Manufacturing resource dynamic optimization method and system supporting multi-objective tradeoff

Abstract

The invention provides a manufacturing resource dynamic optimization method and system supporting multi-objective tradeoff, and relates to the field of intelligent manufacturing, and the method comprises the steps: distinguishing data flow and execution sequence dependence through process dependence analysis, and determining a process capable of being started in advance based on data completeness; monitoring a resource load to construct a release window; embedding a process window and a resource release window; and calculating a matching distance between a demand feature and a capability feature for an overload process which cannot be embedded, determining a migration mapping relation, and identifying an optimal scheduling moment. According to the invention, the utilization rate of manufacturing resources can be improved, the production time is shortened, and multi-target dynamic balance optimization is realized.

Description

Technical Field

[0001] This invention relates to intelligent manufacturing technology, and more particularly to a method and system for dynamic optimization of manufacturing resources that supports multi-objective trade-offs. Background Technology

[0002] Manufacturing resource optimization scheduling refers to the process of rationally allocating and scheduling various manufacturing resources to achieve multiple objectives, such as maximizing production efficiency and minimizing production costs, while meeting the requirements of manufacturing tasks. Traditional manufacturing resource scheduling methods are mainly based on static programming models, which cannot adapt to dynamically changing manufacturing environments and are difficult to meet the demands of modern manufacturing systems for efficient resource utilization.

[0003] With the deepening of the concept of intelligent manufacturing, dynamic optimization scheduling technology for manufacturing resources has received widespread attention. This technology improves resource utilization and production efficiency by dynamically adjusting resource allocation strategies through real-time monitoring of the status and load of manufacturing resources. Currently, dynamic optimization methods for manufacturing resources mainly include rule-based scheduling methods, simulation-based scheduling methods, and artificial intelligence-based scheduling methods.

[0004] Existing technologies oversimplify the handling of process dependencies, often treating them uniformly and ignoring the distinction between data flow dependencies and execution order dependencies. This leads to unnecessary delays in processes that could have started earlier, reducing overall resource utilization efficiency. Identification methods lack precision, mostly relying on fixed periods or simple thresholds, failing to accurately capture resource load inflection points. This results in inaccurate resource release window identification and hinders fine-grained resource scheduling. When facing manufacturing resource overload, simple load balancing strategies are typically employed, lacking consideration of multi-dimensional matching characteristics such as process precision and processing speed. This leads to low adaptability between migrated processes and resources, preventing the achievement of optimal scheduling under a multi-objective trade-off. Summary of the Invention

[0005] This invention provides a method and system for dynamic optimization of manufacturing resources that supports multi-objective trade-offs, and can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for dynamic optimization of manufacturing resources that supports multiple objective trade-offs, comprising: Obtain the set of manufacturing tasks and the set of manufacturing resources. Perform process dependency analysis on the manufacturing tasks in the set of manufacturing tasks. Divide the process dependencies into data flow dependencies and execution order dependencies. For processes with only data flow dependencies, calculate the data completeness threshold to determine the processes that can be started in advance and generate a decoupled sequence diagram. Load monitoring of manufacturing resource sets is performed, and gradient analysis is used to identify load inflection points as the starting point of resource release windows, thereby constructing a resource release window distribution map. Align the early start time window of the early start process in the decoupled timing diagram with the resource release window distribution diagram. When there is an overlap between the early start time window and the resource release window, embed the early start process into the manufacturing resource corresponding to the overlap interval to generate the window embedding result. For overloaded processes that cannot be embedded, the demand characteristics of the overloaded processes in terms of process accuracy and processing speed are extracted, and the capability characteristics of idle resources in the corresponding dimensions are extracted. By calculating the matching distance between the demand characteristics and the capability characteristics, the migration mapping relationship is determined. The window embedding result and migration mapping relationship are transformed into a process resource state sequence. The coupling strength value of the state sequence is calculated. The optimal scheduling time is identified based on the coupling strength value. The optimal scheduling time is transformed into resource start and stop instructions to control the operation of the manufacturing resource set.

[0007] Process dependency resolution is performed on the manufacturing tasks in the manufacturing task set, distinguishing process dependencies into data flow dependencies and execution order dependencies. For processes with only data flow dependencies, a data completeness threshold is calculated to determine which processes can be started early, and a decoupled sequence diagram is generated, including: Extract the input data requirements and output data content of each process in the manufacturing task set, identify the data flow relationship between processes, mark the data flow relationship with incremental data transmission characteristics as data flow dependency, and mark the data flow relationship with state synchronization constraints as execution order dependency; For processes that only have data flow dependencies, the calculation process of the process is analyzed to identify the data range accessed during the process start-up phase. A mapping matrix is ​​constructed between the data range and the input data requirements. The mapping matrix is ​​decomposed to extract the principal components, and the proportion of the principal components to the total amount of input data is calculated as the benchmark completeness threshold. Obtain historical execution records of the process, extract the data receiving progress sequence and process start time sequence from the historical execution records, embed the two sequences into a multi-dimensional space to construct the execution trajectory, perform manifold recognition on the execution trajectory to obtain the dominant orientation direction, and correct the baseline completeness threshold based on the projection of the dominant orientation direction on the data dimension to generate a dynamic completeness threshold. Monitor the data output progress of the preceding process. When the data output progress reaches the dynamic completeness threshold, determine the process that can be started in advance and the start time. Calculate the time offset between processes according to the main director's direction and mark the time offset on the process node to construct a decoupled sequence diagram.

[0008] Load monitoring is performed on the manufacturing resource set. Gradient analysis is used to identify load inflection points as the starting points of resource release windows. A resource release window distribution map is constructed, including: Collect the load time series of each manufacturing resource in the manufacturing resource set, construct a two-way gradient field of the load time series, and calculate the directional deviation of the two-way gradient field; The load attenuation curve is extracted at the position where the directional deviation exceeds the preset reversal judgment threshold. The load attenuation curve is then subjected to morphological feature extraction to generate an attenuation morphology code. The moment corresponding to the attenuation morphology code is marked as the load turning point and used as the starting point of the resource release window. The decay pattern code is input into the pre-built release trajectory knowledge base for encoding and matching to obtain the historical release trajectory set. The matching degree between the decay pattern code and each historical release trajectory in the historical release trajectory set is calculated. The release completion time of each historical release trajectory is weighted according to the matching degree to obtain the resource release window width. Map the starting point and width of the resource release window for all manufacturing resources to the time axis to construct a coverage matrix. Perform density projection on the coverage matrix to generate a resource release density curve. Perform topological decomposition on the resource release density curve to extract the density peak region and density valley region, and construct a resource release window distribution map.

[0009] The early start time windows of early startable processes in the decoupled timing diagram are aligned with the resource release window distribution diagram. When there is an overlap between the early start time window and the resource release window, the early startable process is embedded into the manufacturing resource corresponding to the overlap interval. The window embedding results include: Project the early start time window and resource release window in the decoupled timing diagram onto a unified time axis, identify time overlap intervals, and extract the process identifiers and resource identifiers corresponding to the time overlap intervals to form a candidate embedding set. Extract process constraints from process identifiers in the candidate embedding set and encode them into a constraint dependency tree structure. Extract processing capabilities from resource identifiers and encode them into a capability supply tree structure. Perform tree node matching between the constraint dependency tree structure and the capability supply tree structure. Identify constraint nodes in the constraint dependency tree structure that are not covered by the capability supply tree structure. Calculate the process fit degree based on the number of uncovered constraint nodes. Extract the predecessor and successor processes from the process identifiers in the candidate embedding set to construct a dependency chain. Propagate the time perturbation generated by the process embedding along the dependency chain to obtain the perturbation influence range. Convert the perturbation influence range into a penalty coefficient to correct the process fit. Based on the corrected process fit, select the binding relationship between process identifiers and resource identifiers from the candidate embedding set to generate window embedding results.

[0010] For overloaded processes that cannot be embedded, the requirement features of the overloaded processes in terms of process accuracy and processing speed are extracted, and the capability features of idle resources in the corresponding dimensions are extracted. By calculating the matching distance between the requirement features and the capability features, the migration mapping relationship is determined, including: The process accuracy and processing speed values ​​of the overload process at different processing stages are collected to form a feature sequence. The feature sequence is constructed into a tree structure according to the processing sequence. Each tree node contains the process accuracy value and processing speed value. The feature transfer coefficient between adjacent processing stages is calculated. A state transition matrix is ​​generated based on the feature transfer coefficient. The feature distribution vector of the process at different times is calculated from the state transition matrix. The frequency domain representation is obtained by performing a Fourier transform on the process feature distribution vector. A projection matrix is ​​constructed using the frequency domain representation. The main frequency component is obtained by performing singular value decomposition on the projection matrix. The main frequency component is cross-correlated with the processing capacity value of idle resources to generate a time-series matching function between processes and resources. The curvature change is calculated for the time-series matching function, the fluctuation amplitude of the curvature change is extracted, the continuous time period with the smallest fluctuation amplitude is identified as the stable matching interval, the coupling degree optimization equation is established within the stable matching interval, and the optimal mapping time is solved by the alternating iterative method. A process resource binding table is constructed based on the optimal mapping time, generating a migration mapping relationship that includes timing constraints.

