Multi-laser SLM layer vector data multi-source collaborative dynamic distribution method

Abstract

The invention relates to the field of manufacturing and production process management, in particular to a multi-laser SLM layer vector data multi-source collaborative dynamic allocation method. The method comprises the following steps: processing a three-dimensional model according to a production order containing commercial constraints to generate a hierarchical process topology set carrying process rules; fusing the static attribute and the real-time process state of the topology set to generate a dynamic collaborative matching matrix, and dynamically scheduling multiple laser production resources according to the dynamic collaborative matching matrix; the real-time process state is compared with a preset space-time process potential field graph, a correction strategy is called, and closed-loop management of the manufacturing task is achieved. According to the method, the dynamic decision model is constructed, so that the defect that a static distribution strategy cannot cope with dynamic working conditions in the background technology is overcome, intelligent optimal configuration of production resources under the condition that order business constraints are met is realized, and the quality, efficiency and cost effectiveness of complex component manufacturing are improved.

Description

Technical Field

[0001] This invention relates to the field of manufacturing and production process management, specifically to a multi-source collaborative dynamic allocation method for multi-laser SLM layer vector data. Background Technology

[0002] In the context of digital transformation in high-end manufacturing, selective laser melting (SLM) equipment, as a key flexible production resource, relies on efficient operation management and scheduling to maximize return on investment. However, current mainstream manufacturing execution systems or enterprise resource planning systems still generally use traditional linear production models based on static work orders when planning production tasks and scheduling resources for such additive manufacturing equipment.

[0003] The core flaw of this type of traditional scheduling system lies in the limitations of its information processing architecture. The system's decision-making logic is based solely on static geometric data in production orders and preset fixed working hours, essentially representing an information-poor open-loop management model. It lacks the ability to collect, process, and fuse the massive, heterogeneous data streams generated by multiple laser equipment during task execution in real time. These data streams include: real-time sensor data reflecting the physical state of the equipment, dynamic cost data characterizing resource consumption, and scheduling conflict data affecting delivery cycles.

[0004] The inability to integrate this multi-source dynamic data into the decision-making loop has led to a series of management and business challenges. First, production costs become uncontrollable; the system cannot dynamically calculate costs based on real-time energy consumption and potential scrap risks, making it difficult to guarantee project profitability. Second, production cycles become unreliable; the system cannot dynamically reschedule in response to changes in equipment status or production line bottlenecks, resulting in frequent breaches of delivery schedule commitments in service level agreements. Ultimately, this low operational efficiency and inability to anticipate business risks prevent the company from providing accurate dynamic pricing and reliable delivery guarantees for complex, high-value-added manufacturing services, thus losing its core competitive advantage in a fiercely competitive market.

[0005] To address this, a multi-source collaborative dynamic allocation method for vector data in multi-laser SLM layers is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers, which aims to realize dynamic allocation of laser resources and closed-loop risk management during the manufacturing process.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A multi-source collaborative dynamic allocation method for vector data in multi-laser SLM layers includes: Based on the manufacturing task containing the 3D model and associated production order information, the 3D model is sliced ​​to generate initial layer vector data. Multidimensional feature processing is performed through a vector featureization engine to generate a hierarchical process topology set of vector objects carrying independent process rules. By integrating the static attributes and real-time process status data of each vector object in the hierarchical process topology set, the matching degree between each object to be processed and idle production resources is quantitatively calculated to generate a dynamic collaborative matching matrix; and the multi-laser production resources are dynamically scheduled according to the optimal matching value in the dynamic collaborative matching matrix. By analyzing offline simulation and historical process data, a spatiotemporal process potential field diagram is established. This diagram is used to define the ideal process state of each spatial position of the part at the processing time and is associated with process strategies for correcting state deviations. The assigned processing task is executed, the real-time process status is compared with the ideal state defined by the spatiotemporal process potential field diagram, and the correction strategy is invoked to achieve closed-loop management and control of the manufacturing task while meeting the cost budget and delivery cycle.

[0008] Preferably, the multi-dimensional feature processing steps of the vector featureization engine include: performing geometric and topological analysis on the initial layer vector data to identify and mark vector clusters representing the key shapes of the parts; and classifying and constructing the vector clusters into vector objects of different levels based on the results of the geometric and topological analysis and in conjunction with the quality levels defined in the production order information. The vector objects of different levels include: core entity objects for achieving high-efficiency filling, fine contour objects for ensuring dimensional accuracy and surface quality, and multi-laser overlapping objects for ensuring metallurgical bonding strength, thereby constituting the hierarchical process topology set.

