Material cutting method, system, storage medium and program product

Abstract

The application relates to the technical field of computers and discloses a material cutting method, a material cutting system, a storage medium and a program product, the material cutting method comprising the following steps: receiving geometric data and material attribute data of a cutting piece, generating a layout scheme meeting non-overlapping constraints, texture direction constraints and flaw avoidance constraints based on a nested layout model; constructing a process knowledge graph comprising historical process data, material characteristic parameters and equipment state parameters; generating a process parameter set for the cutting piece through the process knowledge graph, a vector embedding model and a graph neural network inference; controlling a cutting device to perform a cutting operation according to the layout scheme and the process parameter set, and collecting a multi-source sensing data stream in real time; performing abnormality identification and compensation amount prediction on the multi-source sensing data stream based on a target convolutional neural network model, and generating a process compensation instruction; adjusting the cutting parameters in response to the process compensation instruction, and feeding back execution data to the process knowledge graph, so that the cutting precision and efficiency of flexible materials can be improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically to material cutting methods, systems, storage media, and program products. Background Technology

[0002] In related technologies, the cutting of flexible materials often relies on manual experience, which results in insufficient cutting accuracy and low efficiency. Summary of the Invention

[0003] This invention provides a material cutting method, system, storage medium, and program product to solve the problems of insufficient cutting accuracy and low efficiency in the cutting of flexible materials, which rely heavily on manual experience in related technologies.

[0004] In a first aspect, the present invention provides a material cutting method, comprising: receiving geometric data and material property data of a cut piece; generating a layout scheme based on a nested layout model that satisfies non-overlapping constraints, texture direction constraints, and defect avoidance constraints; constructing a process knowledge graph including historical process data, material characteristic parameters, and equipment status parameters; generating a set of process parameters for the cut piece through the process knowledge graph, vector embedding model, and graph neural network inference; controlling a cutting device to perform physical cutting operations according to the layout scheme and the set of process parameters, and collecting multi-source sensor data streams in real time; performing anomaly identification and compensation amount prediction on the multi-source sensor data streams based on a target convolutional neural network model, and generating real-time process compensation instructions; adjusting cutting parameters in response to the process compensation instructions, and feeding back the execution data to the process knowledge graph.

[0005] The material cutting method provided in this invention generates a nested nesting model that satisfies multiple constraints such as non-overlap, texture direction, and defect avoidance, effectively improving material utilization while ensuring that the cutting of pieces meets design specifications and quality requirements, reducing material scrap and rework costs caused by constraint violations. By integrating historical process data, material properties, and equipment status through a process knowledge graph, and combining vector embedding and graph neural network inference to generate a set of process parameters, the method achieves precise and intelligent decision-making for process parameters, avoiding the subjective bias and insufficient adaptability of experience-based parameter settings. Through real-time acquisition of multi-source sensor data streams and the use of a target convolutional neural network model for anomaly identification and compensation prediction, the method can quickly capture abnormal states during the cutting process and generate targeted compensation instructions, significantly reducing cutting errors and improving cutting accuracy and process stability. Through a closed-loop mechanism that feeds execution data back to the process knowledge graph, the method continuously iterates and updates the process knowledge base, enhancing its adaptability and robustness under different material types, equipment conditions, and cutting piece requirements, providing comprehensive assurance for the efficient and high-quality operation of the material cutting process.

[0006] In one optional implementation, the geometric data and material property data of the cut pieces are received, and a nested nesting model is used to generate a nesting scheme that satisfies non-overlapping constraints, texture direction constraints, and defect avoidance constraints. This includes: converting the outline polygons, texture direction vectors, defect area polygons, and minimum process spacing of the cut pieces into a set of constraints based on the geometric data and material property data of the cut pieces; and solving for the optimal nesting scheme under the set of constraints based on the nested nesting model, wherein the nested nesting model includes a genetic algorithm and / or a simulated annealing algorithm.

[0007] The material cutting method provided in this invention transforms elements such as the outline of the cut piece, texture direction, defect area, and process spacing into a structured set of constraints. This achieves precise quantification and unified expression of multi-dimensional constraints, effectively avoiding the problems of incomplete constraint consideration and vague expression when relying on manual experience for layout. It provides a solid and rigorous constraint framework for the scientific solution of layout schemes. By adopting a nested model that integrates genetic algorithms and simulated annealing algorithms, it combines the global search breadth of genetic algorithms with the local optimization depth of simulated annealing algorithms. This enables it to quickly converge to the optimal layout scheme in complex constraint spaces, significantly reducing material waste.

[0008] In one optional implementation, the constraint set includes non-overlapping constraints, texture direction constraints, defect avoidance constraints, and process spacing constraints: non-overlapping constraints describe that the boundaries and internal spaces of any two pieces have no intersection; texture direction constraints describe that the placement angle and texture direction angle of the pieces are consistent with a preset reference texture direction, or are within a preset allowable deviation angle range; defect avoidance constraints describe that the outline polygons of the pieces have no intersection with the defective areas on the material surface; and process spacing constraints describe that the process spacing between the outline polygons of adjacent pieces is greater than the minimum process spacing.

[0009] The material cutting method provided in this invention ensures the rationality of the cut piece layout from a spatial dimension through non-overlapping constraints, eliminating cutting errors and material waste caused by overlapping cut pieces; it accurately matches the process requirements of different materials for the texture direction through texture direction constraints, ensuring the consistency of the finished product in appearance and performance, and improving the standardization of the product; it limits the spatial relationship between cut pieces and defect areas through defect avoidance constraints, maximizing the use of the effective area of ​​the material and significantly reducing material loss caused by improper defect avoidance; and it reserves sufficient process space for actual cutting operations through process spacing constraints, avoiding mutual interference between adjacent cut pieces or quality problems such as rough edges during the cutting process, ensuring the stability of the cutting process and the integrity of the finished product.

[0010] In one optional implementation, a set of process parameters for the cut piece is generated through a process knowledge graph, a vector embedding model, and graph neural network reasoning. This includes: based on the process knowledge graph, mapping the entities and relationships corresponding to the cut piece to a continuous vector space through a vector embedding model, so that the head entity, relationship, and tail entity satisfy vector operation constraints; receiving the task feature vector for the cut piece, performing similarity matching in the vector space to determine candidate entities, and performing forward propagation on the candidate entities based on the graph neural network to output a recommended set of process parameters.

[0011] The material cutting method provided in this invention, through the fusion of process knowledge graph and vector space mapping, realizes a structured and computable presentation of entities and relationships related to the cut pieces, thereby improving the retrieval efficiency of process knowledge. Through the collaborative mechanism of vector similarity matching and graph neural network reasoning, it can quickly and accurately locate related process knowledge based on the task characteristics of different cut pieces, significantly improving the accuracy and response speed of process parameter recommendations, reducing excessive reliance on human experience, and strengthening the interpretability of the reasoning process through the forward propagation mechanism of graph neural network, ensuring that the recommendation results not only conform to the process logic but also meet the actual production needs, thus greatly improving the intelligence level and production adaptability of the material cutting system.