[0011] The curvature change is calculated for the time-series matching function, the fluctuation amplitude of the curvature change is extracted, and the continuous time interval with the smallest fluctuation amplitude is identified as the stable matching interval. A coupling optimization equation is established within the stable matching interval, and the optimal mapping time is solved using an alternating iterative method. Construct a sampling point sequence for a time-series matching function, adjust the sampling density by adjusting the sampling point spacing, calculate the directional change value of adjacent points in the sampling point sequence to obtain the curvature change, sort the curvature change by time to form a curvature sequence, calculate the fluctuation amplitude from the curvature sequence, and obtain the fluctuation amplitude distribution. Perform interval division operation on the fluctuation amplitude distribution to generate a fluctuation intensity statistical matrix. Extract the local minimum value position from the fluctuation intensity statistical matrix. Determine the time interval boundary based on the local minimum value position. Determine the interval where the value in the fluctuation intensity statistical matrix is ​​less than the mean as the stable matching interval. The process resource coupling degree equation is constructed by using the stable matching interval as the constraint boundary. A coefficient matrix is ​​constructed by the process execution order and resource capacity. The coefficient matrix is ​​split into process constraint subproblems and resource constraint subproblems. The coupling coefficient is updated by alternately solving the two subproblems. The iteration is repeated until the coupling coefficient converges. The optimal mapping time is calculated based on the converged coupling coefficient.

[0012] The window embedding results and migration mapping relationship are transformed into a process resource state sequence. The coupling strength value of the state sequence is calculated, and the optimal scheduling time is identified based on the coupling strength value. The optimal scheduling time is then transformed into resource start / stop instructions. Controlling the operation of the manufacturing resource set includes: Extract process time sequence data from window embedding results, extract resource time sequence data from migration mapping relationship, construct time axis mapping between process time sequence data and resource time sequence data, and calculate the mapping overlap interval to form process resource state sequence; Calculate the time point offset in the process resource state sequence, generate a correction coefficient based on the time point offset, use the correction coefficient to calibrate the state sequence position, and calculate the coupling strength value of the calibrated state sequence at each time point. The coupling strength values ​​are arranged in chronological order to form a strength change curve. The fluctuation amplitude of the strength change curve is calculated. The moment with the largest coupling strength in the time interval with the smallest fluctuation amplitude is selected as the optimal scheduling moment. The optimal scheduling moment is decomposed into the processing preparation stage and the processing execution stage. Staged start and stop instructions are generated, and the operation of the manufacturing resource set is controlled according to the staged start and stop instructions.

[0013] A second aspect of the present invention provides a dynamic optimization system for manufacturing resources that supports multiple objective trade-offs, comprising: The first unit is used to obtain the set of manufacturing tasks and the set of manufacturing resources, perform process dependency parsing on the manufacturing tasks in the set of manufacturing tasks, distinguish process dependencies into data flow dependencies and execution order dependencies, calculate the data completeness threshold for processes with only data flow dependencies to determine the processes that can be started in advance, and generate a decoupled sequence diagram. The second unit is used to monitor the load of the manufacturing resource set, identify the load inflection point as the starting point of the resource release window through gradient analysis, and construct a resource release window distribution map. The third unit is used to align the early start time window of the early start process in the decoupled timing diagram with the resource release window distribution diagram. When there is an overlapping interval between the early start time window and the resource release window, the early start process is embedded into the manufacturing resource corresponding to the overlapping interval to generate the window embedding result. The fourth unit is used to extract the demand characteristics of overloaded processes in terms of process accuracy and processing speed for overloaded processes that cannot be embedded, extract the capability characteristics of idle resources in the corresponding dimensions, and determine the migration mapping relationship by calculating the matching distance between demand characteristics and capability characteristics. The fifth unit is used to convert the window embedding result and migration mapping relationship into a process resource state sequence, calculate the coupling strength value of the state sequence, identify the optimal scheduling time based on the coupling strength value, convert the optimal scheduling time into resource start and stop instructions, and control the operation of the manufacturing resource set.

[0014] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0016] In this embodiment, by parsing the process dependencies of manufacturing tasks, distinguishing between data flow dependencies and execution order dependencies, and determining processes that can be started early based on data completeness thresholds, a decoupled sequence diagram is generated, enabling flexible scheduling of manufacturing tasks. This effectively reduces unnecessary waiting time between processes and improves overall production efficiency. By monitoring the load status of manufacturing resources, the load inflection point is identified as the starting point of the resource release window. The time window for processes that can be started early is time-aligned with the resource release window, making full use of idle resource periods, improving the utilization rate of manufacturing resources, and reducing resource idle costs. For overloaded processes, by extracting process demand characteristics and resource capacity characteristics, the matching distance is calculated to determine the migration mapping relationship, and the optimal scheduling time is identified based on the coupling strength value. This enables dynamic optimization of manufacturing resource allocation, ensuring production quality while balancing production speed and resource utilization efficiency, supporting an effective trade-off between multiple objectives. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a dynamic optimization method for manufacturing resources that supports multiple objective trade-offs, as described in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the process resource migration and mapping logic of an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 This is a flowchart illustrating a dynamic optimization method for manufacturing resources that supports multi-objective trade-offs, as described in an embodiment of the present invention. Figure 1 As shown, the method includes: Obtain the set of manufacturing tasks and the set of manufacturing resources. Perform process dependency analysis on the manufacturing tasks in the set of manufacturing tasks. Divide the process dependencies into data flow dependencies and execution order dependencies. For processes with only data flow dependencies, calculate the data completeness threshold to determine the processes that can be started in advance and generate a decoupled sequence diagram. Load monitoring of manufacturing resource sets is performed, and gradient analysis is used to identify load inflection points as the starting point of resource release windows, thereby constructing a resource release window distribution map. Align the early start time window of the early start process in the decoupled timing diagram with the resource release window distribution diagram. When there is an overlap between the early start time window and the resource release window, embed the early start process into the manufacturing resource corresponding to the overlap interval to generate the window embedding result. For overloaded processes that cannot be embedded, the demand characteristics of the overloaded processes in terms of process accuracy and processing speed are extracted, and the capability characteristics of idle resources in the corresponding dimensions are extracted. By calculating the matching distance between the demand characteristics and the capability characteristics, the migration mapping relationship is determined. The window embedding result and migration mapping relationship are transformed into a process resource state sequence. The coupling strength value of the state sequence is calculated. The optimal scheduling time is identified based on the coupling strength value. The optimal scheduling time is transformed into resource start and stop instructions to control the operation of the manufacturing resource set.

[0021] In one optional implementation, process dependency resolution is performed on the manufacturing tasks in the manufacturing task set, distinguishing process dependencies into data flow dependencies and execution order dependencies. For processes with only data flow dependencies, a data completeness threshold is calculated to determine processes that can be started in advance, and a decoupled sequence diagram is generated, including: Extract the input data requirements and output data content of each process in the manufacturing task set, identify the data flow relationship between processes, mark the data flow relationship with incremental data transmission characteristics as data flow dependency, and mark the data flow relationship with state synchronization constraints as execution order dependency; For processes that only have data flow dependencies, the calculation process of the process is analyzed to identify the data range accessed during the process start-up phase. A mapping matrix is ​​constructed between the data range and the input data requirements. The mapping matrix is ​​decomposed to extract the principal components, and the proportion of the principal components to the total amount of input data is calculated as the benchmark completeness threshold. Obtain historical execution records of the process, extract the data receiving progress sequence and process start time sequence from the historical execution records, embed the two sequences into a multi-dimensional space to construct the execution trajectory, perform manifold recognition on the execution trajectory to obtain the dominant orientation direction, and correct the baseline completeness threshold based on the projection of the dominant orientation direction on the data dimension to generate a dynamic completeness threshold. Monitor the data output progress of the preceding process. When the data output progress reaches the dynamic completeness threshold, determine the process that can be started in advance and the start time. Calculate the time offset between processes according to the main director's direction and mark the time offset on the process node to construct a decoupled sequence diagram.

[0022] A manufacturing task set can be a combination of related processing, assembly, and inspection processes on a production line, with each process having clearly defined input data requirements and output data content. The process of process dependency resolution first extracts the input and output data characteristics of each process and identifies the data flow relationships between processes. Data flow relationship analysis is completed by comparing the mapping relationship between the output data of preceding processes and the input data of subsequent processes. When data exhibits incremental characteristics during its transfer from preceding to subsequent processes, such as when data can be generated and used in batches, the dependency is marked as a data flow dependency. When data transfer requires state synchronization constraints, such as waiting for the preceding process to completely finish before the subsequent process can begin, it is marked as an execution order dependency. Data incremental transfer characteristics are determined by detecting data generation and consumption patterns. Typical cases include data that can be processed in segments, data exhibiting spatial or temporal locality, and data supporting streaming processing. State synchronization constraints typically manifest as resource mutual exclusion, state transition points, quality acceptance, and other scenarios that must be completed sequentially.