[0009] Preferably, the independent process rules carried by the vector object cover multiple dimensions, including: geometric constraints for ensuring forming accuracy, the geometric constraints including limits on wall thickness and aperture; and process parameter constraints for guiding energy input, the process parameter constraints including preset ranges for scanning speed and laser power.

[0010] Preferably, the real-time process status data comes from a multimodal sensing system, which includes: a visual sensing unit for identifying molten pool morphology and powder bed geometric anomalies, and a heat flow sensing unit for monitoring molten pool temperature and thermal field distribution.

[0011] Preferably, the method for constructing the dynamic collaborative matching matrix is ​​as follows: for pairing the vector object to be processed with the idle production resource, a quantified matching degree is calculated according to a multi-objective function; the multi-objective function weightedly integrates the following indicators: risk indicator, used to assess the quantified value of the process risks of local heat accumulation, thermal stress concentration and splash interference that will be caused after assigning the production resource to the vector object; benefit indicator, used to assess the contribution of implementing the indicator to improving the metallurgical quality of the overlap area and ensuring the key quality objectives of fine contour forming accuracy; cost indicator, used to assess the time cost and energy cost required for the production resource to move from its current position to the starting point of the vector object.

[0012] Preferably, to reflect the process priority of different regions, the weights of the multi-objective function are dynamically adjusted according to the hierarchy of the vector object: for multi-laser overlap objects, the weight of the benefit index is set to the highest; for fine contour objects, the weight of the benefit index is given priority; for core entity objects, the cost index that improves processing efficiency is given priority.

[0013] Preferably, the dynamic scheduling steps are a continuous iterative loop process, and the steps are as follows: at the beginning of the scheduling cycle, the dynamic cooperative matching matrix is ​​generated; for the currently idle laser, the corresponding optimal matching vector object is found from the matrix; the processing task of the vector object is assigned to the corresponding laser; when the laser completes the assigned task, the laser state becomes idle, and the processing state of the part is updated. The system immediately triggers the scheduling cycle until all vector objects in the current layer are processed.

[0014] Preferably, the method for establishing the spatiotemporal process potential field diagram includes: performing finite element mesh generation on the three-dimensional model and conducting thermo-mechanical coupling simulation to obtain theoretical distribution data of temperature field, stress field, and phase transformation history of the part at any spatiotemporal node during the entire processing under ideal process conditions; extracting case data from a historical database, wherein the part material type, preset quality level, and geometric complexity classification recorded in the case data are completely consistent with the current manufacturing task, and the case data includes complete sensor process records and final quality inspection results; fusing and calibrating the theoretical distribution data with the case data to generate the spatiotemporal process potential field diagram; the potential field diagram is used to define the ideal process state that each spatial position of the part should reach at the processing moment using a four-dimensional data structure.

[0015] Preferably, the step of invoking the correction strategy specifically includes: when the real-time process state deviates from the ideal state defined by the spatiotemporal process potential field diagram, the system selects and executes a correction action from a preset strategy library. The correction action includes: adjusting the power, scanning speed, and spot size of the currently working laser in real time without changing the scanning path; replanning the scanning order and scanning strategy of subsequent vector objects to be processed in the affected area; temporarily changing the task assignment of idle lasers and prioritizing the handling of critical compensation tasks caused by the state deviation.

[0016] Preferably, the closed-loop management and control steps include: before implementing the correction strategy, the system estimates the impact of the strategy on the total processing time and potential material loss; the estimated impact is compared with the total budget and total working hours initially set for the manufacturing task; the correction strategy is only implemented if the revised estimated total cost and total time are still within the allowable fluctuation range; if they exceed the range, the system marks the deviation as a high-risk event and prompts manual intervention.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This application constructs a hierarchical process topology set containing core entities, fine contours, and overlapping areas through a vector feature generation engine. This enables the data to not only carry geometric information but also pre-define independent process rules and objectives. This fundamentally overcomes the shortcomings of prior art that relies solely on discrete, static geometric information for decision-making. It elevates the dimension of data processing from a single geometric path to a deep definition of manufacturing intent and physical processes, providing a high-quality, high-dimensional decision-making foundation for subsequent dynamic scheduling and closed-loop control.

[0018] 2. This application establishes a spatiotemporal process potential field diagram as an ideal state benchmark and constructs a dynamic collaborative matching matrix with risk, benefit, and cost as indicators. This transforms resource allocation from a passive and rigid task assignment in the background technology to a proactive risk assessment and strategy generation. It can quantitatively assess risks such as local heat accumulation before task assignment, elevating quality control from a post-event remedial mode to a level of in-process intervention and pre-event prevention, significantly enhancing the process robustness and yield of complex component manufacturing.