[0012] In one optional implementation, the cutting equipment is controlled to perform physical cutting operations according to the layout scheme and process parameter set, and multi-source sensor data streams are collected in real time: the cutting parameters of the cutting equipment are determined according to the process parameter set, including cutting speed, cutting pressure and tool type; the coordinates of the cut pieces in the layout scheme are converted into motion trajectory instructions for the cutting equipment; the cutting equipment is controlled to perform physical cutting operations based on the motion trajectory instructions and cutting parameters, and multi-source sensor data streams are acquired at a preset sampling frequency through force sensors, vibration sensors and acoustic emission sensors deployed in the cutting area.

[0013] The material cutting method provided in this invention ensures a high degree of consistency between the cutting path and the layout plan by accurately mapping the coordinates of the cut pieces and the movement trajectory of the cutting equipment. This effectively reduces material waste and cut piece quality defects caused by trajectory deviations, and improves the conversion efficiency from the layout plan to actual cutting. By dynamically adjusting the cutting speed, pressure, and tool type based on a set of process parameters, adaptive process matching for different materials and cut piece complexities is achieved, avoiding the problem of cutting quality fluctuations under a single process parameter and significantly improving the stability and consistency of the cutting process. By collecting real-time time-series data such as force, vibration, and acoustic emission from multiple source sensors, potential problems such as tool wear and material deformation during the cutting process can be identified in a timely manner, further improving the overall quality and production efficiency of material cutting.

[0014] In one optional implementation, anomaly identification and compensation amount prediction are performed on the multi-source sensor data stream based on a target convolutional neural network model to generate real-time process compensation instructions. This includes: filtering, normalizing, and performing sliding window segmentation on the multi-source sensor data stream to generate a standard input tensor; extracting features and classifying anomalies from the input tensor using the target convolutional neural network model to identify anomaly types such as material defects, tool wear, or mechanical interference in normal and abnormal states; querying predefined compensation rules based on the identified anomaly types, or predicting compensation amounts through neural network regression branches, to generate process compensation instructions that include speed and pressure adjustment amounts.

[0015] The material cutting method provided in this invention effectively filters out interference from environmental noise and data acquisition errors by filtering, normalizing, and segmenting multi-source sensor data streams through a data preprocessing module, ensuring the standardization and consistency of input data. Through feature extraction and anomaly classification capabilities based on a target convolutional neural network model, it achieves accurate identification and rapid differentiation of normal states and various anomaly types, improving the accuracy and response speed of anomaly identification and enabling timely capture of potential problems in the early stages of anomalies. By employing a dual decision-making logic based on predefined compensation rules and neural network regression branches, it ensures the reliability of compensation strategies in common anomaly scenarios while generating optimal compensation amounts for complex and variable anomalies, dynamically adjusting cutting speed and pressure parameters. This effectively mitigates the negative impact of material defects and tool wear on cutting quality, significantly reducing the scrap rate and material waste caused by anomalies, while also reducing unnecessary equipment wear and extending the service life of tools and mechanical components.

[0016] Secondly, the present invention provides a material cutting system, comprising: a layout agent configured to receive geometric data and material property data of the cut pieces, and generate a layout scheme that satisfies non-overlapping constraints, texture direction constraints, and defect avoidance constraints based on a nested layout model; a process knowledge agent configured to construct a process knowledge graph including historical process data, material characteristic parameters, and equipment status parameters, and generate a set of process parameters for the cut pieces through the process knowledge graph, vector embedding model, and graph neural network inference; an execution agent configured to control the cutting equipment to perform physical cutting operations according to the layout scheme and the set of process parameters, and to collect multi-source sensor data streams in real time; a dynamic compensation agent configured to receive multi-source sensor data streams, perform abnormal pattern recognition and compensation amount prediction based on a target convolutional neural network model, and generate real-time process compensation instructions; and a coordination layer configured to establish an information interaction channel between the layout agent, the process knowledge agent, the execution agent, and the dynamic compensation agent.

[0017] The material cutting system provided in this invention features a nested layout model constructed by a layout agent based on multiple constraints such as non-overlap, texture direction adaptation, and defect avoidance, significantly improving material utilization. The process knowledge agent, relying on a process knowledge graph integrating historical data and multi-dimensional parameters, along with an intelligent reasoning mechanism, can quickly output the optimal set of process parameters adapted to specific cut pieces and materials, effectively shortening the process debugging cycle and avoiding parameter deviations caused by experience dependence. The execution agent and dynamic compensation agent collaborate, relying on real-time multi-source sensor data acquisition and the anomaly recognition and compensation prediction capabilities of the target convolutional neural network to correct deviations caused by factors such as temperature drift and tool wear during the cutting process in real time. The information interaction channel constructed by the coordination layer ensures low latency and high synergy in data flow between agents, allowing the system to flexibly respond to changes in production needs for different material types and cut piece specifications, further improving cutting efficiency.

[0018] In one optional implementation, the coordination layer includes: a communication management module for establishing communication channels between agents; a task scheduling module for receiving manufacturing orders and decomposing them into sub-task sequences, dynamically allocating execution priorities based on the current load index of each agent, with the load index provided in real time by the state synchronization module; a conflict resolution module for detecting spatiotemporal conflicts between changes in the sizing scheme and ongoing cutting tasks, pausing or rolling back conflicting tasks through negotiation protocols, and generating updated joint decisions; and a state synchronization module for maintaining a global state vector describing the execution progress of each agent, using a causal consistency protocol to ensure consistent observation order of key state changes by each agent, and predicting the remaining task completion time based on a time series model trained on historical execution data.

[0019] The material cutting system provided in this invention has a communication management module that establishes stable and efficient communication channels between agents, breaking down information barriers and ensuring real-time interaction of key data and instructions such as anomaly detection results and compensation decision commands. This lays the foundation for information exchange for multi-agent collaborative execution of cutting tasks. The task scheduling module dynamically allocates sub-task priorities based on real-time load indices, achieving optimal system resource allocation and effectively avoiding imbalances such as overload or idle local agents. This improves the overall execution efficiency of manufacturing orders and shortens order delivery cycles. The conflict resolution module detects changes in the nesting scheme and the cutting process in real-time. The system addresses spatiotemporal conflicts in tasks by employing negotiation protocols to pause or roll back conflicting tasks, thus avoiding issues such as cutting position deviations, material scrap, and equipment collisions caused by conflicts and ensuring the continuity and stability of the production process. The state synchronization module maintains the consistency of the global state vector based on a causal consistency protocol, ensuring that all agents have a unified understanding of the production state. At the same time, it predicts the remaining completion time of tasks based on a time series model, providing data support for adjusting scheduling strategies in advance. This significantly improves the collaborative decision-making capability and production scheduling flexibility of the material cutting system, providing a core guarantee for the efficient and high-quality operation of the system in complex and ever-changing manufacturing scenarios.

[0020] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the material cutting method described in the first aspect or any corresponding embodiment thereof.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the material cutting method described in the first aspect or any corresponding embodiment thereof.