[0023] For processes with only data flow dependencies, further analysis of the process's computational flow is needed to identify the actual data range accessed during the startup phase. Process computational flow analysis is achieved through static code analysis or execution path tracing, recording the data items accessed in the initial stage of process startup and their access order, forming a data access feature map. A mapping matrix is ​​constructed by comparing the identified data access range with all input data requirements declared by the process. Rows in the matrix represent data items, columns represent access time points, and matrix element values ​​represent access intensity. Singular value decomposition (SVD) is applied to this mapping matrix for dimensionality reduction, extracting principal components. The proportion of the data item corresponding to the principal component to the total input data is calculated as the baseline completeness threshold. Principal component extraction follows the principle of a cumulative variance contribution rate greater than 85% to ensure the capture of key data dependencies. The baseline completeness threshold represents the minimum data proportion required for process startup, typically set between 40% and 75%, depending on the process's computational characteristics and data dependency structure.

[0024] To improve threshold accuracy, historical execution records of the process are obtained, including a data reception progress sequence and a process start time sequence. The data reception progress is represented as a vector, recording the percentage of data completeness at each moment; the process start time records the actual start time of the process. These two sequences are used as coordinate points in a multi-dimensional space to construct the process execution trajectory. The execution trajectory reflects the temporal correlation between data reception and process start. Nonlinear manifold learning methods, such as isometric mapping or local linear embedding, are applied to the execution trajectory to identify the dominant orientation direction. The dominant orientation direction represents the optimal correlation path between data reception and process start. Based on the projection value of the dominant orientation direction onto the data dimension, the baseline completeness threshold is adjusted to generate a dynamic completeness threshold that adapts to the actual execution characteristics of the process. The dynamic completeness threshold typically fluctuates within 15% above or below the baseline completeness threshold, adaptively adjusting as the data flow characteristics change.

[0025] Continuously monitor the data output progress of preceding processes, with the monitoring frequency set between 100 milliseconds and 5 seconds depending on the process complexity. Data output progress is monitored through database record counting or file system change listening. When the data output progress of a preceding process reaches the dynamic completeness threshold calculated for subsequent processes, it is determined that the subsequent process can start ahead of schedule, and the optimal start time is calculated. The optimal start time is based on the subjective adaptive direction prediction, projecting the data reception progress onto the time dimension to obtain the offset relative to the standard start time. The time offset represents the length of time by which the process can start ahead of schedule, typically 15% to 40% of the original process interval.

[0026] The decoupling sequence diagram generation process represents all processes as nodes, and dependencies as directed edges. For process nodes identified as capable of being started earlier, the calculated time offset is annotated. The time offset annotation uses a relative time format, representing the proportion of time that can be advanced relative to the strictly sequential execution. The decoupling sequence diagram presents the optimized execution sequence between processes in a visual manner, supporting reference execution by the production scheduling system.

[0027] In this embodiment, by accurately distinguishing the dependency types between processes, the optimization space in data flow dependencies is identified, enabling partial parallel execution of manufacturing processes and reducing waiting time during production. Dynamic calculation of the data completeness threshold ensures the feasibility and safety of starting processes in advance. Execution trajectory analysis based on historical data allows for adaptive adjustment of process initiation strategies, improving the manufacturing process's responsiveness to changes. The generation of decoupled sequence diagrams provides a more precise basis for manufacturing resource scheduling, resulting in more rational resource allocation and reducing resource idleness and production bottlenecks.

[0028] In one optional implementation, load monitoring of the manufacturing resource set is performed, and gradient analysis is used to identify load inflection points as the starting points of resource release windows. A resource release window distribution map is then constructed, including: Collect the load time series of each manufacturing resource in the manufacturing resource set, construct a two-way gradient field of the load time series, and calculate the directional deviation of the two-way gradient field; The load attenuation curve is extracted at the position where the directional deviation exceeds the preset reversal judgment threshold. The load attenuation curve is then subjected to morphological feature extraction to generate an attenuation morphology code. The moment corresponding to the attenuation morphology code is marked as the load turning point and used as the starting point of the resource release window. The decay pattern code is input into the pre-built release trajectory knowledge base for encoding and matching to obtain the historical release trajectory set. The matching degree between the decay pattern code and each historical release trajectory in the historical release trajectory set is calculated. The release completion time of each historical release trajectory is weighted according to the matching degree to obtain the resource release window width. Map the starting point and width of the resource release window for all manufacturing resources to the time axis to construct a coverage matrix. Perform density projection on the coverage matrix to generate a resource release density curve. Perform topological decomposition on the resource release density curve to extract the density peak region and density valley region, and construct a resource release window distribution map.

[0029] This method first collects the load time series of each manufacturing resource in the manufacturing resource set. During the collection process, resource load rate data, including indicators such as CPU utilization, memory usage, and work queue length, are recorded at a preset sampling frequency (e.g., every 5 seconds) to form time series data. The load time series within a day exhibits fluctuating characteristics: the load remains between 75% and 85% during the morning shift, drops to 30%-40% during the lunch break, rises to 80%-90% in the afternoon, and gradually decreases to below 20% in the evening. The forward gradient and backward gradient are calculated for this load time series. The forward gradient represents the rate of change of the current moment relative to future moments, and the backward gradient represents the rate of change of the current moment relative to past moments. The two gradient values ​​are mapped to two-dimensional vectors, and the cosine of the angle between the two vectors is calculated as the directional divergence index. When the directional divergence is close to -1, it indicates that the forward gradient and backward gradient are in opposite directions, signifying a reversal in the load trend.

[0030] The load attenuation curve is extracted at the location where the directional deviation exceeds the preset reversal judgment threshold. Morphological features of the load attenuation curve are extracted to generate an attenuation morphology code. The time corresponding to the attenuation morphology code is marked as the load inflection point, serving as the starting point of the resource release window. The reversal judgment threshold is set to -0.85. When the directional deviation of the CNC machine tool load reaches -0.92 around 17:30, the load inflection point is identified, and the load attenuation curve for the next 30 minutes is extracted from that moment. Morphological features of the attenuation curve are extracted, including initial descent rate, mid-term stability, final attenuation rate, and fluctuation frequency. These features are combined into an attenuation morphology code; for example, "rapid descent - slow fluctuation - rapid descent" can be represented as "101". 17:30 is determined as the starting point of the resource release window for this CNC machine tool.

[0031] The decay pattern code is input into a pre-built release trajectory knowledge base for encoding matching to obtain a set of historical release trajectories. The matching degree between the decay pattern code and each historical release trajectory in the set is calculated. Based on the matching degree, the release completion time of each historical release trajectory is weighted and calculated to obtain the resource release window width. The release trajectory knowledge base stores the load decline patterns and their corresponding release completion times for the past three months. Matching the current decay pattern code "101" with the historical trajectories in the knowledge base reveals five historical records with high similarity, with similarity scores of 0.92, 0.88, 0.85, 0.78, and 0.72, and corresponding release completion times of 45 minutes, 42 minutes, 48 ​​minutes, 40 minutes, and 50 minutes, respectively. A weighted average is calculated based on the similarity score, yielding an estimated release completion time of 44.5 minutes, i.e., a resource release window width of 44.5 minutes.

[0032] A coverage matrix is ​​constructed by mapping the resource release window start point and width of all manufacturing resources to the time axis. Density projection is then applied to the coverage matrix to generate resource release density curves. Topological decomposition of these curves extracts density peak and valley regions, thus constructing a resource release window distribution map. For example, a factory contains 15 CNC machine tools, 8 industrial robots, and 5 sets of automatic detection systems. The resource release window start point and width for each are calculated using the aforementioned method. These time periods are mapped onto a 24-hour time axis to form a coverage matrix, where each cell indicates whether a resource is within its release window at that time. For each time point, the number of resources within the release window is calculated, generating a resource release density curve. Utilizing the topological properties of the density curve, time periods with density values ​​above a 70% threshold are extracted as density peak regions, and time periods below a 30% threshold are extracted as density valley regions. Finally, a resource release window distribution map is generated, with 12:00-13:30 and 17:30-19:00 representing peak release density periods, and 20:30-22:00 representing low release density periods.

[0033] Based on information from the resource release window distribution map, and combined with production plans and resource demands, resource allocation is dynamically optimized. In peak density areas, a large release of resources creates a release surge; pre-planning can prevent resource contention. In trough density areas, resource release is sparse; critical tasks can be scheduled to secure resources. During the peak release period from 17:30 to 19:00, some low-priority tasks can be postponed until after 20:30 to avoid conflicts with resource releases during peak hours. During the midday peak period from 12:00 to 13:30, equipment maintenance and system update tasks can be scheduled, utilizing the brief downtime to complete necessary maintenance work.