[0019] 3. This application, by integrating the above content, constructs a fully automated management and control method for the entire process: "data definition (topology set) → state perception (sensor data) → ideal state comparison (potential field diagram) → optimization decision (matching matrix) → closed-loop execution (correction strategy)". This method, through continuous iterative scheduling loops and hierarchical correction strategies, achieves optimal coordinated allocation of multiple laser resources in the spatiotemporal dimension while meeting cost and cycle constraints. This enables the present invention to possess adaptive capabilities in dynamic and nonlinear manufacturing environments, fundamentally solving the deficiency in the prior art where efficiency and quality are difficult to balance. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the structure of a multi-source collaborative dynamic allocation method for vector data in a multi-laser SLM layer according to the present invention; Figure 2 This is a schematic diagram of a multi-source collaborative dynamic allocation method for vector data in a multi-laser SLM layer according to the present invention; Figure 3 This diagram illustrates the construction and decision-making process of the dynamic collaborative matching matrix in a multi-source collaborative dynamic allocation method for multi-laser SLM layer vector data according to the present invention. Detailed Implementation

[0021] 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.

[0022] Example 1: This embodiment provides a specific application of the multi-source collaborative dynamic allocation method for vector data of multiple laser SLM layers. The technical solution of this invention is applied to a large-format SLM additive manufacturing equipment with four sets of lasers and galvanometer systems. The application scenario of this embodiment is to print a large, thin-walled aerospace-grade titanium alloy rectifier casing containing a complex lattice structure. The difficulty of this manufacturing task lies in the fact that the thin-walled lattice structure is extremely sensitive to heat accumulation and thermal stress, and is prone to deformation or cracking. At the same time, the huge part size requires multiple lasers to work together efficiently to ensure a reasonable production cycle, which places extremely high demands on the level of intelligence in task allocation.

[0023] Reference Figure 1 and Figure 2The method in this embodiment first generates a hierarchical process topology set based on the production order information and three-dimensional model contained in the manufacturing task through a vector feature engine; at the same time, the method acquires real-time process status data through a multimodal sensing system, and establishes a spatiotemporal process potential field diagram to guide the processing through offline simulation and historical data analysis.

[0024] Furthermore, when preprocessing a 3D model of a large aerospace structural component, the vector feature engine performs multi-dimensional feature processing as follows: First, geometric and topological analysis is performed on the initial layer vector data to identify and mark vector clusters that characterize the key morphology of the part; second, based on the analysis results and in conjunction with the "aerospace grade" quality level defined in the production order, the vector clusters are classified and constructed into vector objects of different levels. The vector objects of different levels include core entity objects for achieving high-efficiency filling, fine contour objects for ensuring dimensional accuracy and surface quality, and multi-laser overlapping objects for ensuring metallurgical bonding strength, thereby constituting the hierarchical process topology set.

[0025] In one specific implementation of this embodiment, the geometric and topological analysis steps of the vector feature engine specifically involve constructing a kd-tree data structure to spatially index the endpoints of the vector data, thereby quickly querying neighboring vectors within a specified radius and determining the spatial adjacency relationship. The step of classifying vector clusters and constructing them into vector objects of different levels is implemented through a preset set of rules. For example, when the density of the area covered by a vector cluster is greater than a threshold A (e.g., 80%) and it is not adjacent to the outer boundary of a part, it is classified as a core entity object; when it forms a closed loop and its aspect ratio is greater than a threshold B, it is classified as a fine contour object.

[0026] This embodiment, by defining process objects differently, enables the system to execute completely different manufacturing philosophies for different geometric regions of a part within a single layer. It no longer employs a single, compromise-based process parameter, but rather assigns drastically different process objectives to solid parts requiring high efficiency and delicate parts requiring high quality. This ability to decompose objectives based on regional functions is an innovative approach that fundamentally resolves the core contradiction between efficiency and quality in large, complex parts.

[0027] The data implementation methods for hierarchical process topology sets also include: The hierarchical process topology set is constructed as a graphical data structure, where vector objects serve as nodes, and the relationships between objects are represented by edges connecting the nodes. There are two types of topological relationships: the first is spatial adjacency, where an adjacency edge is established between corresponding nodes when the physical distance between two vector objects is less than a preset threshold; the second is process dependency, where a directed dependency edge is established between corresponding nodes when the processing quality of one object is a prerequisite for the successful processing of another. These topological relationships are calculated and stored when the vector feature engine processes the initial data. In the scenario of printing aerospace structural components, when the vector feature engine processes data from one layer, it not only classifies vectors into objects such as "core" and "fine contour," but also constructs a topological graph between them. For example, for a fine reinforcing rib structure, all vector object nodes constituting the structure will have close "spatial adjacency" edges established between them. More importantly, for a "multi-laser-overlapping object," the system analyzes the thermal impact between it and neighboring objects, thus establishing a "process-dependent" edge. This edge may indicate that optimal metallurgical bonding can only be achieved by waiting for the adjacent areas to cool to below 800K before proceeding with the overlap processing. This complete topology map is a crucial input for subsequent scheduling decisions.

