Numerical control machining simulation system based on artificial intelligence

Abstract

The invention discloses a numerical control machining simulation system based on artificial intelligence, and relates to the field of numerical control machining control. Comprising a demand analysis module, a processing modeling module, a data perception module, a knowledge modeling module, a parameter recommendation module, a path generation module, an error prediction module, a digital simulation module, a dynamic correction module, a risk detection module and an auxiliary decision module. Structured integration and multiplexing of historical experience and multi-source data are achieved, efficient knowledge migration can be carried out among different machining tasks, the cold start capacity and parameter optimization efficiency under the new working condition are remarkably improved, the machining performance can be continuously improved according to real-time feedback, the manual trial and error cost is greatly reduced, and the machining efficiency is improved. High-precision modeling of error distribution is achieved, and the consistency and usability between simulation and actual machining are improved.

Description

Technical Field

[0001] The present invention relates to the field of numerical control machining control, and in particular to a numerical control machining simulation system based on artificial intelligence. Background Art

[0002] As the global manufacturing industry accelerates its transformation toward digitalization, intelligence, and greening, the importance of CNC machining technology in advanced manufacturing is becoming increasingly prominent, especially in industries such as aerospace, medical devices, precision molds, and rail transportation, where it undertakes the task of high-precision, high-reliability manufacturing of key parts. However, traditional CNC machining simulation systems mainly rely on static geometric modeling and empirical formula derivation, and are insufficiently responsive to the numerous dynamic and uncertain factors in the actual machining process (such as material anisotropy, tool wear, equipment drift, and ambient temperature fluctuations). This leads to deviations between simulation results and actual effects, making it difficult to meet the needs of the new generation of intelligent manufacturing systems for high-precision prediction and high-reliability control.

[0003] After searching, Chinese patent number CN101739884A discloses a CNC machining simulation system. Although this invention adopts a simulation mode that combines virtual CNC equipment with a hard panel simulator, it has the characteristics of reasonable design, strong simulation realism, wide applicability and easy expansion. It is very suitable for large-scale operation training of CNC systems and effectively improves the training effect. However, it cannot perform structured integration and reuse of historical experience and multi-source data, which reduces the cold start capability and parameter optimization efficiency under new working conditions. It cannot continuously improve machining performance based on real-time feedback, and increases the cost of manual trial and error. Therefore, we propose a CNC machining simulation system based on artificial intelligence. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects in the prior art and provide a numerical control machining simulation system based on artificial intelligence.

[0005] The present invention proposes an artificial intelligence-based numerical control machining simulation system, which includes: a demand analysis module, a machining modeling module, a data perception module, a knowledge modeling module, a parameter recommendation module, a path generation module, an error prediction module, a digital simulation module, a dynamic correction module, a risk detection module, and a decision support module;

[0006] The demand analysis module is used to receive the geometric design and quality requirements of the product and analyze the part processing technology requirements;

[0007] The processing modeling module constructs a processing model of the part based on the geometric design and process requirements;

[0008] The data sensing module is used to collect sensor data including tool vibration, spindle temperature, stress and strain, and cutting force;

[0009] The knowledge modeling module integrates historical processing data, equipment performance, material mechanical properties and tool geometry information to build a cross-material-machine-tool map;

[0010] The parameter recommendation module matches processing equipment, materials and tools according to process requirements, processing models and material-machine-tool-tool maps, and recommends initial processing parameters;

[0011] The path generation module generates an initial processing path by combining the processing model and the initial processing parameters, and plans the path in layers;

[0012] The error prediction module predicts the errors caused by various factors based on various sensor data and processing paths, and generates a high-precision error distribution map;

[0013] The digital simulation module simulates the machining process in real time based on the sensor data, machining path, error information and machining parameters, verifies the rationality of the path, the validity of the parameters and the conformity of the machining results, and outputs a simulation report;

[0014] The dynamic correction module fine-tunes and replans the machining path based on real-time simulation data and error prediction results;

[0015] The risk detection module analyzes the real-time processing path, generates a collision heat map, and performs visual monitoring and retrospective analysis of potential collision risks;

[0016] The auxiliary decision module assists in optimizing the current processing parameters based on the simulation data.