[0034] By implementing the above method, the load inflection points and release windows of manufacturing resources can be accurately identified, providing a time-dimensional decision-making basis for dynamic resource optimization. This method accurately captures the points of change in load trends through bidirectional gradient analysis of load time series; predicts the resource release completion time through attenuation morphology feature encoding and historical release trajectory matching; and reveals the temporal distribution pattern of manufacturing resource release by constructing a resource release window distribution map. This dynamic resource optimization method based on load monitoring and release window analysis improves the utilization efficiency of manufacturing resources, reduces resource competition conflicts, optimizes task scheduling strategies, and supports multi-objective trade-offs in the production process.

[0035] In one optional implementation, the early start time window of the early startable process in the decoupled timing diagram is time-aligned with the resource release window distribution diagram. When there is an overlap between the early start time window and the resource release window, the early startable process is embedded into the manufacturing resource corresponding to the overlap interval. The window embedding result includes: Project the early start time window and resource release window in the decoupled timing diagram onto a unified time axis, identify time overlap intervals, and extract the process identifiers and resource identifiers corresponding to the time overlap intervals to form a candidate embedding set. Extract process constraints from process identifiers in the candidate embedding set and encode them into a constraint dependency tree structure. Extract processing capabilities from resource identifiers and encode them into a capability supply tree structure. Perform tree node matching between the constraint dependency tree structure and the capability supply tree structure. Identify constraint nodes in the constraint dependency tree structure that are not covered by the capability supply tree structure. Calculate the process fit degree based on the number of uncovered constraint nodes. Extract the predecessor and successor processes from the process identifiers in the candidate embedding set to construct a dependency chain. Propagate the time perturbation generated by the process embedding along the dependency chain to obtain the perturbation influence range. Convert the perturbation influence range into a penalty coefficient to correct the process fit. Based on the corrected process fit, select the binding relationship between process identifiers and resource identifiers from the candidate embedding set to generate window embedding results.

[0036] First, the advance start time windows and resource release windows in the decoupling timing diagram are projected onto a unified time axis to identify overlapping time intervals. The corresponding process identifiers and resource identifiers are then extracted to form a candidate embedding set. Specifically, in the decoupling timing diagram, the advance start time window for a CNC milling process W1 is 9:30-10:15, while the resource release window distribution diagram shows that the resource release window for CNC milling machine R1 is 9:20-10:00, and the resource release window for CNC milling machine R2 is 9:45-10:30. After projecting these time windows onto the same time axis, it is identified that process W1 overlaps with resource R1 in the 9:30-10:00 interval and with resource R2 in the 9:45-10:15 interval.

[0037] Extract the process identifier W1 and resource identifiers R1 and R2 corresponding to the overlapping intervals to form a candidate embedding set {(W1, R1, 9:30-10:00), (W1, R2, 9:45-10:15)}, indicating that process W1 can be embedded in the time period of 9:30-10:00 of resource R1 or the time period of 9:45-10:15 of resource R2.

[0038] Process constraints are extracted from process identifiers in the candidate embedding set and encoded into a constraint dependency tree structure. Machining capabilities are extracted from resource identifiers and encoded into a capability supply tree structure. Tree node matching is performed between the constraint dependency tree structure and the capability supply tree structure. Constraint nodes in the constraint dependency tree structure that are not covered by the capability supply tree structure are identified, and the process fit is calculated based on the number of uncovered constraint nodes. The process constraints for process W1 include accuracy requirements, tool type, cooling method, and feed rate. A constraint dependency tree structure is constructed, with process W1 as the root node, the first-level nodes representing the main constraint categories, and the second-level nodes representing the specific constraint parameters. The milling process W1 requires a machining accuracy ≤0.01mm, uses a ball end mill, employs oil cooling, and has a feed rate ≤200mm / min. Resource R1 has a machining capability of 0.008mm accuracy, supports ball end mills and flat end mills, offers both oil and air cooling, and has a maximum feed rate of 250mm / min. Resource R2 has a machining capability of 0.015mm accuracy, supports ball end mills, offers air cooling, and has a maximum feed rate of 300mm / min. The resource capabilities are encoded as a capability supply tree structure, and node matching is performed with the constraint dependency tree structure. Resource R1 completely covers the constraints of process W1, with 0 uncovered constraint nodes; however, resource R2's accuracy and cooling method do not meet the requirements, resulting in 2 uncovered constraint nodes. Based on the number of uncovered constraint nodes, the process fit between process W1 and resource R1 is calculated to be 1.0, and the process fit between process W1 and resource R2 is calculated to be 0.6.

[0039] For the process identifiers in the candidate embedding set, the predecessor and successor processes are extracted to construct a dependency chain. The time perturbation generated by the process embedding is propagated along the dependency chain to obtain the perturbation's influence range. This influence range is then converted into a penalty coefficient to correct the process fit. Based on the corrected process fit, the binding relationship between process identifiers and resource identifiers is selected from the candidate embedding set to generate the window embedding result. The predecessor process of process W1 is blank preparation process W0, and the successor processes are surface treatment process W2 and quality inspection process W3. According to the original plan, process W1 starts at 10:15 and finishes at 11:30; process W2 starts at 11:40; and process W3 starts at 11:45. If process W1 is embedded into the 9:30-10:00 time window of resource R1, W1 can start 45 minutes earlier and is expected to finish at 10:45, allowing subsequent processes W2 and W3 to start 55 minutes and 60 minutes earlier, respectively. The disturbance affects two processes, both of which are positively advanced, resulting in a penalty coefficient of 0.8. The corrected process fit is 1.0 × 0.8 = 0.8. If process W1 is embedded into the 9:45-10:15 time window of resource R2, W1 can start 30 minutes earlier and is expected to complete at 11:00, advancing subsequent processes by approximately 40 minutes. The disturbance also affects two processes, resulting in a penalty coefficient of 0.8. The corrected process fit is 0.6 × 0.8 = 0.48. Comparing the corrected process fit of the two schemes, the binding relationship between process W1 and resource R1 within the 9:30-10:00 time window is selected, and the window embedding result is generated.

[0040] The window embedding result, when applied in actual production, needs to be dynamically adjusted based on resource preparation and task execution status. Resource R1 requires necessary cleanup and tool preparation before embedding into process W1, which takes approximately 15 minutes. In actual execution, the window embedding result is translated into specific scheduling instructions: resource R1 preparation begins at 9:15, and process W1 starts at 9:30. Simultaneously, the execution status of process W1 is monitored; if any anomalies are detected, a re-evaluation of the embedding scheme can be triggered. During the execution of process W1, relevant personnel and equipment for processes W2 and W3 are notified in advance to prepare for receiving the task.

[0041] To address uncertainties in the production environment, this method also includes a feedback adjustment mechanism. If the actual execution time of process W1 on resource R1 exceeds expectations, the system will calculate the impact on subsequent processes in real time and optimize the start time of those processes. If an abnormality is detected in the state of resource R1, an alternative solution can be immediately triggered, transferring process W1 to resource R2 for execution. This dynamic adjustment mechanism ensures the robustness of the embedded results and the continuity of the production plan.

[0042] In this embodiment, the feasibility and rationality of process embedding are ensured through time window matching and process constraint evaluation; the relationship between local optimization and global stability is balanced through disturbance impact analysis; and the system's ability to cope with uncertainty is enhanced through dynamic feedback adjustment. This optimization method makes full use of the idle time of resource release windows while ensuring strict execution of process requirements, achieving a dual improvement in manufacturing efficiency and processing quality. Simultaneously, by shortening process waiting time and improving resource utilization, production costs are reduced, and the company's market responsiveness and competitiveness are enhanced.

[0043] like Figure 2 The diagram illustrates the logical flow of process resource migration and mapping in this embodiment.

[0044] In one optional implementation, for overloaded processes that cannot be embedded, the requirement characteristics of the overloaded processes in the dimensions of process accuracy and processing speed are extracted, and the capability characteristics of idle resources in the corresponding dimensions are extracted. The migration mapping relationship is determined by calculating the matching distance between the requirement characteristics and the capability characteristics, including: The process accuracy and processing speed values ​​of the overload process at different processing stages are collected to form a feature sequence. The feature sequence is constructed into a tree structure according to the processing sequence. Each tree node contains the process accuracy value and processing speed value. The feature transfer coefficient between adjacent processing stages is calculated. A state transition matrix is ​​generated based on the feature transfer coefficient. The feature distribution vector of the process at different times is calculated from the state transition matrix. The frequency domain representation is obtained by performing a Fourier transform on the process feature distribution vector. A projection matrix is ​​constructed using the frequency domain representation. The main frequency component is obtained by performing singular value decomposition on the projection matrix. The main frequency component is cross-correlated with the processing capacity value of idle resources to generate a time-series matching function between processes and resources. The curvature change is calculated for the time-series matching function, the fluctuation amplitude of the curvature change is extracted, the continuous time period with the smallest fluctuation amplitude is identified as the stable matching interval, the coupling degree optimization equation is established within the stable matching interval, and the optimal mapping time is solved by the alternating iterative method. A process resource binding table is constructed based on the optimal mapping time, generating a migration mapping relationship that includes timing constraints.