[0022] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the material cutting method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a material cutting method according to an embodiment of the present invention;

[0025] Figure 2 This is a structural block diagram of a material cutting system according to an embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0027] 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, 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.

[0028] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0029] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0030] According to an embodiment of the present invention, a material cutting method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0031] This embodiment provides a material cutting method applied to flexible materials. Figure 1 This is a flowchart of a material cutting method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0032] Step S101: Receive the geometric data and material property data of the cut pieces, and generate a nested nesting model that satisfies non-overlapping constraints, texture direction constraints, and defect avoidance constraints.

[0033] The geometric data of the cut pieces may include the outline boundary coordinates of the cut pieces, size specifications (such as specific values ​​for length, width, curvature, angle, etc.), topological characteristics (such as whether it is a regular rectangle, circle, or irregular irregular shape), and key information such as the quantity ratio of each cut piece in the same batch, the preset splicing relationship, or symmetry requirements. The material property data may include one or more of the following: physical property data (such as mechanical parameters such as material thickness, density, elastic modulus, elongation at break, etc., and thermal properties such as coefficient of thermal expansion, thermal conductivity, etc.); and surface and texture property data (texture direction, texture density, surface smoothness, and pre-detected surface defect information), processing adaptability property data (such as material cutting resistance coefficient, shear strength, whether there is a tendency to stick to the knife, thermal shrinkage rate), and morphological and batch property data (material width, overall length, winding state, material type, and uniformity index of materials in the same batch).

[0034] In some optional implementations, when receiving the geometric data and material property data of the cut pieces, and generating a nested nesting model that satisfies non-overlapping constraints, texture direction constraints, and defect avoidance constraints, the outline polygons, texture direction vectors, defect area polygons, and minimum process spacing of the cut pieces can be converted into a set of constraints based on the geometric data and material property data of the cut pieces; the optimal nesting scheme is then solved under the set of constraints based on the nested nesting model, which includes a genetic algorithm and / or a simulated annealing algorithm.

[0035] The constraint set includes non-overlapping constraints, texture direction constraints, defect avoidance constraints, and process spacing constraints: Non-overlapping constraints describe that the boundaries and internal spaces of any two pieces have no intersection; Texture direction constraints describe that the placement angle and texture direction angle of the pieces are consistent with the preset reference texture direction, or within the preset allowable deviation angle range; Defect avoidance constraints describe that the outline polygon of the pieces has no intersection with the defect area on the material surface; Process spacing constraints describe that the process spacing between the outline polygons of adjacent pieces is greater than the minimum process spacing.

[0036] Specifically, if a genetic algorithm is used, the placement parameters of the cut pieces (such as coordinate position and placement orientation angle) can be encoded with real numbers to generate an initial population containing multiple candidate layout schemes. Secondly, based on the objective function, material utilization rate is used as the core indicator. At the same time, gradient penalty terms are assigned to schemes that violate non-overlap, texture direction, defect avoidance, and process spacing constraints. For example, a high penalty is added to schemes that seriously violate non-overlap constraints, and a lower penalty is given to schemes that slightly deviate from the texture direction. Then, individuals with high fitness are selected, and the encoded fragments of the individuals are recombined using a uniform crossover operator. Then, a small range of random adjustments are made to some encoded bits through Gaussian mutation to generate a new generation of population. Finally, the best individual in each generation is directly retained to the next generation. The above selection, crossover, and mutation process is repeated until the number of iterations reaches the preset upper limit or the fitness value does not improve significantly for several consecutive generations. The layout scheme with the highest fitness is then output. If simulated annealing is used, a randomly generated initial nesting scheme can be started. Using the reciprocal of material utilization rate combined with constraint violation penalties as the energy function, the position or angle of the cut pieces is randomly adjusted at the initial high temperature to generate a new scheme. If the new scheme has lower energy, it is accepted directly; if the energy is higher, it is accepted with a certain probability according to the Metropolis criterion. As the temperature gradually decreases, the probability of accepting poorer schemes decreases until the temperature drops to a preset threshold, at which point the current optimal scheme is output. In practical applications, the two algorithms can be combined. A genetic algorithm is used for global search to obtain potential optimal solution regions, and then the simulated annealing algorithm is used for local fine-tuning search within these regions, balancing nesting efficiency and scheme quality.

[0037] Furthermore, let the set of cut pieces to be arranged be P = {p i |i=1,...,N}:p i This represents the i-th cut piece, where N is the total number of cut pieces, and each cut piece p... i The set of vertices of its contour polygon, the minimum envelope rectangle, and the texture direction vector τ i Definition. The pattern layout H(S) is determined by the placement position (x) of each cut piece. i ,y i ) and placement direction angle φ i Decide.

[0038] Objective function (maximizing material utilization):

[0039]

[0040] in, Given a fixed width for the material, H(S) represents the layout scheme; S is the required total length (height), which is the variable to be minimized, and is therefore equivalent to min H(S).

[0041] Non-overlapping constraints can be expressed as follows: for any two different cut pieces p i and pj Its outline polygon C i With C j The intersection of the Minkowski set and the origin O must be empty:

[0042] Ci⊕(-C j ) {O}=

[0043] If NFP ij Let the NFP of piece i relative to piece j be:

[0044] (x j -x i ,y j -y i ) Interior (NFP) ij )

[0045] Texture direction constraints can be represented as the cut piece p. i Placement orientation angle φ on the material plane i With respect to the material texture direction θ material The deviation must be within the allowable range:

[0046] |(φ i +arg(τ i ))-θ material |≤Δθ max

[0047] Where, τ i For cut pieces p i Texture direction vector, arg(τ) i ) represents the argument of the texture direction vector, Δθ max The maximum allowable texture orientation deviation.

[0048] The defect avoidance constraint can be expressed as follows: Let the defective regions be a set of polygons, and for any piece of fabric, its outline polygons must not intersect with any defective region.

[0049] C i ∩(∪ k f k )= , i

[0050] Among them, C i f is the i-th cut piece. k Let ∪ be the polygon representing the k-th defect region. k f k The union of all defective regions, It is an empty set. i represents each piece of fabric, indicating that the constraint applies to all pieces of fabric, where i = 1, 2, ..., N, and N is the total number of pieces of fabric.

[0051] Process spacing constraints can be expressed as the minimum cutting spacing d that must be maintained between cut pieces. min This is equivalent to offsetting the outline of the cut piece outwards by d. min Even after / 2, the non-overlapping constraint still needs to be satisfied.

[0052] crossover probability p in genetic algorithm c With the mutation probability p b Dynamically adjusted with the iteration algebra g:

[0053] p c =p c0 -(p c0 -p c1 )×g / G max

[0054] p b =p b0 +(p b1 -p b0 )×g / G max

[0055] Among them, the initial high crossover rate p c0 >Late-stage high crossover rate p c1 High initial mutation rate p b0 <Late-stage high mutation rate p b1 G max The maximum number of iterations, i.e., the initial high crossover rate p. c0 Promotes global exploration, with a high mutation rate in the later stages. b1 Enhance local search.