[0017] As a further solution of the present invention, the specific steps of the requirement analysis module for analyzing the part processing technology requirements are as follows:

[0018] Q1.1: Use a feature extraction method based on boundary representation and topology rules to decompose the CAD model into a set of topological surfaces, extract the connection relationships between the surfaces, and identify basic geometric units on the part surface from the decomposed topological surfaces, including but not limited to holes, slots, steps, and contours. Classify each identified basic geometric unit as a machining feature.

[0019] Q1.2: Analyze the impact of each processing feature on product performance. If the impact exceeds a preset threshold, the processing feature is marked as "critical" and will be used as a key point to control processing accuracy in subsequent processing steps, and corresponding controls will be strengthened. The specific calculation formula for the impact is as follows:

[0020]

[0021] Where, Representative Dimensional sensitivity of each machined feature; represents the performance indicator function; Representative Dimensional parameters of each machining feature; Representative The weight of each processing feature in the structure;

[0022] Q1.3: Based on the roughness parameters specified in the product drawing or quality specification, deduce the required machining level. Select the appropriate tool, cutting parameters, and machining method based on the obtained machining level. Convert the tolerance range to the machining error limit for the corresponding product and establish an initial relationship with the machining parameters. Finally, based on the product machining characteristics and clamping conditions, infer a reasonable machining sequence. The specific calculation formula for the machining level is as follows:

[0023]

[0024] Where, Representative The root mean square value of the roughness of the machined surface; represents the total number of sampling points; Representative Surface height at each sampling point; Represents the average value of the height of all sampling points;

[0025] The specific calculation formula for the machining error limit is as follows:

[0026]

[0027] Where, Representative The allowable variation range of process parameters; Representative Tolerance bandwidth of each process parameter; Representative The participation coefficient of each process parameter in the current dimensional control;

[0028] The specific reasoning formula for the processing sequence is as follows:

[0029]

[0030] Where, Representative The processing priority of each processing feature; Representative Feature complexity evaluation of each processing feature; Representative The fixture accessibility score of each machining feature; Representative The distance between a machining feature and the datum; 、 as well as Represent the experience weight coefficients respectively.

[0031] As a further solution of the present invention, the specific steps of the processing modeling module to construct the processing model of the part are as follows:

[0032] S1.1: Based on the requirements analysis module, the processing characteristics obtained by analyzing the part processing requirements are analyzed, and standard processing modeling is performed on the product to be processed. Its geometry, dimensions, tolerance attributes and processing methods are defined. The processing process stages are divided according to the product form, equipment capabilities and clamping method after processing modeling.

[0033] S1.2: Calculate the compatibility of each product feature with each stage of the processing process, then sort all candidate combinations and select the combination with the highest compatibility as the optimal allocation solution. Then, assign an ideal processing angle to each processing feature and calculate the deviation between the ideal processing angle and the actual clamping. If the deviation exceeds the preset tolerance range, reset the clamping posture.

[0034] S1.3: The processing modeling module establishes the dependency order and spatial Boolean relationship between processing paths according to the spatial inclusion relationship, processing allowance dependency relationship and clamping dependency relationship of each processing feature, and generates a corresponding processing path topology map. Based on this processing path topology map, the envelope area of the processing path corresponding to each processing feature is constructed.