[0045] First, the process accuracy and processing speed values ​​of the overload process at different machining stages are collected to form a feature sequence. Specifically, for a precision bearing inner ring turning process, the complete machining process includes three stages: rough turning, semi-finish turning, and finish turning. In the rough turning stage, the process accuracy requirement is 0.05 mm, and the processing speed is 120 mm / min; in the semi-finish turning stage, the process accuracy requirement is 0.02 mm, and the processing speed is 80 mm / min; in the finish turning stage, the process accuracy requirement is 0.005 mm, and the processing speed is 40 mm / min. The feature values ​​of these three stages are constructed into a tree structure according to the machining sequence, with the rough turning stage as the root node, connecting to the semi-finish turning node, and the semi-finish turning node connecting to the finish turning node. The characteristic transfer coefficients between adjacent nodes are calculated. The process accuracy transfer coefficient from rough turning to semi-finish turning is 0.4 (0.02 / 0.05), and the processing speed transfer coefficient is 0.67 (80 / 120). The process accuracy transfer coefficient from semi-finish turning to finish turning is 0.25 (0.005 / 0.02), and the processing speed transfer coefficient is 0.5 (40 / 80). Based on these transfer coefficients, a state transition matrix is ​​constructed with a dimension of 2×2, representing the transition relationship between process accuracy and processing speed. Using the state transition matrix and initial eigenvalues, the characteristic distribution vector of the process at any time during processing is calculated. For example, when processing reaches 25%, the characteristic distribution vector is a process accuracy of 0.043 mm and a processing speed of 110 mm / min; when processing reaches 60%, the characteristic distribution vector is a process accuracy of 0.015 mm and a processing speed of 65 mm / min.

[0046] A Fourier transform is performed on the process feature distribution vector to obtain its frequency domain representation. A projection matrix is ​​constructed using this frequency domain representation, and singular value decomposition (SVD) is performed on the projection matrix to obtain the principal frequency components. The principal frequency components are then cross-correlated with the processing capacity values ​​of idle resources to generate a time-series matching function between the process and the resources. A Fourier transform is performed on the sequence of feature distribution vectors of the process throughout the entire processing cycle to obtain its frequency domain representation. The main frequency components of the process accuracy dimension are concentrated in the low-frequency band, indicating that the accuracy requirement changes smoothly with the processing progress; the frequency distribution of the processing speed dimension is relatively dispersed, indicating that speed adjustments are more frequent. A projection matrix is ​​constructed using the frequency domain representation, with a size consistent with the feature dimensions (2×2). Singular value decomposition is performed on the projection matrix to obtain singular values ​​of 0.75 and 0.35, whose corresponding principal frequency components represent the trends in process accuracy and processing speed, respectively. The factory has three idle lathe resources: a conventional lathe, a CNC lathe, and a precision lathe. Their processing capacity is expressed as a time-varying function of process accuracy and processing speed. The machining accuracy range of a conventional lathe is 0.04-0.08 mm, and the processing speed is 80-150 mm / min; the machining accuracy range of a CNC lathe is 0.01-0.05 mm, and the processing speed is 60-120 mm / min; the machining accuracy range of a precision lathe is 0.003-0.02 mm, and the processing speed is 30-80 mm / min. By cross-correlating the main frequency component of the process with the capability parameters of each lathe, three timing matching functions are obtained. The function values ​​represent the degree of matching between the process and resources at different times.

[0047] The curvature change of the temporal matching function is calculated, and the fluctuation amplitude of the curvature change is extracted. The continuous time period with the smallest fluctuation amplitude is identified as the stable matching interval. Within the stable matching interval, a coupling degree optimization equation is established, and the optimal mapping time is solved using an alternating iterative method. The curvature change of the three temporal matching functions is calculated, and the curvature change represents the changing trend of the matching degree. The curvature change of the conventional lathe is smaller in the early stage of machining, smaller in the middle stage of CNC lathe, and smaller in the later stage of precision lathe. The fluctuation amplitude is extracted: the fluctuation amplitude of the conventional lathe is 0.08 in the 0-30% machining progress, the fluctuation amplitude of the CNC lathe is 0.05 in the 35-70% machining progress, and the fluctuation amplitude of the precision lathe is 0.03 in the 75-100% machining progress. Three stable matching intervals are identified: the conventional lathe is suitable for the 0-30% machining progress, the CNC lathe is suitable for the 35-70% machining progress, and the precision lathe is suitable for the 75-100% machining progress. Within their respective stable matching intervals, an optimization equation for the coupling degree between processes and resources is established. The objective function of the equation comprehensively considers the matching degree of process accuracy, the matching degree of processing speed, and the resource switching cost. An alternating iterative method is used to solve this equation, with the initial value set as the midpoint of the interval, an iteration step size of 0.01, a convergence threshold of 0.001, and a maximum number of iterations of 100. The results show that the optimal mapping time is: for a conventional lathe, accepting the process at 15% of the machining progress; for a CNC lathe, accepting the process at 48% of the machining progress; and for a precision lathe, accepting the process at 82% of the machining progress.

[0048] Based on the optimal mapping time, a process resource binding table is constructed, generating a migration mapping relationship with timing constraints. The process resource binding table contains three binding relationships: processing progress 0-15% is completed by the original resource; processing progress 15-48% is completed by a conventional lathe; processing progress 48-82% is completed by a CNC lathe; and processing progress 82-100% is completed by a precision lathe. To ensure processing continuity, timing constraints are added at resource switching times: the conventional lathe must be ready when the process reaches 15% progress; the CNC lathe must be ready when the process reaches 48% progress; and the precision lathe must be ready when the process reaches 82% progress. Simultaneously, the preparation time for each switch is calculated: the conventional lathe needs 5 minutes of preparation time, the CNC lathe needs 8 minutes, and the precision lathe needs 12 minutes. Combining this information, a complete migration mapping relationship is generated, including the correspondence between process segments and resources, the precise timing of the switch, and the necessary preparatory work. This migration mapping relationship will be imported into the Manufacturing Execution System (MES) to guide resource scheduling and process execution in the actual production process.

[0049] To ensure the effectiveness of the migration mapping, a real-time monitoring mechanism is also required. During the process, actual processing parameters, including real-time accuracy and speed values, are continuously collected and compared with the expected feature distribution vector. If the deviation exceeds a preset threshold, the mapping relationship is dynamically adjusted, updating the optimal mapping time to ensure that process requirements are met and production efficiency is improved.

[0050] In this embodiment, based on the refined analysis of process characteristics and the precise matching of resource capabilities, the process can dynamically migrate to the most suitable manufacturing resources at different processing stages, avoiding the inefficient use of high-precision resources and the overload of low-precision resources. By expressing the feature sequence in a tree structure and constructing the state transition matrix, the changing patterns of process requirements during processing are accurately captured; by frequency domain analysis and singular value decomposition, the main changing patterns of process characteristics are identified; by extracting the temporal matching function and stable matching interval, the optimal matching of process requirements and resource capabilities is achieved; and by using a migration mapping relationship that includes temporal constraints, the continuity and stability of the production process are ensured.

[0051] In one optional implementation, the curvature change of the time-series matching function is calculated, the fluctuation amplitude of the curvature change is extracted, the continuous time interval with the smallest fluctuation amplitude is identified as the stable matching interval, the coupling optimization equation is established within the stable matching interval, and the optimal mapping time is solved using an alternating iterative method, including: Construct a sampling point sequence for a time-series matching function, adjust the sampling density by adjusting the sampling point spacing, calculate the directional change value of adjacent points in the sampling point sequence to obtain the curvature change, sort the curvature change by time to form a curvature sequence, calculate the fluctuation amplitude from the curvature sequence, and obtain the fluctuation amplitude distribution. Perform interval division operation on the fluctuation amplitude distribution to generate a fluctuation intensity statistical matrix. Extract the local minimum value position from the fluctuation intensity statistical matrix. Determine the time interval boundary based on the local minimum value position. Determine the interval where the value in the fluctuation intensity statistical matrix is ​​less than the mean as the stable matching interval. The process resource coupling degree equation is constructed by using the stable matching interval as the constraint boundary. A coefficient matrix is ​​constructed by the process execution order and resource capacity. The coefficient matrix is ​​split into process constraint subproblems and resource constraint subproblems. The coupling coefficient is updated by alternately solving the two subproblems. The iteration is repeated until the coupling coefficient converges. The optimal mapping time is calculated based on the converged coupling coefficient.