[0056] Next, simulated annealing perturbation is applied to the individuals generated by the genetic algorithm. A penalty function method can be used to handle complex constraints, transforming the degree of constraint violation into a penalty term added to the objective function. The overall utility function is expressed as:

[0057] E(S) = H(S) + λ × Penalty(S)

[0058] Where H(S) is the nesting scheme, Penalty(S) is the weighted sum of texture deviation penalty, defect overlap penalty and insufficient spacing penalty, and λ is the penalty coefficient.

[0059] For a new solution S' generated by the neighborhood operation, if its objective function value (considering the penalty) E(S')=H(S')+λ×Penalty(S') is worse, then it is accepted with probability:

[0060] P accept=exp(-(E(S')-E(S)) / (k B ×T))

[0061] Temperature T is calculated as T = T0 × α g Cooling (α∈(0,1)), T0 is the initial temperature, k B is the Boltzmann constant.

[0062] The termination condition for the above nested layout model is:

[0063]

[0064] in, The height value is the optimal solution for the g-th generation. This is the preset convergence threshold.

[0065] As shown above, by transforming elements such as cut piece outline, texture direction, defect area, and process spacing into a structured set of constraints, the precise quantification and unified expression of multi-dimensional constraints are achieved. This effectively avoids the problems of incomplete constraint consideration and vague expression when relying on manual experience for pattern making, and provides a solid and rigorous constraint framework for the scientific solution of pattern making schemes. By adopting a nested model that integrates genetic algorithm and simulated annealing algorithm, it combines the global search breadth of genetic algorithm with the local optimization depth of simulated annealing algorithm, and can quickly converge to the optimal pattern making scheme in complex constraint space, greatly reducing material waste. Non-overlapping constraints ensure the rationality of the pattern layout from a spatial perspective, eliminating cutting errors and material waste caused by pattern overlap; texture direction constraints precisely match the process requirements of different materials for texture direction, ensuring the consistency of finished products in appearance and performance, and improving the standardization of products; defect avoidance constraints limit the spatial relationship between pattern pieces and defect areas, maximizing the effective use of material area and significantly reducing material loss caused by improper defect avoidance; process spacing constraints reserve sufficient process space for actual cutting operations, avoiding mutual interference between adjacent pattern pieces or quality problems such as rough edges during the cutting process, ensuring the stability of the cutting process and the integrity of the finished product.

[0066] Step S102: Construct a process knowledge graph that includes historical process data, material property parameters, and equipment status parameters. Generate a set of process parameters for the cut pieces through the process knowledge graph, vector embedding model, and graph neural network reasoning.

[0067] In some optional implementations, when constructing a process knowledge graph that includes historical process data, material property parameters, and equipment status parameters, the historical process data, material property parameters, and equipment status parameters can first undergo structured preprocessing, including cleaning redundant data, correcting outliers, and converting unstructured process experience text (such as operation notes) into structured entity information. Next, based on the process domain ontology model, core entity types (such as "cutting process," "material properties," and "equipment components") and the relationships between entities (such as "material thickness affects cutting pressure" and "equipment tool wear is related to cutting accuracy") are defined. Then, entity extraction tools are used to extract entities and attribute values ​​from the preprocessed data, and semantic relationships between entities are identified through relation extraction algorithms and mapped to the corresponding nodes and edges in the ontology model. Afterward, a process knowledge graph completion method is used to fill in missing entity attributes and potential relationships, for example, completing the recommended cutting parameter range for a new material based on the property parameters of similar materials. Finally, a dynamic update mechanism for the process knowledge graph is established, regularly importing new process experimental data, material batch test results, and equipment maintenance records to ensure that the process knowledge graph can reflect the latest process knowledge in real time.

[0068] In some optional implementations, when generating a set of process parameters for the cut piece using process knowledge graphs, vector embedding models, and graph neural network inference, the entities and relationships corresponding to the cut piece can be mapped to a continuous vector space using vector embedding models based on the process knowledge graph, so that the head entity, relationship, and tail entity satisfy vector operation constraints; the task feature vector for the cut piece is received, similarity matching is performed in the vector space to determine candidate entities, and forward propagation is performed on the candidate entities based on the graph neural network to output the recommended set of process parameters.

[0069] Specifically, entities and relationships can be mapped to a low-dimensional continuous vector space using TransE or RotatE models. The task feature vector v for the cutout is... new Encode the query vector q, and find the closest material entity in the vector space using cosine similarity:

[0070]

[0071] in, E is the material entity that best matches the task feature vector q. m Let be a set of material entities, q be a task feature vector, e be an entity vector, and ||·|| denote the magnitude of the vector. This will include... The subgraph of the entity and its neighboring entities is input into the GraphSAGE or GAT model, and neighbor information is aggregated through a message passing mechanism to output a recommended process parameter vector v. p .

[0072] The forward propagation formula for a graph neural network is:

[0073]

[0074] in, H is the normalized adjacency matrix. (l) Let W be the feature matrix of the nodes in the l-th layer. (l) Let be the learnable parameter matrix, and σ be the nonlinear activation function.

[0075] As an example, suppose the cut piece is a polyester-cotton blend fabric with a thickness of 1.8mm, and its task feature vector v new This includes information such as material type, thickness, and cutting accuracy requirements (±0.1mm). (The last part, "v," appears to be an unrelated fragment and is left untranslated.) new After being encoded into a query vector q, material entities are matched using cosine similarity. The specified entity is "polyester-cotton blended fabric (thickness 1.5-2.0mm)," and its adjacent entities include the cutting equipment entity "CNC cutting machine M-200" and the basic process parameter entity "p." m0 =0.6MPa, pm1=2.2MPa, cutting speed entity “v0=30mm / s”, etc. The subgraph containing these entities is input into the GraphSAGE model, with the initial feature matrix H. (0) It covers the attribute characteristics of each entity (such as the thickness of a material entity, the rated power of a device entity, and the numerical range of a parameter entity). In the normalized adjacency matrix A, The connection weights with each adjacent entity are set to 0.8 (direct association) and 0.5 (indirect association). The graph neural network is configured with two layers, and the first layer can learn the parameter matrix W. (0) The dimension is (128, 64), and the activation function σ is ReLU; the second layer learnable parameter matrix W (1) The dimension is (64, 32), and the activation function is Sigmoid. After forward propagation, H is calculated. (1) =ReLU(A×H (0) ×W (0) ), H (2) =Sigmoid(A×H (1) ×W (1) The final output recommended process parameter vector v includes the cutting pressure p. m Combining formula p m =p m0 +(p m1 -p m0 )×g x / G maxx The current task's cutting requirement coefficient g x Based on the thickness of the polyester-cotton blended fabric (1.8mm) and the required precision, the result is calculated to be 0.7 (G). maxxTaking the maximum demand factor of the equipment as 1.0), then p m =0.6+(2.2-0.6)×0.7=0.6+1.12=1.72MPa, which is the cutting pressure parameter recommended by the graph neural network. Meanwhile, v p It also outputs recommended cutting speeds of 32mm / s, die angles of 15°, and other process parameters, forming a complete set of process parameters. This example demonstrates that the reasoning method based on process knowledge graphs and graph neural networks can effectively combine material properties, equipment characteristics, and historical process experience to output process parameters that are accurately adapted to the current cutting task, improving cutting efficiency and quality.