[0035] As a further solution of the present invention, the specific calculation formula of the standard processing modeling described in S1.1 is as follows:

[0036]

[0037] Where, Representative The volume of a feature; Representative The three-dimensional space area of the processing feature; Representative voxel density function of the three-dimensional spatial region of the processed feature;

[0038] The specific calculation formula for the adaptability described in S1.2 is as follows:

[0039]

[0040] Where, Representative machining features on the machine tool Arranged to the next process The adaptability; Representative The processing difficulty coefficient of each processing feature; Representative machine tools For the first Processing capability score of each processing feature; Representative processes For the first The adaptation factor of each processing feature;

[0041] The specific calculation formula for the deviation between the ideal machining angle and the actual clamping angle mentioned in S1.2 is as follows:

[0042]

[0043] Where, Representative The angle between the ideal machining angle of a machining feature and the current clamping direction; Representative Normal vector of each machined feature; Represents the preset processing direction vector.

[0044] As a further solution of the present invention, the specific steps of constructing a cross-material-machine-tool map by the knowledge modeling module are as follows:

[0045] S2.1: Extract text data from machine tool operation logs, historical processing documents, material technical manuals, CAD / CAM templates, and tool manuals, and construct a manually annotated sample set. Then, use NER technology to identify entities in each text data in the manually annotated sample set.

[0046] S2.2: Calculate the semantic matching confidence between the identified entity and the known processed semantic words. If the semantic matching confidence is higher than the preset threshold, the alignment is considered successful. Otherwise, the entity is marked as a new entity. The successfully aligned entity is then normalized into the data format used internally by the system.

[0047] S2.3: Calculate the machining compatibility between different entities. Establish relationship edges between corresponding entities based on the machining compatibility of each entity. Then, treat each group of entities as a node, and connect different sets of nodes through the established relationship edges. At the same time, the machining compatibility of each relationship edge is used as the weight value of each relationship edge to construct the corresponding cross-material-machine-tool-tool graph.

[0048] S2.4: After the graph is constructed, the semantic similarity of different entities in the cross-material-machine-tool-tool graph is calculated, and the corresponding similarity matrix is established. The entities and relationships in the cross-material-machine-tool-tool graph are then embedded into the vector space. The marked new entities are added to the cross-material-machine-tool-tool graph to expand the graph and update the processing adaptation strength, graph structure and embedding vector of the relationship edge.

[0049] As a further solution of the present invention, the specific steps of the parameter recommendation module for recommending initial processing parameters are as follows:

[0050] S3.1: Based on the machining model, material properties, and process requirements, an initial set of process states is set. Using a cross-material-machine-tool graph, the similarity between all historical combinations and the current process state is calculated. Historical combinations with a similarity higher than a preset threshold are selected as candidate "equipment-material-tool" combinations, i.e., migration sources, to establish the initial action space.

[0051] S3.2: Assign a corresponding migration weight to each migration source in the action space based on the similarity, and perform weighted fusion on multiple groups of migration sources based on the migration weight to generate an initial processing strategy for the current processing task, and then migrate the initial processing strategy to the current processing task;

[0052] S3.3: Calculate the immediate reward of the initial machining strategy based on machining accuracy, efficiency, and energy consumption, execute the state change of any machining parameter combination in the action space according to the machining state of the machining model module, and obtain the corresponding state transition probability;

[0053] S3.4: Based on the processing state, action space, state transition probability, and immediate reward, a corresponding parameter recommendation set is established. The strategy function is initialized based on the current parameter recommendation set. The current processing strategy is input into the processing model for simulation interaction to obtain the state transition trajectory. The initial processing strategy is used for the first simulation interaction.

[0054] S3.5: Based on the acquired state transition trajectory, the current policy parameters are optimized through the policy gradient function to maximize the expected reward. The processing strategy update, simulation interaction, and policy parameter optimization are repeated multiple times until the policy gradient function converges to the preset range after multiple rounds of iteration. The optimal processing parameter combination and equipment tool matching recommendations under the current process state are output.