[0052] Before performing curvature analysis of the temporal matching function, a sampling point sequence needs to be constructed. The sampling density is adjusted by the sampling point spacing, and the directional change value of adjacent points in the sampling point sequence is calculated to obtain the curvature change. The curvature changes are then sorted by time to form a curvature sequence. The fluctuation amplitude is calculated from the curvature sequence to obtain the fluctuation amplitude distribution. Specifically, for the temporal matching function of the precision bearing inner ring turning process and CNC lathe resources, data points are collected within a time window of 0 to 240 minutes. The initial sampling interval is set to 10 minutes. In regions where the matching function value changes rapidly, such as the 60-90 minute and 150-180 minute segments, the sampling interval is reduced to 2 minutes to capture fine changes; in regions with gradual changes, a 10-minute interval is maintained, resulting in a total of 48 sampling points. The directional change value is calculated for adjacent points in the sampling point sequence. The method for calculating the directional change value is to divide the difference in function value between the current point and the previous point by the time interval, and then subtract the difference in function value between the previous point and the point before that, divided by the time interval, to obtain the curvature change of each point. For example, at t=60 minutes, the function value is 0.82; at the previous time t=50 minutes, the function value is 0.79; and at the time before that, t=40 minutes, the function value is 0.77, resulting in a calculated curvature change of 0.005. At t=70 minutes, the function value rapidly rises to 0.88, resulting in a calculated curvature change of 0.025. The curvature changes at all sampling points are arranged chronologically to form a curvature sequence. The standard deviation of the curvature sequence is calculated within a sliding window of 5 adjacent points to obtain the fluctuation amplitude. For example, the fluctuation amplitude is 0.012 in the 30-80 minute interval; 0.008 in the 90-140 minute interval; and 0.028 in the 150-200 minute interval. These fluctuation amplitudes constitute the fluctuation amplitude distribution.

[0053] An interval division operation is performed on the fluctuation amplitude distribution to generate a fluctuation intensity statistical matrix. Local minimum locations are extracted from the fluctuation intensity statistical matrix, and time interval boundaries are determined based on these local minimum locations. Intervals in the fluctuation intensity statistical matrix where values ​​are less than the mean are defined as stable matching intervals. The fluctuation amplitude distribution is divided into 24 intervals, each 10 minutes long. The average fluctuation intensity within each interval is calculated, forming a 24×1 fluctuation intensity statistical matrix. The values ​​of the first 6 intervals in the fluctuation intensity statistical matrix are [0.015, 0.014, 0.012, 0.011, 0.013, 0.018], representing fluctuations from 0 to 60 minutes; the values ​​of the middle 6 intervals are [0.009, 0.008, 0.007, 0.008, 0.009, 0.010], representing fluctuations from 60 to 120 minutes; the values ​​of the subsequent intervals increase sequentially. Local minima were identified from the fluctuation intensity statistical matrix. A local minimum of 0.007 was found in interval 8-10, corresponding to a time period of 80-100 minutes; a local minimum of 0.010 was also found in interval 14-16, corresponding to a time period of 140-160 minutes. Based on these local minimum locations, the boundaries of the time intervals were determined. The mean of the fluctuation intensity statistical matrix was calculated to be 0.014. Continuous intervals with values ​​less than 0.014 in the matrix were identified as stable matching intervals, resulting in two stable intervals: 70-120 minutes and 135-155 minutes.

[0054] A process-resource coupling degree equation is constructed using the stable matching interval as the constraint boundary. A coefficient matrix is ​​built using the process execution order and resource capacity. The coefficient matrix is ​​then decomposed into process constraint subproblems and resource constraint subproblems. The coupling coefficients are updated by alternately solving the two subproblems, and this process is repeated iteratively until the coupling coefficients converge. The optimal mapping time is calculated based on the converged coupling coefficients. Within the stable matching intervals of 70-120 minutes and 135-155 minutes, process-resource coupling degree equations are constructed. The goal of this equation is to maximize the matching degree between processes and resources, while simultaneously satisfying the process execution order and resource capacity constraints.

[0055] For the precision bearing inner ring turning process, there are four sub-processes: roughing, semi-finishing, finishing, and inspection. The corresponding CNC lathe resources have four capability dimensions: power output, tool accuracy, cooling capacity, and measurement accuracy. Based on the process execution sequence and resource capability constraints, a 4×4 coefficient matrix is ​​constructed. Each element of the coefficient matrix represents the degree of matching between a specific sub-process and a specific resource capability; for example, the matching coefficient between the roughing sub-process and power output capability is 0.9, and the matching coefficient with tool accuracy is 0.3. The coefficient matrix is ​​then divided into process constraint sub-problems and resource constraint sub-problems. The process constraint sub-problems consider the continuity of the process execution sequence, while the resource constraint sub-problems consider the balance of resource capacity. An alternating iterative method is used to solve the two sub-problems: the coupling coefficients are initialized to a mean of 0.5; the resource constraint sub-problems are fixed, and the process constraint sub-problems are solved to obtain updated process-related coupling coefficients; the process constraint sub-problems are fixed again, and the resource constraint sub-problems are solved to obtain updated resource-related coupling coefficients. This process is repeated until the change in coupling coefficients is less than the threshold of 0.001 or the maximum number of iterations (50) is reached. After 27 iterations, the coupling coefficients converged to [0.82, 0.75, 0.68, 0.77]. Based on the converged coupling coefficients, the optimal mapping time was calculated to be 95 minutes, which falls within the first stable matching interval.

[0056] The curvature change reflects the trend of the temporal matching function; a large value indicates an unstable matching relationship, while a small value indicates a stable matching relationship. By extracting the time period with the smallest fluctuation amplitude, the most stable time window for process and resource matching can be found, ensuring that migration operations are performed at the optimal time. The idea behind solving the coupling optimization equation using the alternating iterative method is to decompose the complex problem into two relatively simple sub-problems, and gradually approach the global optimum through iterative optimization, avoiding the computational complexity of directly solving large-scale nonlinear equations.

[0057] When implementing this method, it is important to consider the impact of sampling point density on the accuracy of curvature change calculation. Increase the sampling density in regions where the matching function changes drastically, and appropriately reduce the number of sampling points in regions with gradual changes to balance computational accuracy and efficiency. The interval division of the fluctuation amplitude distribution should also be adjusted according to the actual situation; intervals that are too large may mask local variation characteristics, while intervals that are too small may introduce noise interference. During alternating iterations, the iteration step size can be dynamically adjusted according to the convergence speed to accelerate the convergence process.

[0058] In this embodiment, by analyzing curvature changes and extracting fluctuation amplitudes, the most stable time window for matching processes and resources can be accurately identified, reducing scheduling risks. Through the decomposition and alternating optimization of process constraints and resource constraints, multi-objective balance is achieved while ensuring scheduling feasibility. Precise calculation of the optimal mapping time improves the success rate and efficiency of resource migration operations. This method significantly improves the utilization rate of manufacturing resources, reduces processing quality fluctuations and production efficiency declines caused by improper resource matching, and provides an effective tool for manufacturing enterprises to achieve flexible production and intelligent scheduling.

[0059] In one optional implementation, the window embedding result and migration mapping relationship are transformed into a process resource state sequence. The coupling strength value of the state sequence is calculated, the optimal scheduling time is identified based on the coupling strength value, and the optimal scheduling time is transformed into resource start / stop instructions. Controlling the operation of the manufacturing resource set includes: Extract process time sequence data from window embedding results, extract resource time sequence data from migration mapping relationship, construct time axis mapping between process time sequence data and resource time sequence data, and calculate the mapping overlap interval to form process resource state sequence; Calculate the time point offset in the process resource state sequence, generate a correction coefficient based on the time point offset, use the correction coefficient to calibrate the state sequence position, and calculate the coupling strength value of the calibrated state sequence at each time point. The coupling strength values ​​are arranged in chronological order to form a strength change curve. The fluctuation amplitude of the strength change curve is calculated. The moment with the largest coupling strength in the time interval with the smallest fluctuation amplitude is selected as the optimal scheduling moment. The optimal scheduling moment is decomposed into the processing preparation stage and the processing execution stage. Staged start and stop instructions are generated, and the operation of the manufacturing resource set is controlled according to the staged start and stop instructions.

[0060] In practical implementation, the first step is to construct a sequence of process resource states. Process timing data is extracted from the window embedding results, and resource timing data is extracted from the migration mapping relationship. A timeline mapping is constructed between the process timing data and the resource timing data, and the overlapping intervals of the mappings are calculated to form the process resource state sequence. Specifically, the timing data of the precision gear machining process is extracted from the window embedding results, including the roughing process W1 scheduled to execute from 8:00 to 9:30, the semi-finishing process W2 scheduled to execute from 9:40 to 10:50, and the finishing process W3 scheduled to execute from 11:00 to 12:30. Resource timing data of the gear machining equipment is extracted from the migration mapping relationship, including the availability of CNC milling machine R1 from 7:30 to 9:00 and 10:00 to 12:00, and the availability of CNC grinding machine R2 from 8:30 to 11:30. Process timing data and resource timing data are plotted on a unified timeline, and overlapping intervals are identified: W1 and R1 overlap from 8:00 to 9:00, W1 and R2 overlap from 8:30 to 9:30, W2 and R1 have no overlap, W2 and R2 overlap from 9:40 to 10:50, W3 and R1 overlap from 11:00 to 12:00, and W3 and R2 overlap from 11:00 to 11:30. Based on the overlapping interval information, a process-resource status sequence is formed, and the status sequence records the binding status of processes and resources at specific time points. During the period from 8:00 to 9:00, the status is W1-R1; during the period from 8:30 to 9:30, the status is W1-R2; during the period from 9:40 to 10:50, the status is W2-R2; during the period from 11:00 to 11:30, the status may be W3-R1 or W3-R2; during the period from 11:30 to 12:00, the status is W3-R1.