[0028] By digitizing the relationship between one vector object and another into a process topology diagram, the decision-making system is no longer facing isolated processing tasks, but rather understanding the structural connectivity of parts and the internal correlation of processes from a macroscopic perspective. This profound understanding of the context enables scheduling decisions to shift from pursuing local optima to pursuing global optima.

[0029] Furthermore, in the above steps, the independent process rules represented by the vector objects are set from multiple dimensions. Specifically, the geometric constraints of the fine contour object define the lower limit of the minimum wall thickness (e.g., 0.5 mm), and the process parameter constraints of the fine contour object define a narrow preset range for scanning speed and laser power to maintain stable energy input; while the process parameter constraints of the core entity object are set to a wider preset range to achieve efficiency.

[0030] This embodiment transforms abstract process specifications and quality requirements into machine-readable and executable rigid constraints by embedding independent process rules into the vector object. This allows manufacturing intentions to be transmitted seamlessly from the design end to the execution end, ensuring that all subsequent automated decisions do not exceed predefined process limits. This shifts the starting point of quality control from passive monitoring during manufacturing to proactive definition during data preparation, significantly enhancing process stability and predictability.

[0031] Furthermore, during the printing process, the real-time process status data is generated by the multimodal sensing system. Specifically, the visual sensing unit uses a high-speed camera to determine the morphology of the molten pool and geometric anomalies such as scratches and defects on the powder bed surface in real time; while the heat flow sensing unit uses a pyrometer to determine the dynamic conditions of the molten pool, such as temperature, cooling rate, and surrounding thermal field distribution. This data from different modalities is fused to form a complete description of the current manufacturing process.

[0032] This embodiment constructs a process sensing system that is far more robust and richer in information dimensions than a single sensor by fusing multimodal sensing data. It can cross-validate information from different physical dimensions, effectively filter noise or false alarms from a single sensor, and form a high-confidence comprehensive judgment on the state of the molten pool, greatly improving the system's accuracy in perceiving complex physical phenomena.

[0033] The steps for fusing real-time process status data specifically include: multimodal data fusion and feature engineering. This process first performs millisecond-level spatial and temporal alignment of the original pixel matrix from the visual sensing unit and the temperature matrix from the thermal sensing unit to form a unified multi-layer data map. Subsequently, the feature extraction module calculates a set of structured, low-dimensional feature vectors from this data map in real time. These feature vectors include: morphological features characterizing molten pool stability; thermodynamic features characterizing thermal cycling history; and anomalous features characterizing the process environment. This final output feature vector is the real-time process status data used to quantify the matching degree. During the actual printing process of the casing, while a laser is scanning, a high-speed camera and a thermal imaging camera synchronously record the dynamic process of the molten pool. In this embodiment, the feature engineering process runs in real time on an edge computing node, rapidly processing and transforming hundreds of megabytes of raw video streams generated per second into a concise feature vector. this Instead of massive amounts of raw pixel data, a feature vector containing rich physical meaning is used to update the risk indicators related to the region in the dynamic collaborative matching matrix in real time, thereby achieving efficient real-time decision-making.

[0034] This embodiment successfully bridges the gap between the numerous and chaotic raw sensor signals and the low-dimensional but highly information-dense decision factors by introducing multimodal data fusion and feature engineering processes. It transforms complex physical phenomena into operational intelligence that machines can understand and apply to computation. This not only enables closed-loop control based on real-time feedback at the computational level, but also significantly improves the accuracy and robustness of subsequent scheduling decisions by refining and reducing noise at the data source.

[0035] Furthermore, referring to Figure 3 During the printing process, regarding the construction method of the dynamic collaborative matching matrix, within a scheduling cycle, the system needs to select the optimal task from among many vector objects to be processed for the currently idle Laser 2. One candidate object is a core entity object (A) located in the central region of the part, and the other is a fine contour object (B) located in a thin-walled lattice structure. The system will pair up (Laser 2 - Object A) and (Laser 2 - Object B) and calculate a quantized matching degree according to a preset multi-objective function. The quantization method for each index in the multi-objective function when calculating the matching degree is as follows: Risk indicators are obtained by analyzing the real-time thermal field distribution map of the area where the object to be processed is located, extracting features such as the average temperature and maximum temperature gradient of the area, and inputting these features into a preset risk assessment function to calculate a normalized risk score. Higher average temperature and temperature gradient correspond to higher risk scores. The revenue metric is obtained by reading the static attributes of the object to be processed, especially its level and preset process sensitivity, and mapping them to a preset revenue lookup table to obtain a quantitative revenue score. Objects with more critical levels and higher process sensitivity correspond to higher revenue scores. The cost index is calculated by directly determining the Euclidean distance required for the laser galvanometer to jump based on the current physical location of idle production resources and the starting position of the object to be processed, and then converting this distance into a quantified time cost score.