[0076] As shown above, by integrating the process knowledge graph and vector space mapping, a structured and computable representation of entities and relationships related to the cut pieces is achieved, improving the retrieval efficiency of process knowledge. Through the collaborative mechanism of vector similarity matching and graph neural network reasoning, the associated process knowledge can be quickly and accurately located based on the task characteristics of different cut pieces, significantly improving the accuracy and response speed of process parameter recommendations and reducing excessive reliance on human experience. At the same time, the forward propagation mechanism of the graph neural network enhances the interpretability of the reasoning process, ensuring that the recommendation results not only conform to the process logic but also meet the actual production needs, greatly improving the intelligence level and production adaptability of the material cutting system.

[0077] Step S103: Control the cutting equipment to perform physical cutting operations according to the layout plan and process parameter set, and collect multi-source sensor data streams in real time.

[0078] In some optional implementations, when controlling the cutting equipment to perform physical cutting operations according to the layout scheme and process parameter set, and acquiring multi-source sensor data streams in real time, the cutting parameters of the cutting equipment can be determined according to the process parameter set, including cutting speed, cutting pressure, and tool type; the coordinates of the cut pieces in the layout scheme can be converted into motion trajectory instructions for the cutting equipment; the cutting equipment can be controlled to perform physical cutting operations based on the motion trajectory instructions and cutting parameters, and multi-source sensor data streams can be acquired at a preset sampling frequency through force sensors, vibration sensors, and acoustic emission sensors deployed in the cutting area.

[0079] Specifically, the cutting pressure p m The calculation can be performed using formula p. m =p m0+ (p m1 -p m0 )×g m / G maxm , where p m0 p is the basic pressure threshold of the cutting equipment. m1 g is the maximum allowable pressure value of the equipment. m G represents the actual thickness of the current cut material.maxm This setting defines the maximum material thickness supported by the cutting equipment. The cutting speed can be adapted to the hardness coefficient of the cut material and the selected blade type; for example, a high-speed mode (e.g., 300-500 mm / s) can be used when cutting flexible textile materials. The blade type selection must match the complexity of the cut material's shape and its hardness. For example, circular rotary blades are preferred for cuts with dense curves to ensure smooth curve cutting, while diamond-coated blades are preferred for thick, hard materials to improve cutting accuracy. In the motion trajectory command conversion stage, the Cartesian coordinate system in the nesting scheme needs to be converted to the pulse equivalent coordinate system of the cutting equipment, and circular and linear interpolation algorithms should be incorporated to ensure the continuity and accuracy of the cut material edge cutting. During multi-source sensor data acquisition, the preset sampling frequency can be set to 1 kHz. Force sensors monitor the deviation between the cutting pressure and the calculated value in real time, vibration sensors capture abnormal vibration amplitudes of the equipment spindle (e.g., triggering an alarm when exceeding 5 μm), and acoustic emission sensors detect acoustic emission signal characteristics (e.g., energy value, count rate) during material fracture.

[0080] As mentioned above, by accurately mapping the coordinates of the cut pieces to the movement trajectory of the cutting equipment, the cutting path is ensured to be highly consistent with the layout plan, effectively reducing material waste and cut piece quality defects caused by trajectory deviations, and improving the conversion efficiency from the layout plan to actual cutting. By dynamically adjusting the cutting speed, pressure, and tool type based on the process parameter set, adaptive process matching for different materials and cut piece complexities is achieved, avoiding the problem of cutting quality fluctuations under a single process parameter, and significantly improving the stability and consistency of the cutting process. By collecting time-series data such as force, vibration, and acoustic emission in real time through multi-source sensors, potential problems such as tool wear and material deformation during the cutting process can be identified in a timely manner, further improving the overall quality and production efficiency of material cutting.

[0081] Step S104: Based on the target convolutional neural network model, perform anomaly identification and compensation amount prediction on the multi-source sensor data stream, and generate real-time process compensation instructions.

[0082] The target convolutional neural network model includes: a Conv1D layer with a kernel size of 5, 32 filters, and ReLU activation function; a MaxPooling1D layer with a pooling size of 2; a Conv1D layer with a kernel size of 3, 64 filters, and ReLU activation function; a GlobalAveragePooling1D layer; a Dense layer with 128 units and ReLU activation function; and an output layer that uses Softmax activation for classification tasks and linear activation for regression tasks.

[0083] In some optional implementations, when receiving multi-source sensor data streams, performing anomaly pattern recognition and compensation amount prediction based on a target convolutional neural network model, and generating real-time process compensation instructions, the multi-source sensor data streams can be filtered, normalized, and segmented using a sliding window to generate standard input tensors. The input tensors are then used for feature extraction and anomaly classification via the target convolutional neural network model to identify anomaly types such as material defects, tool wear, or mechanical interference in normal and abnormal states. Based on the identified anomaly types, predefined compensation rules are queried, or compensation amounts are predicted through neural network regression branches to generate process compensation instructions that include speed and pressure adjustments. Material defects can include localized hardening, cracks, and bubbles; tool wear can include varying degrees of wear such as edge dulling, chipping, and coating peeling; and mechanical interference includes transmission component jamming, guide rail misalignment, and fixture loosening. For different anomaly types, predefined compensation rules will match corresponding adjustment strategies: for example, when local hardening of the material is detected, the cutting pressure is appropriately increased and the cutting speed is reduced to ensure that the cutting edge is neat; if the tool edge dulling is detected, the pressure compensation amount is increased or a tool replacement prompt is triggered; for guide rail offset in mechanical interference, the coordinate parameters of the cutting path are finely adjusted in real time.

[0084] Specifically, the filtering process can employ wavelet threshold filtering to remove high-frequency interference noise from the multi-source sensor data stream while retaining the effective components of force, vibration, and acoustic emission signals. Normalization maps the data from each sensor channel to the [0,1] interval, eliminating the characteristic scale bias caused by differences in the measurement ranges of different sensors. Sliding window segmentation divides the preprocessed data with a window duration of 100ms and a step size of 50ms, forming an input tensor with a dimension of [window length × number of sensor channels]. The first Conv1D layer of the target convolutional neural network model extracts local temporal features of the data stream using a 5×32 convolutional kernel. After the feature dimension is compressed by the MaxPooling1D layer, the second Conv1D layer further mines deep correlation features using a 3×64 convolutional kernel. The GlobalAveragePooling1D layer performs global average pooling on the deep features to obtain a global feature vector of fixed dimensions. This vector is input to a 128-unit Dense layer for nonlinear transformation and then fed into the classification and regression branches respectively. The classification branch uses Softmax activation to output the probability distribution of four categories: normal state, material defect, tool wear, and mechanical interference, and selects the category with the highest probability as the anomaly type. The regression branch uses linear activation to output the predicted values ​​of the speed adjustment Δv and the pressure adjustment Δpm, where the pressure adjustment can be combined with the formula p. m =p m0 +(p m1 -p m0 )×g o / G maxo Calculate go G is the quantized value of the current anomaly level (obtained by mapping the global feature vector through a fully connected layer). maxo p is the preset maximum anomaly threshold. m0 As the reference cutting pressure, p m1 To maximize the compensation pressure, a process compensation instruction is generated based on the identified anomaly type and the adjustment amount predicted by regression. This instruction specifies the direction and magnitude of the speed adjustment and the target value of the pressure adjustment, thereby achieving real-time dynamic compensation of the cutting process parameters based on the direction and magnitude of the speed adjustment and the target value of the pressure adjustment.