[0061] The time point offset in the process resource state sequence is calculated. Correction coefficients are generated based on these offsets, and the state sequence positions are calibrated using these coefficients. The coupling strength of the calibrated state sequence at each time point is then calculated. Offsets are calculated for key time points in the process resource state sequence; these offsets represent the difference between the actual execution time and the planned time. Production site monitoring data reveals that process W1 typically starts 5 minutes earlier than planned and ends 10 minutes later; process W2 starts 15 minutes later and ends 8 minutes later; and process W3 starts 5 minutes later and ends 12 minutes later. Resource R1 has a preparation time of 15 minutes and a switchover time of 10 minutes; resource R2 has a preparation time of 20 minutes and a switchover time of 12 minutes. Correction coefficients are generated based on these offsets. The calculation of the correction coefficients considers process duration, resource preparation time, and historical deviations. Specifically, it is calculated by dividing the planned process duration by the historical average actual duration and then multiplying by the resource preparation efficiency factor. The preparation efficiency factor for resource R1 is 0.9, and for resource R2 it is 0.85. The correction coefficients for W1-R1 are calculated to be 0.92, W1-R2 to 0.88, W2-R2 to 0.90, W3-R1 to 0.94, and W3-R2 to 0.91. Using these correction coefficients, the process resource state sequences are calibrated, resulting in the following calibrated time intervals: W1-R1 7:55-9:10, W1-R2 8:25-9:40, W2-R2 9:55-11:00, W3-R1 10:55-12:12, and W3-R2 10:55-11:42. The coupling strength value of the calibrated state sequences at each time point is calculated. This coupling strength value reflects the tightness of the matching between the process and resources. The calculation of the coupling strength value comprehensively considers the weighted average of three factors: process complexity, resource capability matching degree, and time window adaptability. Process complexity is determined by process parameters, resource capability matching is determined by the degree of conformity between equipment specifications and process requirements, and time window fit is determined by the overlap ratio between calibration time and resource availability window. The calculated average coupling strength values ​​for each time interval are: W1-R1 = 0.82, W1-R2 = 0.78, W2-R2 = 0.85, W3-R1 = 0.79, and W3-R2 = 0.88.

[0062] The coupling strength values ​​are arranged chronologically to form a strength variation curve. The fluctuation amplitude of the strength variation curve is calculated. The moment with the largest coupling strength within the time interval with the smallest fluctuation amplitude is selected as the optimal scheduling moment. The optimal scheduling moment is decomposed into a processing preparation stage and a processing execution stage, generating staged start and stop instructions. The operation of the manufacturing resource set is controlled according to the staged start and stop instructions. The coupling strength values ​​at each time point after calibration are arranged chronologically to form a strength variation curve.

[0063] Within the overlapping interval, the value with the higher coupling strength among the process resource combinations is selected. The fluctuation amplitude of each time window in the strength variation curve is calculated; the fluctuation amplitude represents the stability of the coupling strength within that window. The fluctuation amplitude is calculated as the difference between the maximum and minimum coupling strength within the window. The time window length is set to 30 minutes, and the sliding step is 5 minutes. The calculated fluctuation amplitudes for each window are: 0.15 for 7:30-8:00, 0.06 for 8:00-8:30, 0.04 for 8:30-9:00, 0.08 for 9:00-9:30, and so on.

[0064] The time intervals with the smallest fluctuation amplitudes were identified as 8:30-9:00 and 10:30-11:00, with fluctuation amplitudes of 0.04 and 0.05 respectively, indicating relatively stable coupling strength during these periods. Within these two time intervals, the moments with the largest coupling strength were found: 0.82 at 8:45 and 0.87 at 10:40. Comparing the coupling strengths at these two moments, 10:40 was selected as the optimal scheduling time. The optimal scheduling time was decomposed into a processing preparation phase and a processing execution phase. The processing preparation phase includes resource preparation and process preparation. For resource R2, the preparation time is 20 minutes, so resource preparation should begin at 10:20; for process W2, the preparation time is 15 minutes, so process preparation should begin at 10:25. The processing execution phase begins at 10:40 and includes three sub-phases: process initiation, monitoring, and completion. Generate phased start / stop instructions: at 10:20, issue the resource R2 preparation instruction; at 10:25, issue the process W2 preparation instruction; at 10:40, issue the process W2 start instruction; and at 11:00, issue the process W2 completion check instruction. Based on these instructions, control the operation of the manufacturing resource set to achieve precise scheduling.

[0065] In practical applications, this method also includes an exception handling mechanism. When a deviation in process execution is detected to exceed a threshold, the coupling strength value and optimal scheduling time are recalculated in real time, and subsequent instructions are dynamically adjusted. For example, if grinding accuracy is found to be substandard during the execution of process W2, the system will extend the execution time and adjust the start time of the subsequent process W3. This method also supports dynamic resource switching. When resource R2 fails, the feasibility of switching to resource R1 can be quickly assessed based on the process resource status sequence, and corresponding switching instructions can be generated.

[0066] In this embodiment, static planning results are transformed into executable control instructions, bridging the gap between planning and execution. Dynamic mapping in the time dimension is achieved through the construction of process resource state sequences. The matching degree between processes and resources is accurately assessed through coupling strength calculation. Stable scheduling intervals are identified through fluctuation amplitude analysis. Phased start / stop instructions ensure the continuity and controllability of the execution process. This dynamic optimization method can adapt to the uncertainties of the manufacturing environment, balance multiple objectives such as production efficiency, processing quality, and resource utilization, reduce production costs, improve product delivery rates, and enhance the flexibility and robustness of the manufacturing system.

[0067] A second aspect of the present invention provides a dynamic optimization system for manufacturing resources that supports multiple objective trade-offs, the system comprising: The first unit is used to obtain the set of manufacturing tasks and the set of manufacturing resources, perform process dependency parsing on the manufacturing tasks in the set of manufacturing tasks, distinguish process dependencies into data flow dependencies and execution order dependencies, calculate the data completeness threshold for processes with only data flow dependencies to determine the processes that can be started in advance, and generate a decoupled sequence diagram. The second unit is used to monitor the load of the manufacturing resource set, identify the load inflection point as the starting point of the resource release window through gradient analysis, and construct a resource release window distribution map. The third unit is used to align the early start time window of the early start process in the decoupled timing diagram with the resource release window distribution diagram. When there is an overlapping interval between the early start time window and the resource release window, the early start process is embedded into the manufacturing resource corresponding to the overlapping interval to generate the window embedding result. The fourth unit is used to extract the demand characteristics of overloaded processes in terms of process accuracy and processing speed for overloaded processes that cannot be embedded, extract the capability characteristics of idle resources in the corresponding dimensions, and determine the migration mapping relationship by calculating the matching distance between demand characteristics and capability characteristics. The fifth unit is used to convert the window embedding result and migration mapping relationship into a process resource state sequence, calculate the coupling strength value of the state sequence, identify the optimal scheduling time based on the coupling strength value, convert the optimal scheduling time into resource start and stop instructions, and control the operation of the manufacturing resource set.

[0068] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0069] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0070] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic optimization method for manufacturing resources that supports multi-objective trade-offs, characterized in that, include: Obtain the set of manufacturing tasks and the set of manufacturing resources. Perform process dependency analysis on the manufacturing tasks in the set of manufacturing tasks. Divide the process dependencies into data flow dependencies and execution order dependencies. For processes with only data flow dependencies, calculate the data completeness threshold to determine the processes that can be started in advance and generate a decoupled sequence diagram. Load monitoring of manufacturing resource sets is performed, and gradient analysis is used to identify load inflection points as the starting point of resource release windows, thereby constructing a resource release window distribution map. Align the early start time window of the early start process in the decoupled timing diagram with the resource release window distribution diagram. When there is an overlap between the early start time window and the resource release window, embed the early start process into the manufacturing resource corresponding to the overlap interval to generate the window embedding result. For overloaded processes that cannot be embedded, the demand characteristics of the overloaded processes in terms of process accuracy and processing speed are extracted, and the capability characteristics of idle resources in the corresponding dimensions are extracted. By calculating the matching distance between the demand characteristics and the capability characteristics, the migration mapping relationship is determined. The window embedding result and migration mapping relationship are transformed into a process resource state sequence. The coupling strength value of the state sequence is calculated. The optimal scheduling time is identified based on the coupling strength value. The optimal scheduling time is transformed into resource start and stop instructions to control the operation of the manufacturing resource set.