[0036] Finally, through computational logic, the above-mentioned profit scores are positively weighted, while the risk scores and cost scores are negatively weighted, thereby calculating the final quantitative matching degree.

[0037] This embodiment transforms the complex scheduling problem into a clear and quantifiable multi-objective optimization task. It abandons the rigid allocation logic based on fixed rules or regional divisions in the background technology, allowing the system to dynamically and comprehensively consider the combined situation of risk, quality and cost for each potential decision, and thus make a highly intelligent resource allocation that fits the current optimal interests.

[0038] Furthermore, to reflect the process priorities of different regions, the weights of the multi-objective function are dynamically adjusted according to the hierarchy of the vector objects. This adjustment mechanism is implemented through a preset weight lookup table, which establishes a mapping relationship between the hierarchy of the vector object and its corresponding weight coefficient vector. For example, when the system evaluates a fine contour object, it calls a preset quality-priority weight vector from the lookup table. This vector can be set to a revenue indicator weight of 0.7, a risk indicator weight of 0.15, and a cost indicator weight of 0.15. When evaluating a core entity object, it calls an efficiency-priority weight vector, which can be set to a revenue indicator weight of 0.2, a risk indicator weight of 0.2, and a cost indicator weight of 0.6. During the weighted calculation, the system first obtains the hierarchy of the object to be evaluated and retrieves the corresponding weight vector from the lookup table to complete the calculation.

[0039] This embodiment tightly couples high-level manufacturing strategies with low-level mathematical optimization by introducing a dynamic weight adjustment mechanism. As a result, the system's decision-making is no longer solely focused on a single objective, but can automatically switch between different modes, such as quality priority and efficiency priority, based on the functional importance of the current processing area. This gives the system a high degree of decision-making flexibility.

[0040] Furthermore, the steps involved in dynamic scheduling constitute a continuously iterative cyclical process. In the scenario mentioned earlier, the system, based on the results calculated from the dynamic collaborative matching matrix, assigns object B, with the optimal matching degree, to laser #2 for processing. Several hundred milliseconds after laser #2 begins its task, laser #4, located in another area of ​​the part, completes its assigned task, and its state changes to idle. At this moment, the system immediately initiates a new scheduling cycle: it first updates the current actual processing status of the entire part; then, based on this new and real-time state, the system reconstructs a new dynamic collaborative matching matrix for all remaining unprocessed objects and all idle lasers, while simultaneously searching for the optimal task option for laser #4. This cyclical process of "completing task operation → updating actual state → reconstructing matrix → finding optimal solution → assigning new task" continues at a considerably high frequency until all vector objects in the current layer have completed their processing.

[0041] This embodiment employs a continuously iterative scheduling loop, ensuring that each allocation decision is based on the latest and most accurate system state. This mechanism of frequent re-decision-making is a key manifestation of the dynamism of this invention. It allows the system to respond instantly to any subtle changes during processing, proactively seize fleeting process opportunities, and quickly re-plan and reschedule for unexpected anomalies, thereby ensuring that the entire manufacturing process remains in an optimal state.

[0042] Furthermore, before the printing task officially begins, the method for establishing the spatiotemporal process potential field diagram is as follows: First, finite element meshing is performed on the 3D model of the aerospace structural component. At the same time, related work on thermo-mechanical coupling simulation is also carried out. The purpose is to obtain data on the theoretical temperature field and stress field distribution of the part at any spatiotemporal node during the entire processing under ideal process conditions. Next, successful case data that match the material type, quality grade, and geometric complexity of the current part are extracted from the historical database. The theoretical simulation data is then fused with the real process data from the successful cases and calibrated to generate the final spatiotemporal process potential field diagram.

[0043] Furthermore, during the execution of assigned processing tasks, the specific operation of the step involving invoking correction strategies is as follows: If a deviation occurs between the real-time process state and the ideal state defined by the spatiotemporal process potential field diagram, the system will immediately select and execute a matching correction action from the pre-set strategy library. For example, if the local temperature exceeds the ideal value, the system will perform a real-time adjustment to the current laser power; if the deviation is significant, it may execute a temporary change to the task assignment of other idle lasers, prioritizing critical compensation tasks, and this correction action is of a higher priority.