[0085] Furthermore, the target compensation amount can be determined based on the rule-based compensation amount and the predicted compensation amount from the neural network regression branch. The weights of the rule-based compensation amount and the predicted compensation amount can be dynamically adjusted based on the probability entropy output by the neural network regression branch.

[0086] Furthermore, sensor anomalies can be detected based on dynamically updated statistical confidence intervals, and a safety threshold [L] can be established. low ,L high ].

[0087] L low =μ-k×σ

[0088] L high =μ+k×σ

[0089] Where μ is the mean, σ is the standard deviation, which can be estimated online based on historical normal data; and k is the confidence level coefficient.

[0090] As described above, the data preprocessing module effectively filters out interference from environmental noise and data acquisition errors by filtering, normalizing, and segmenting multi-source sensor data streams through filtering, normalization, and sliding window processing, ensuring the standardization and consistency of input data. Through feature extraction and anomaly classification capabilities based on a target convolutional neural network model, accurate identification and rapid differentiation of normal states and various anomaly types are achieved, improving the accuracy and response speed of anomaly identification and enabling timely detection of potential problems in the early stages of anomalies. By employing a dual decision-making logic based on predefined compensation rules and neural network regression branches, the reliability of compensation strategies in common anomaly scenarios is ensured, while optimal compensation amounts are generated for complex and variable anomaly situations. Dynamic adjustments to cutting speed and pressure parameters effectively mitigate the negative impact of material defects and tool wear on cutting quality, significantly reducing the scrap rate and material waste caused by anomalies, while also minimizing unnecessary equipment wear and extending the service life of tools and mechanical components.

[0091] Step S105: Adjust the cutting parameters in response to the process compensation command, and feed the execution data back to the process knowledge graph.

[0092] In some optional implementations, when adjusting cutting parameters in response to process compensation commands and feeding the execution data back to the process knowledge graph, cutting quality indicators (such as edge smoothness of cut pieces and dimensional deviation rate), equipment operating status data (such as motor load and tool temperature), and anomaly mitigation effect data (such as changes in scrap rate and material utilization improvement) before and after parameter adjustment can be recorded simultaneously. These multi-dimensional data are then mapped to corresponding anomaly scenarios and compensation strategies, supplementing or correcting the triplet relationship of "anomaly type - compensation parameter - effect evaluation" in the process knowledge graph. Simultaneously, an effectiveness score for the current compensation strategy is calculated based on the feedback data. If the score is lower than a preset threshold, the self-learning module of the process knowledge graph is triggered to re-optimize the parameter weights in the compensation rules based on similar historical cases, further improving the accuracy and adaptability of subsequent process compensation. Furthermore, key features in the feedback data (such as the optimal cutting pressure p under specific anomalies) can also be used to... m The range of values ​​is extracted into new rule entries in the process knowledge graph, providing a more direct decision-making reference for similar cutting scenarios in the future.

[0093] The material cutting method provided in this embodiment generates a nested nesting model that satisfies multiple constraints such as non-overlap, texture direction, and defect avoidance, effectively improving material utilization while ensuring that the cutting of pieces meets design specifications and quality requirements, reducing material scrap and rework costs caused by constraint violations. By integrating historical process data, material properties, and equipment status through a process knowledge graph, and combining vector embedding and graph neural network inference to generate a set of process parameters, it achieves precise and intelligent decision-making of process parameters, avoiding the subjective bias and insufficient adaptability of empirical parameter settings. By acquiring multi-source sensor data streams in real time and using a target convolutional neural network model for anomaly identification and compensation prediction, it can quickly capture abnormal states in the cutting process and generate targeted compensation instructions, significantly reducing cutting errors and improving cutting accuracy and process stability. Through a closed-loop mechanism that feeds execution data back to the process knowledge graph, the process knowledge base is continuously updated iteratively, enhancing the adaptability and robustness of the method under different material types, equipment conditions, and cutting piece requirements, providing comprehensive assurance for the efficient and high-quality operation of the material cutting process.

[0094] As a specific application example of the present invention (taking the cutting of a flawed natural leather shoe upper component as an example):

[0095] If using imported Italian full-grain cowhide leather, with dimensions of 1800mm × 1200mm and a thickness controlled within the range of 2.1 ± 0.3mm, and three natural scars or blemishes on the surface. Cut 12 pieces of upper components for a certain brand of athletic shoes, including parts with complex curved contours such as the toe, sides, and tongue. The cut roughness must meet the requirement of Ra < 50μm, with no scorched edges or burrs; the grain direction deviation should be less than 10 degrees. Input feature = [Material Material] Emb Thickness=2.1, Elasticity=85, Roughness=0.6, PartArea=1500]^T. GNN Inference: Select a subgraph containing historical order #2023-0892 (similar thickness 1.9mm, elasticity 82) and related parameters for analysis. Recommended Output: Use vibrating knife type A, corresponding parameter set P001 (cutting speed v1=350mm / s, pressure p1=45N, frequency... f 1=180Hz); Fine-tuning: Due to a thickness deviation Δh=0.1mm, adjust the pressure. for p1′=47N. The vector embedding model, based on an initial population size of 100 individuals, converges to the optimal solution after 500 generations of iterations. Key metrics include material utilization: η=79.2%; defect avoidance effect: complete avoidance of all three scar areas, with an avoidance cost area accounting for 8.3%; and texture consistency: texture deviation of all cut pieces is less than 8 degrees. During the cutting process, the standard deviation is calculated within a window T based on the force sensor reading sequence. A threshold alarm is triggered when the absolute value of the standard deviation exceeds a threshold of 3, enabling preliminary identification of abnormal data. The standardized window data is input into a 1D-CNN model, outputting a probability distribution. (e.g., [0.02, 0.85, 0.10, 0.03]), corresponding to states such as "normal," "local hardening," "tool wear," and "mechanical interference," is diagnosed as "local hardening" based on the highest probability (0.85). The suggested compensation amount is directly output through the model regression branch, for example, Δv = -35 mm / s, Δp = +2.3 N. This is combined with rule suggestions (e.g., Δv...). rule =-28 mm / s) and model confidence (con) fnn =0.85), using the weighted formula (Δv) final = (1-conf nn )×Δv rule +conf nn The core indicator Δv is used to calculate the final compensation amount. An example result is Δv. final =-33.05 mm / s, Δp final≈+1.96 N. After compensation, new data pairs are collected and added to the online learning buffer for subsequent model fine-tuning. Local hardening is a specific manifestation of material defects (e.g., abnormal local hardness in the material).