2. The method according to claim 1, characterized in that, Process dependency resolution is performed on the manufacturing tasks in the manufacturing task set, distinguishing process dependencies into data flow dependencies and execution order dependencies. For processes with only data flow dependencies, a data completeness threshold is calculated to determine which processes can be started early, and a decoupled sequence diagram is generated, including: Extract the input data requirements and output data content of each process in the manufacturing task set, identify the data flow relationship between processes, mark the data flow relationship with incremental data transmission characteristics as data flow dependency, and mark the data flow relationship with state synchronization constraints as execution order dependency; For processes that only have data flow dependencies, the calculation process of the process is analyzed to identify the data range accessed during the process start-up phase. A mapping matrix is ​​constructed between the data range and the input data requirements. The mapping matrix is ​​decomposed to extract the principal components, and the proportion of the principal components to the total amount of input data is calculated as the benchmark completeness threshold. Obtain historical execution records of the process, extract the data receiving progress sequence and process start time sequence from the historical execution records, embed the two sequences into a multi-dimensional space to construct the execution trajectory, perform manifold recognition on the execution trajectory to obtain the dominant orientation direction, and correct the baseline completeness threshold based on the projection of the dominant orientation direction on the data dimension to generate a dynamic completeness threshold. Monitor the data output progress of the preceding process. When the data output progress reaches the dynamic completeness threshold, determine the process that can be started in advance and the start time. Calculate the time offset between processes according to the main director's direction and mark the time offset on the process node to construct a decoupled sequence diagram.

3. The method according to claim 1, characterized in that, Load monitoring is performed on the manufacturing resource set. Gradient analysis is used to identify load inflection points as the starting points of resource release windows. A resource release window distribution map is constructed, including: Collect the load time series of each manufacturing resource in the manufacturing resource set, construct a two-way gradient field of the load time series, and calculate the directional deviation of the two-way gradient field; The load attenuation curve is extracted at the position where the directional deviation exceeds the preset reversal judgment threshold. The load attenuation curve is then subjected to morphological feature extraction to generate an attenuation morphology code. The moment corresponding to the attenuation morphology code is marked as the load turning point and used as the starting point of the resource release window. The decay pattern code is input into the pre-built release trajectory knowledge base for encoding and matching to obtain the historical release trajectory set. The matching degree between the decay pattern code and each historical release trajectory in the historical release trajectory set is calculated. The release completion time of each historical release trajectory is weighted according to the matching degree to obtain the resource release window width. Map the starting point and width of the resource release window for all manufacturing resources to the time axis to construct a coverage matrix. Perform density projection on the coverage matrix to generate a resource release density curve. Perform topological decomposition on the resource release density curve to extract the density peak region and density valley region, and construct a resource release window distribution map.

4. The method according to claim 1, characterized in that, The early start time windows of early startable processes in the decoupled timing diagram are aligned with the resource release window distribution diagram. When there is an overlap between the early start time window and the resource release window, the early startable process is embedded into the manufacturing resource corresponding to the overlap interval. The window embedding results include: Project the early start time window and resource release window in the decoupled timing diagram onto a unified time axis, identify time overlap intervals, and extract the process identifiers and resource identifiers corresponding to the time overlap intervals to form a candidate embedding set. Extract process constraints from process identifiers in the candidate embedding set and encode them into a constraint dependency tree structure. Extract processing capabilities from resource identifiers and encode them into a capability supply tree structure. Perform tree node matching between the constraint dependency tree structure and the capability supply tree structure. Identify constraint nodes in the constraint dependency tree structure that are not covered by the capability supply tree structure. Calculate the process fit degree based on the number of uncovered constraint nodes. Extract the predecessor and successor processes from the process identifiers in the candidate embedding set to construct a dependency chain. Propagate the time perturbation generated by the process embedding along the dependency chain to obtain the perturbation influence range. Convert the perturbation influence range into a penalty coefficient to correct the process fit. Based on the corrected process fit, select the binding relationship between process identifiers and resource identifiers from the candidate embedding set to generate window embedding results.

5. The method according to claim 1, characterized in that, For overloaded processes that cannot be embedded, the requirement features of the overloaded processes in terms of process accuracy and processing speed are extracted, and the capability features of idle resources in the corresponding dimensions are extracted. By calculating the matching distance between the requirement features and the capability features, the migration mapping relationship is determined, including: The process accuracy and processing speed values ​​of the overload process at different processing stages are collected to form a feature sequence. The feature sequence is constructed into a tree structure according to the processing sequence. Each tree node contains the process accuracy value and processing speed value. The feature transfer coefficient between adjacent processing stages is calculated. A state transition matrix is ​​generated based on the feature transfer coefficient. The feature distribution vector of the process at different times is calculated from the state transition matrix. The frequency domain representation is obtained by performing a Fourier transform on the process feature distribution vector. A projection matrix is ​​constructed using the frequency domain representation. The main frequency component is obtained by performing singular value decomposition on the projection matrix. The main frequency component is cross-correlated with the processing capacity value of idle resources to generate a time-series matching function between processes and resources. The curvature change is calculated for the time-series matching function, the fluctuation amplitude of the curvature change is extracted, the continuous time period with the smallest fluctuation amplitude is identified as the stable matching interval, the coupling degree optimization equation is established within the stable matching interval, and the optimal mapping time is solved by the alternating iterative method. A process resource binding table is constructed based on the optimal mapping time, generating a migration mapping relationship that includes timing constraints.

6. The method according to claim 5, characterized in that, The curvature change is calculated for the time-series matching function, the fluctuation amplitude of the curvature change is extracted, and the continuous time interval with the smallest fluctuation amplitude is identified as the stable matching interval. A coupling optimization equation is established within the stable matching interval, and the optimal mapping time is solved using an alternating iterative method. Construct a sampling point sequence for a time-series matching function, adjust the sampling density by adjusting the sampling point spacing, calculate the directional change value of adjacent points in the sampling point sequence to obtain the curvature change, sort the curvature change by time to form a curvature sequence, calculate the fluctuation amplitude from the curvature sequence, and obtain the fluctuation amplitude distribution. Perform interval division operation on the fluctuation amplitude distribution to generate a fluctuation intensity statistical matrix. Extract the local minimum value position from the fluctuation intensity statistical matrix. Determine the time interval boundary based on the local minimum value position. Determine the interval where the value in the fluctuation intensity statistical matrix is ​​less than the mean as the stable matching interval. The process resource coupling degree equation is constructed by using the stable matching interval as the constraint boundary. A coefficient matrix is ​​constructed by the process execution order and resource capacity. The coefficient matrix is ​​split into process constraint subproblems and resource constraint subproblems. The coupling coefficient is updated by alternately solving the two subproblems. The iteration is repeated until the coupling coefficient converges. The optimal mapping time is calculated based on the converged coupling coefficient.

7. The method according to claim 1, characterized in that, The window embedding results and migration mapping relationship are transformed into a process resource state sequence. The coupling strength value of the state sequence is calculated, and the optimal scheduling time is identified based on the coupling strength value. The optimal scheduling time is then transformed into resource start / stop instructions. Controlling the operation of the manufacturing resource set includes: Extract process time sequence data from window embedding results, extract resource time sequence data from migration mapping relationship, construct time axis mapping between process time sequence data and resource time sequence data, and calculate the mapping overlap interval to form process resource state sequence; Calculate the time point offset in the process resource state sequence, generate a correction coefficient based on the time point offset, use the correction coefficient to calibrate the state sequence position, and calculate the coupling strength value of the calibrated state sequence at each time point. The coupling strength values ​​are arranged in chronological order to form a strength change curve. The fluctuation amplitude of the strength change curve is calculated. The moment with the largest coupling strength in the time interval with the smallest fluctuation amplitude is selected as the optimal scheduling moment. The optimal scheduling moment is decomposed into the processing preparation stage and the processing execution stage. Staged start and stop instructions are generated, and the operation of the manufacturing resource set is controlled according to the staged start and stop instructions.

8. A dynamic optimization system for manufacturing resources supporting multi-objective trade-offs, used to implement the method of any one of claims 1-7, characterized in that, include: The first unit is used to obtain the set of manufacturing tasks and the set of manufacturing resources, perform process dependency parsing on the manufacturing tasks in the set of manufacturing tasks, distinguish process dependencies into data flow dependencies and execution order dependencies, calculate the data completeness threshold for processes with only data flow dependencies to determine the processes that can be started in advance, and generate a decoupled sequence diagram. The second unit is used to monitor the load of the manufacturing resource set, identify the load inflection point as the starting point of the resource release window through gradient analysis, and construct a resource release window distribution map. The third unit is used to align the early start time window of the early start process in the decoupled timing diagram with the resource release window distribution diagram. When there is an overlapping interval between the early start time window and the resource release window, the early start process is embedded into the manufacturing resource corresponding to the overlapping interval to generate the window embedding result. The fourth unit is used to extract the demand characteristics of overloaded processes in terms of process accuracy and processing speed for overloaded processes that cannot be embedded, extract the capability characteristics of idle resources in the corresponding dimensions, and determine the migration mapping relationship by calculating the matching distance between demand characteristics and capability characteristics. The fifth unit is used to convert the window embedding result and migration mapping relationship into a process resource state sequence, calculate the coupling strength value of the state sequence, identify the optimal scheduling time based on the coupling strength value, convert the optimal scheduling time into resource start and stop instructions, and control the operation of the manufacturing resource set.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images