped to the weights of graph edges, thus constructing a multi-dimensional production status feature graph representing the global state of the production system, including: Obtain the feature vector of equipment health status and the feature matrix coupled with production tasks, and determine the mapping rules of the two in the graph respectively; The device health status feature vector is used as the attribute of the graph node. Each node is initialized to generate an initial node set. The production task coupling feature matrix is ​​used as the weight of the graph edge, and each edge is initialized to generate an initial edge set. Based on the initial set of nodes and the initial set of edges, an initial multidimensional production state feature map is generated; The initial multidimensional production state feature map is optimized in terms of topology to eliminate redundant nodes and edges, thereby generating the final multidimensional production state feature map.

7. The method according to claim 6, characterized in that, The multidimensional production state feature map is input into a preset production process optimization model. The production process optimization model analyzes the topology and node attributes of the multidimensional production state feature map, identifies potential bottlenecks and risk points in the production process, performs global optimization reasoning, and generates preliminary scheduling adjustment instructions, including: The topological structure information and node attribute information of the multidimensional production state feature map are obtained and parsed to extract key nodes and key paths in the map. Based on the key nodes and key paths, potential bottlenecks in the production process are identified using a preset bottleneck identification algorithm. For each potential bottleneck, a risk value is calculated using a preset risk assessment algorithm, and the bottlenecks are prioritized based on the risk values. Based on the priority ranking results of the bottleneck points, a preliminary scheduling adjustment instruction set is generated through a global optimization inference algorithm.

8. The method according to claim 7, characterized in that, Based on a pre-defined production rule base and resource constraints, the initial scheduling adjustment instruction set is optimized and adjusted to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, including: Obtain the set of scheduling adjustment instructions after feasibility verification and conflict resolution, and group them to generate multiple scheduling adjustment subsets; For each scheduling adjustment subset, its corresponding resource requirements and execution priority are determined by combining the preset production rule base and resource constraints; Based on the resource requirements and execution priorities, corresponding resources are allocated to each scheduling adjustment subset using a resource allocation algorithm; The scheduling adjustment subset after resource allocation is integrated to generate a preliminary resource reallocation plan; The preliminary resource reallocation scheme is optimized and adjusted to ensure that it meets the balance requirements of production efficiency and resource utilization, and the final scheduling adjustment instruction set is generated.

9. The method according to claim 8, characterized in that, The training process of the production process optimization model includes: Acquire historical production data, which includes historical equipment operating parameter data and historical production execution process data; The historical production data is timestamped and cleaned to generate a standardized historical time-series production data stream. Based on the standardized historical time-series production data stream, extract the historical equipment health status feature vector and the historical production task coupling feature matrix; The historical equipment health status feature vector and the historical production task coupling feature matrix are input into the initial production process optimization model. The model is trained by using the unit time production capacity as the reward function through a reinforcement learning algorithm. Based on the reward value generated during training, the initial production process optimization model is iteratively optimized until the reward value converges to above a preset threshold, thus generating the final production process optimization model.

10. A production process AI management system based on time-series production data flow, characterized in that, The system includes: The acquisition module is used to acquire multi-source heterogeneous production data covering the entire production process. The multi-source heterogeneous production data includes real-time collected equipment operating parameter data and production execution process data retrieved from the manufacturing execution system. The preprocessing module is used to preprocess the multi-source heterogeneous production data to generate a standardized time-series production data stream. The feature construction module is used to construct a feature vector representing the health status of the equipment by performing multi-scale time-frequency domain analysis on the equipment operating parameter data in the standardized time-series production data stream. The matrix construction module is used to construct a production task coupling feature matrix that characterizes the correlation of production tasks based on the work order information and material flow records in the production execution process data in the standardized time-series production data stream, as well as the health status feature vector. The graph generation module is used to map the device health status feature vector to graph node attributes through a graph construction algorithm, and to map the production task coupling feature matrix to the weights of graph edges, thereby constructing a multi-dimensional production status feature graph that represents the global state of the production system. In this graph, nodes represent generating devices, and edges represent the logical relationships between production tasks. The optimization module is used to input the multidimensional production state feature map into a preset production process optimization model, analyze the topology and node attributes of the multidimensional production state feature map through the production process optimization model, identify potential bottlenecks and risk points in the production process, perform global optimization reasoning, and generate a preliminary scheduling adjustment instruction set. The verification module is used to verify the feasibility and resolve conflicts of the preliminary scheduling adjustment instruction set. Combined with the preset production rule base and resource constraints, the preliminary scheduling adjustment instruction set is optimized and adjusted to generate a final scheduling adjustment instruction set containing targeted scheduling adjustment suggestions and resource reallocation schemes, so as to realize intelligent management of production process.

Images