[0096] This embodiment also provides a material cutting system for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, hardware implementations, or combinations of software and hardware, are also possible and contemplated.

[0097] This embodiment provides a material cutting system, such as Figure 2 As shown, it includes:

[0098] The layout agent is configured to receive geometric data and material property data of the cut pieces, and generate a layout scheme that satisfies non-overlapping constraints, texture direction constraints and defect avoidance constraints based on the nested layout model.

[0099] The process knowledge intelligent agent is configured to construct a process knowledge graph that includes historical process data, material property parameters, and equipment status parameters. It generates a set of process parameters for the cut pieces through the process knowledge graph, vector embedding model, and graph neural network reasoning.

[0100] The execution agent is configured to control the cutting equipment to perform physical cutting operations according to the nesting scheme and process parameter set, and to collect multi-source sensor data streams in real time;

[0101] The dynamic compensation agent is configured to perform anomaly identification and compensation amount prediction on multi-source sensor data streams based on a target convolutional neural network model, and generate real-time process compensation instructions.

[0102] The coordination layer is configured to establish information exchange channels between the sorting agent, the process knowledge agent, the execution agent, and the dynamic compensation agent.

[0103] In some alternative implementations, the sorting agent includes:

[0104] The constraint modeling module is used to transform the outline polygons, texture direction vectors, defect area polygons, and minimum process spacing of the cut pieces into a set of constraint conditions based on the geometric data and material property data of the cut pieces.

[0105] The nesting solution module is used to solve for the optimal nesting scheme under a set of constraints based on a nested nesting model, which includes a genetic algorithm and / or a simulated annealing algorithm.

[0106] In some alternative implementations, the constraint set includes non-overlapping constraints, texture orientation constraints, defect avoidance constraints, and process spacing constraints:

[0107] Non-overlapping constraints describe that the boundaries and internal spaces of any two pieces do not intersect; texture direction constraints describe that the placement angle and texture direction angle of the pieces are consistent with the preset reference texture direction, or within the preset allowable deviation angle range; defect avoidance constraints describe that the outline polygon of the pieces does not intersect with the defect area on the material surface; process spacing constraints describe that the process spacing between the outline polygons of adjacent pieces is greater than the minimum process spacing.

[0108] In some alternative implementations, the process knowledge agent includes:

[0109] The knowledge representation module is used to map the entities and relationships corresponding to the cut pieces to a continuous vector space through a vector embedding model based on the process knowledge graph, so that the head entity, relationship, and tail entity satisfy the vector operation constraints.

[0110] The reasoning query module receives the task feature vector for the cut piece, performs similarity matching in the vector space to determine candidate entities, and performs forward propagation on the candidate entities based on a graph neural network to output a recommended set of process parameters.

[0111] In some alternative implementations, the executing agent includes:

[0112] The process configuration module is used to determine the cutting parameters of the cutting equipment based on the process parameter set. The cutting parameters include cutting speed, cutting pressure and tool type.

[0113] The motion control module is used to convert the coordinates of the cut pieces in the layout plan into motion trajectory instructions for the cutting equipment.

[0114] The data acquisition module is used to control the cutting equipment to perform physical cutting operations based on motion trajectory commands and cutting parameters, and to acquire multi-source sensor data streams at a preset sampling frequency through force sensors, vibration sensors and acoustic emission sensors deployed in the cutting area.

[0115] In some alternative implementations, the dynamically compensated agent includes:

[0116] The data preprocessing module is used to filter, normalize, and perform sliding window segmentation on multi-source sensor data streams to generate standard input tensors.

[0117] The anomaly detection module is used to extract features and classify anomalies from the input tensor through the target convolutional neural network model, and to identify the anomaly types such as material defects, tool wear or mechanical interference in normal and abnormal states.

[0118] The compensation decision module is used to query predefined compensation rules based on the identified anomaly type, or predict the compensation amount through a neural network regression branch, and generate process compensation instructions that include speed adjustment and pressure adjustment.

[0119] In some alternative implementations, the coordination layer includes:

[0120] The communication management module is used to establish communication channels between various intelligent agents;

[0121] The task scheduling module is used to receive manufacturing orders and decompose them into a sequence of sub-tasks. It dynamically allocates execution priorities based on the current load index of each agent. The load index is provided in real time by the state synchronization module.

[0122] The conflict resolution module is used to detect spatiotemporal conflicts between changes to the nesting scheme and ongoing cutting tasks, suspend or roll back conflicting tasks through negotiation protocols, and generate updated joint decisions.

[0123] The state synchronization module is used to maintain a global state vector describing the execution progress of each agent. It adopts a causal consistency protocol to ensure that the observation order of key state changes by each agent is consistent, and predicts the remaining completion time of the task based on a time series model trained on historical execution data.

[0124] Furthermore, the information flow in multi-agent collaboration can be formalized as an extension of the partially observable Markov decision process (POMDP) ​​in a multi-agent environment. At time t, each agent selects an action (adjusting parameters, sending messages) based on observed local information, and obtains global rewards from the environment (such as overall utilization, quality score, and time penalty), with the overall objective being to maximize the expected cumulative reward.

[0125] The key transformation of the data stream is that the raw sensor data is converted into observations through a feature extraction function. The coordination layer decomposes the data through a joint action value function to guide the decision-making of each agent. The physical state transitions of each agent can be described by a dynamic model.

[0126] The node positions in the process knowledge graph are calculated iteratively by minimizing the system energy formula, including the repulsive forces and spring forces between nodes. The tool wear health index model is as follows:

[0127]

[0128] Where β is the natural degradation rate; p is the historical characteristic lag order = 10; α j The regression coefficients are obtained from historical data; x t j Historical characteristic values ​​(such as tool wear-related characteristics: cutting force increase trend, vibration spectrum changes, etc.); εt To observe noise.

[0129] As mentioned above, the communication management module breaks down information barriers between agents by establishing stable and efficient communication channels, ensuring real-time interaction of key data and instructions such as anomaly detection results and compensation decision commands. This lays the foundation for information exchange for multi-agent collaborative execution of cutting tasks. The task scheduling module dynamically allocates sub-task priorities based on real-time load indices, achieving optimal allocation of system resources. This effectively avoids imbalances caused by overload or idle local agents, improving the overall execution efficiency of manufacturing orders and shortening order delivery cycles. The conflict resolution module detects changes in the nesting scheme and spatiotemporal conflicts during the execution of cutting tasks in real time. The system employs a negotiation protocol to pause or roll back conflicting tasks, avoiding issues such as cutting position deviations, material scrap, and equipment collisions caused by conflicts, thus ensuring the continuity and stability of the production process. The state synchronization module maintains the consistency of the global state vector based on a causal consistency protocol, ensuring that all agents have a unified understanding of the production state. At the same time, it predicts the remaining completion time of tasks based on a time series model, providing data support for adjusting scheduling strategies in advance. This significantly improves the collaborative decision-making capability and production scheduling flexibility of the material cutting system, providing a core guarantee for the efficient and high-quality operation of the system in complex and ever-changing manufacturing scenarios.