[0044] Furthermore, closed-loop management and control specifically involves the following steps: When implementing the aforementioned corrective strategy, the system estimates the impact of the strategy on the total processing time and potential material losses, and then compares the estimates with the initial total budget and total man-hours set for the manufacturing task. Only if the corrected estimated total cost and total time are within the allowable fluctuation range can the system automatically execute the corrective strategy. If they exceed this range, the system will mark this deviation as a high-risk event and immediately send an alarm message to the operator, reminding them to perform manual intervention.

[0045] This embodiment systematically integrates the data definition of hierarchical process topology sets, the optimization decision-making of dynamic collaborative matching matrices, and closed-loop execution based on spatiotemporal process potential field diagrams to construct a fully automated control method. This method overcomes the shortcomings of previous technologies caused by the static nature of decision models and the unpredictability of process risks, providing crucial technical support for achieving unified optimization of quality, efficiency, and cost-effectiveness while meeting order commercial constraints in dynamic and nonlinear manufacturing environments.

[0046] Example 2: This embodiment provides another specific application of the multi-source collaborative dynamic allocation method for multi-laser SLM layer vector data. The technical solution of this invention is applied to the same independently developed large-format four-laser SLM additive manufacturing equipment, but the manufacturing task in this case is a large, high-precision injection mold for the automotive manufacturing industry. This mold contains complex conformal cooling water channels, placing extremely high demands on suppressing part warpage and ensuring final dimensional accuracy. The production order for this manufacturing task clearly defines a strict cost budget and delivery cycle. This embodiment will focus on illustrating the specific implementation methods of this invention in process benchmark establishment, closed-loop correction, and production management.

[0047] Furthermore, in this embodiment, the method for establishing the spatiotemporal process potential field map is as follows: Before the printing task begins, the system's preprocessing module first performs finite element mesh generation on the three-dimensional model of the injection mold and conducts a detailed thermo-mechanical coupling simulation. This simulation aims to obtain the distribution data of the theoretical temperature field, stress field, and phase transformation history of any spatiotemporal node within the mold during the entire processing process, which lasts for several days, under ideal process conditions. Simultaneously, the system automatically retrieves and extracts historical successful case data from the historical process database that best matches the current manufacturing task. This case data includes complete sensor process records and the final three-dimensional scanning quality inspection results. Finally, the system fuses and calibrates the distribution data obtained from the theoretical simulation with process data from real successful cases. For example, it uses real data to correct the prediction deviation of the cooling rate of the conformal water channel region in the simulation model, thereby generating a high-fidelity final spatiotemporal process potential field map defined by a four-dimensional data structure.

[0048] This embodiment introduces prior knowledge of process quality into the real-time control loop by establishing a spatiotemporal process potential field diagram. Background art control systems lack real-time quantitative standards for judging the quality of the current state when executing tasks. The potential field diagram described in this invention provides a high-resolution quality benchmark for the manufacturing process, backed by both physical theory and empirical data. This fundamentally transforms the manufacturing process from an open-loop command-driven mode that executes geometric paths to a closed-loop target optimization mode that actively seeks to approximate the ideal state, providing an objective and accurate basis for all subsequent online corrections and closed-loop control.

[0049] Furthermore, during the printing process of the mold, the step of invoking the correction strategy specifically involves the following: When the equipment prints to a narrow, conformal water channel partition wall of the mold, the multimodal sensing system monitors the real-time process state of that area, which deviates significantly from the ideal state defined by the spatiotemporal process potential field diagram. Specifically, the actual temperature is 30% higher than the ideal temperature, and this deviation lasts for more than 2 seconds. The system determines that this is a serious deviation that could lead to local overheating and deformation. At this point, the system immediately selects and executes a collaborative composite correction action from the preset strategy library. The corrective action comprises three parallel operations: First, the power of laser number 2, currently operating in the area, is adjusted in real time and reduced by 15%; second, the scanning sequence of all subsequent vector objects to be processed within the affected area is immediately replanned, changing from the original short-side filling strategy to a skip-filling strategy to disperse heat; third, the task assignment of the adjacent idle laser number 3 is temporarily changed, ordering it to prioritize a critical compensation task, namely, to perform a low-power, large-spot preheating scan of the adjacent solid area of ​​the overheated region, in order to actively guide heat to conduct to the solid area, thereby rapidly reducing the temperature of the partition wall.

[0050] The correction strategy library and selection mechanism constructed in this embodiment provide the system with a hierarchical response system to address process deviations. Traditional control systems typically employ single and fixed correction methods. This invention, however, can invoke strategies at different levels, ranging from parameter fine-tuning to task replanning and even multi-resource collaborative intervention, based on the severity and type of the deviation. This capability, with multiple correction methods matched to the risk level, ensures that the system's intervention is always precise, appropriate, and efficient, thereby endowing the entire manufacturing process with unprecedented resilience and robustness.

[0051] During the printing process of the mold, the step of invoking the correction strategy specifically includes: The method pre-establishes a structured correction strategy library, which is organized as a multi-level database. The selection logic for the correction strategy is implemented through a decision tree selection model pre-trained with historical data. The input of this model is a multi-dimensional deviation vector between the real-time process state and the ideal state defined by the spatiotemporal process potential field diagram, and its output is the correction strategy to be called from the strategy library.

[0052] The multidimensional deviation vector used as input to the decision tree selection model includes at least the following dimensions: the deviation between real-time temperature and ideal temperature, the deviation between real-time cooling rate and ideal cooling rate, and the deviation between real-time melt pool area and ideal area identified by the vision system. When the mold is printed onto a deep-cavity thin-walled structure, the system detects a deviation between the real-time state of the region and the ideal state of the spatiotemporal process potential field map, forming a deviation vector containing multiple dimensions of information. For example, this vector indicates that the temperature of the current region is 15°C higher than the ideal value, and the cooling rate is 10% lower than the standard. The system immediately inputs this deviation vector into the pre-trained decision tree model. Based on the specific values ​​in the vector, the model performs a series of judgments and branches along the tree nodes. The first node judges whether the temperature deviation is greater than 10°C, and the second node judges whether the melt pool morphology is abnormal. Finally, the deviation vector falls on a leaf node, which explicitly points to the correction strategy to be executed, namely, calling the second-level replanning scan strategy and applying a checkerboard scan pattern to homogenize the thermal field of the region. This instruction is then issued to the execution unit.

[0053] By constructing a structured strategy library and a decision tree-based selection model, the logic of closed-loop correction is elevated from simple rule-based judgment to a more intelligent and systematic level. The hierarchical strategy library enables the system to take measures of varying granularity, from parameter fine-tuning to multi-laser collaborative intervention, based on the severity of the deviation. The application of the decision tree model makes the selection process not only efficient but also provides a clear and interpretable decision path, which is crucial for manufacturing critical components such as precision molds that require high reliability and traceability. This mechanism ensures that the system can match the most reasonable and efficient correction strategy to complex and variable real-time deviations.

[0054] Furthermore, the closed-loop management and control steps are specifically reflected in a key decision-making stage before the aforementioned corrective action is executed. After the system determines that the aforementioned collaborative corrective action needs to be executed, its built-in closed-loop management module will first perform a rapid cost and time impact estimation of the strategy. The estimation results show that executing this set of corrective actions will add approximately 15 minutes of total processing time and corresponding laser operating costs. Subsequently, the module compares this estimated impact with the initial total budget and total man-hours set for this mold manufacturing task. The comparison results show that the added 15 minutes are still within the 8-hour floating time range specified in the production order, and the increased cost does not exceed the 2% redundancy of the budget. Only when the judgment that it is "still within the allowable floating range" is true will the system finally confirm and autonomously execute the corrective strategy. Conversely, if the estimated impact exceeds the range, the system will mark the deviation as a high-risk decision event requiring manual intervention and push alarm information to the operations engineer.

[0055] The closed-loop management steps in this embodiment construct a crucial business objective constraint layer above the technical control loop. This establishes a direct link between the optimization of the machine's physical processes and the factory's production management goals. Background automation systems, lacking this mechanism, may make costly decisions to correct local technical problems. This invention ensures that every autonomous correction by the system has been tested for economic rationality, guaranteeing that technological autonomy always serves macro-level business objectives. This management intelligence layer is the core guarantee for the safe, reliable, and trustworthy deployment of complex automation systems in industrial scenarios.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-source collaborative dynamic allocation of vector data in multi-laser SLM layers, characterized in that, include: Based on the manufacturing task containing the 3D model and associated production order information, the 3D model is sliced ​​to generate initial layer vector data. Multidimensional feature processing is performed through a vector featureization engine to generate a hierarchical process topology set of vector objects carrying independent process rules. By integrating the static attributes and real-time process status data of each vector object in the hierarchical process topology set, the matching degree between each object to be processed and idle production resources is quantitatively calculated, and a dynamic collaborative matching matrix is ​​generated. Dynamic scheduling of multiple laser production resources is performed based on the optimal matching value in the dynamic collaborative matching matrix. By analyzing offline simulation and historical process data, a spatiotemporal process potential field diagram is established. This diagram is used to define the ideal process state of each spatial position of the part at the processing time and is associated with process strategies for correcting state deviations. The assigned processing task is executed, the real-time process status is compared with the ideal state defined by the spatiotemporal process potential field diagram, and the correction strategy is invoked to achieve closed-loop management and control of the manufacturing task while meeting the cost budget and delivery cycle.