[0130] The material cutting system provided in this invention features a nested layout model constructed by a layout agent based on multiple constraints such as non-overlap, texture direction adaptation, and defect avoidance, significantly improving material utilization. The process knowledge agent, relying on a process knowledge graph integrating historical data and multi-dimensional parameters, along with an intelligent reasoning mechanism, can quickly output the optimal set of process parameters adapted to specific cut pieces and materials, effectively shortening the process debugging cycle and avoiding parameter deviations caused by experience dependence. The execution agent and dynamic compensation agent collaborate, relying on real-time multi-source sensor data acquisition and the anomaly recognition and compensation prediction capabilities of the target convolutional neural network to correct deviations caused by factors such as temperature drift and tool wear during the cutting process in real time. The information interaction channel constructed by the coordination layer ensures low latency and high synergy in data flow between agents, allowing the system to flexibly respond to changes in production needs for different material types and cut piece specifications, further improving cutting efficiency.

[0131] The material cutting system provided in this embodiment of the invention can execute the material cutting method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0132] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0133] The following is a detailed reference. Figure 3 The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0134] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0135] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a memory 308, or installed from a ROM 302. When the computer program is executed by the processor 301, it performs the functions defined in the material cutting method of the embodiments of the present invention.

[0136] Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0137] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the material cutting method shown in the above embodiments is implemented.

[0138] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0139] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for cutting materials, characterized in that, The method includes: Receive the geometric data and material property data of the cut pieces, and generate a nested nesting model to meet the non-overlapping constraints, texture direction constraints and defect avoidance constraints. A process knowledge graph is constructed, which includes historical process data, material property parameters, and equipment status parameters. A set of process parameters for the cut piece is generated through the process knowledge graph, vector embedding model, and graph neural network inference. The cutting equipment is controlled to perform physical cutting operations according to the layout scheme and the set of process parameters, and multi-source sensor data streams are collected in real time. Based on the target convolutional neural network model, anomaly identification and compensation amount prediction are performed on the multi-source sensor data stream, and real-time process compensation instructions are generated. The cutting parameters are adjusted in response to the process compensation command, and the execution data is fed back to the process knowledge graph; the geometric data and material property data of the received cut pieces are used to generate a nested nesting model that satisfies non-overlapping constraints, texture direction constraints, and defect avoidance constraints, including: Based on the geometric data and material property data of the cut piece, the outline polygon, texture direction vector, defect area polygon and minimum process spacing of the cut piece are transformed into a set of constraint conditions. The nested nesting model is used to solve for the optimal nesting scheme under the constraint set. The nested nesting model includes a genetic algorithm and / or a simulated annealing algorithm. The step of controlling the cutting equipment to perform physical cutting operations according to the layout scheme and the set of process parameters, and collecting multi-source sensor data streams in real time, includes: The cutting parameters of the cutting equipment are determined according to the set of process parameters, and the cutting parameters include cutting speed, cutting pressure and tool type. Convert the coordinates of the cut pieces in the layout scheme into motion trajectory instructions for the cutting equipment; Based on the motion trajectory command and the cutting parameters, the cutting device is controlled to perform physical cutting operations, and the multi-source sensor data stream is acquired at a preset sampling frequency through force sensors, vibration sensors and acoustic emission sensors deployed in the cutting area. The step of performing anomaly identification and compensation amount prediction on the multi-source sensor data stream based on the target convolutional neural network model to generate real-time process compensation instructions includes: The multi-source sensor data stream is filtered, normalized, and subjected to sliding window segmentation to generate a standard input tensor. The target convolutional neural network model is used to extract features and classify anomalies in the input tensor, identifying the anomaly types such as material defects, tool wear, or mechanical interference in normal and abnormal states. Based on the identified anomaly type, a predefined compensation rule is queried, or the compensation amount is predicted through a neural network regression branch, to generate the process compensation instruction containing speed adjustment and pressure adjustment amounts.

2. The method according to claim 1, characterized in that, The constraint set includes non-overlap constraints, texture direction constraints, defect avoidance constraints, and process spacing constraints: The non-overlapping constraint describes that the boundaries and internal spaces of any two pieces have no intersection; the texture direction constraint describes that the placement angle and texture direction angle of the pieces are consistent with the preset reference texture direction, or are within the preset allowable deviation angle range; the defect avoidance constraint describes that the outline polygon of the pieces has no intersection with the defect area on the material surface. The process spacing constraint is used to describe the distance between the process spacing of the outline polygons of adjacent cut pieces that is greater than the minimum process spacing.

3. The method according to claim 1, characterized in that, The process parameter set generated for the cut piece through the process knowledge graph, vector embedding model, and graph neural network inference includes: Based on the process knowledge graph, the entities and relationships corresponding to the cut pieces are mapped to a continuous vector space through the vector embedding model, so that the head entity, relationship, and tail entity satisfy the vector operation constraints. The system receives the task feature vector for the cut piece, performs similarity matching in the vector space to determine candidate entities, and performs forward propagation on the candidate entities based on a graph neural network to output the recommended set of process parameters.

4. A material cutting system, applied to the material cutting method of claim 1, characterized in that, The system includes: The layout agent is configured to receive geometric data and material property data of the cut pieces, and generate a layout scheme that satisfies non-overlapping constraints, texture direction constraints and defect avoidance constraints based on the nested layout model. The process knowledge intelligent agent is configured to construct a process knowledge graph that includes historical process data, material property parameters, and equipment status parameters, and to generate a set of process parameters for the cut piece through the process knowledge graph, vector embedding model, and graph neural network inference. The execution agent is configured to control the cutting equipment to perform physical cutting operations according to the layout scheme and the set of process parameters, and to collect multi-source sensor data streams in real time; The dynamic compensation agent is configured to receive the multi-source sensor data stream, perform abnormal pattern recognition and compensation amount prediction based on the target convolutional neural network model, and generate real-time process compensation instructions. The coordination layer is configured to establish information exchange channels between the sorting agent, the process knowledge agent, the execution agent, and the dynamic compensation agent.

5. The system according to claim 4, characterized in that, The coordination layer includes: The communication management module is used to establish communication channels between various intelligent agents; The task scheduling module is used to receive manufacturing orders and decompose them into a sequence of sub-tasks, and dynamically allocate execution priorities according to the current load index of each intelligent agent. The load index is provided in real time by the state synchronization module. The conflict resolution module is used to detect spatiotemporal conflicts between the change in the layout scheme and the cutting task being executed, suspend or roll back the conflicting task through negotiation protocol, and generate an updated joint decision. The state synchronization module is used to maintain a global state vector describing the execution progress of each agent. It adopts a causal consistency protocol to ensure that the observation order of key state changes by each agent is consistent, and predicts the remaining completion time of the task based on a time series model trained on historical execution data.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method described in any one of claims 1 to 3.

7. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the method according to any one of claims 1 to 3.

Images