2. The multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers according to claim 1, characterized in that, The steps of the vector feature processing engine for multidimensional feature processing include: performing geometric and topological analysis on the initial layer vector data, identifying and marking vector clusters that characterize the key shape of the part; and classifying the vector clusters and constructing them into vector objects of different levels based on the results of the geometric and topological analysis and in conjunction with the quality level defined in the production order information.

3. The multi-source collaborative dynamic allocation method for vector data in a multi-laser SLM layer according to claim 1, characterized in that, The independent process rules carried by the vector object cover multiple dimensions, including: geometric constraints to ensure forming accuracy, which include limits on wall thickness and aperture; and process parameter constraints to guide energy input, which include preset ranges for scanning speed and laser power.

4. The multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers according to claim 1, characterized in that, The real-time process status data comes from a multimodal sensing system, which includes a visual sensing unit for identifying molten pool morphology and powder bed geometric anomalies, and a heat flow sensing unit for monitoring molten pool temperature and thermal field distribution.

5. The multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers according to claim 1, characterized in that, The method for constructing the dynamic collaborative matching matrix is ​​as follows: for pairing vector objects to be processed with idle production resources, a quantified matching degree is calculated based on a multi-objective function; the multi-objective function weightedly integrates the following indicators: risk indicators, used to assess the quantified value of the process risks of local heat accumulation, thermal stress concentration, and splash interference that will be caused by assigning the production resources to the vector objects; and benefit indicators, used to assess the contribution of the execution to improving the metallurgical quality of the overlap area and ensuring the precision of fine contour forming, which are key quality objectives. Cost metrics are used to assess the time and energy costs required to move the production resource from its current location to the starting point of the vector object.

6. The multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers according to claim 5, characterized in that, The weights of the multi-objective function are dynamically adjusted according to the hierarchy of the vector object: for multi-laser overlap objects, the weight of the benefit index is set to the highest; for fine contour objects, the weight of the benefit index is given priority; for core entity objects, the cost index that improves processing efficiency is given priority.

7. The multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers according to claim 1, characterized in that, The dynamic scheduling process is a continuous iterative loop, consisting of the following steps: at the start of the scheduling cycle, the dynamic collaborative matching matrix is ​​generated; for the currently idle laser, the corresponding optimal matching vector object is found from the matrix; the processing task of the vector object is assigned to the corresponding laser; when the laser completes the assigned task, the laser state becomes idle, and the processing state of the part is updated, and the system immediately triggers the scheduling cycle until all vector objects in the current layer are processed.

8. The multi-source collaborative dynamic allocation method for vector data in a multi-laser SLM layer according to claim 1, characterized in that, The method for establishing the spatiotemporal process potential field diagram includes: performing finite element mesh generation on the three-dimensional model and conducting thermo-mechanical coupling simulation to obtain theoretical distribution data of temperature field, stress field, and phase transformation history of the part at any spatiotemporal node during the entire processing under ideal process conditions; extracting case data from a historical database, wherein the part material type, preset quality level, and geometric complexity classification recorded in the case data are completely consistent with the current manufacturing task, and the case data includes complete sensor process records and final quality inspection results; fusing and calibrating the theoretical distribution data with the case data to generate the spatiotemporal process potential field diagram; the spatiotemporal process potential field diagram is used to define the ideal process state that the part should reach at each spatial position at the processing moment using a four-dimensional data structure.

9. The multi-source collaborative dynamic allocation method for vector data in a multi-laser SLM layer according to claim 1, characterized in that, The steps of invoking the correction strategy specifically include: when the real-time process state deviates from the ideal state defined by the spatiotemporal process potential field diagram, the system selects and executes a correction action from the preset strategy library. The correction action includes: adjusting the power, scanning speed, and spot size of the currently working laser in real time without changing the scanning path; replanning the scanning order and scanning strategy of subsequent vector objects to be processed in the affected area; temporarily changing the task assignment of idle lasers and prioritizing the handling of critical compensation tasks caused by the state deviation.

10. The multi-source collaborative dynamic allocation method for vector data of multi-laser SLM layers according to claim 1, characterized in that, The closed-loop management and control steps include: before implementing the correction strategy, the system estimates the impact of the correction strategy on the total processing time and potential material loss; the estimated impact is compared with the total budget and total time initially set for the manufacturing task; the correction strategy is only implemented if the revised estimated total cost and total time are still within the allowable fluctuation range; if they exceed the range, the system marks the deviation as a high-risk event and prompts manual intervention.

Images