[0055] As a further solution of the present invention, the specific manifestation of the initial process state described in S3.1 is as follows:

[0056]

[0057] Where, represents the initial process state vector, represents the material density, represents the geometric complexity coefficient, represents the quality requirement weight, Represents the current tool cutting capability coefficient;

[0058] The specific calculation formula for the similarity described in S3.1 is as follows:

[0059]

[0060] Where, Representative historical combination Similarity to the current process state, where Representative (equipment, materials, tools); Representative Dynamic stiffness score of each device; Representative The machinability index of a material; Representative The sharpness rating of each knife; Represents the target material density of the current process; Representative The standard density of a material;

[0061] The specific calculation formula for the migration weight described in S3.2 is as follows:

[0062]

[0063] Where, Representative and historical combination The corresponding migration weight; represents the similarity sensitivity factor; Represents a set of candidate historical tasks; Represents the current processing status Combined with history similarity; Represents the current processing status Combined with history similarity; as well as Obtained respectively through the above similarity calculation formula;

[0064] The specific calculation formula for the instant reward mentioned in S3.3 is as follows:

[0065]

[0066] Where, Representative Moment Total rewards; Representative Moment Satisfaction score of machining accuracy; Representative Moment processing time; Representative Moment Energy consumption index value; 、 as well as Respectively represent the user's preference coefficients for the current machining strategy's machining accuracy, efficiency, and energy consumption;

[0067] The specific form of the parameter recommendation set described in S3.4 is as follows:

[0068]

[0069] In the formula Representative The processing state space of the round iteration, Representative The action space of round iteration, Representative The state transition probability function of the round iteration, Representative The immediate reward function of the round iteration, Represents the discount factor.

[0070] As a further solution of the present invention, the error prediction module predicts the errors caused by various factors and generates a high-precision error distribution map in the following specific steps:

[0071] S4.1: The data perception module synchronously collects multi-source sensor data from multiple sensors and unifies the timestamps, frequencies, and scales of various sensor data. It extracts the modal features of each group of sensor data through the corresponding encoding network, and then uses the alignment network to project all modal features into a common latent space.

[0072] S4.2: Using the modal loss consistency function, unify the outputs of different modal features under the same input conditions. The aligned modal features are weighted summed to generate the final cross-modal shared features. The fused cross-modal shared features are then input into the generator, which uses the forward propagation algorithm to output the prediction error map corresponding to each cross-modal shared feature.

[0073] S4.3: Collect the true error map and input the predicted error map, true error map, current processing status, and cross-modal shared features into the discriminator. The discriminator processes each set of input data through a fully connected layer or a convolutional layer and outputs the authenticity probability of the error image. It then uses a binary adversarial loss function to calculate the error value between the discriminator and the generator, and updates the discriminator and generator parameters through the backpropagation algorithm.

[0074] S4.4: Repeatedly generate the prediction error map and update the parameters until the error values of the discriminator and the generator converge to the preset range. Then stop the iteration, input the latest cross-modal shared features into the generator, and generate the error map through the forward propagation of the generator. Then, through spatial smoothing and abnormal area hot spot amplification processing, generate a high-precision error distribution map.

[0075] Beneficial effects of the present invention:

[0076] The present invention extracts processing semantic entities from multi-source texts such as machine tool logs, process documents, and material manuals to construct a standardized knowledge graph, and then quantifies the processing adaptation relationship between entities to form a cross-material-machine tool-tool graph with semantic similarity and processing weights. Then, the migration source is screened according to the current process state and the graph similarity, and the initial processing strategy is generated through weighted fusion. The strategy parameters are continuously iteratively optimized with the help of transfer reinforcement learning, and the optimal parameter combination and equipment matching suggestions are output. At the same time, the multimodal generative adversarial network is used to fuse sensor data, extract modal features and align them to a unified latent space, and generate high-precision error distribution maps through adversarial training to provide support for error prediction and process adjustment. It realizes the structured integration and reuse of historical experience and multi-source data, can perform efficient knowledge transfer between different processing tasks, significantly improve the cold start capability and parameter optimization efficiency under new working conditions, can continuously improve processing performance based on real-time feedback, greatly reduce the cost of manual trial and error, achieve high-precision modeling of error distribution, and improve the consistency and usability between simulation and actual processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The present invention will be further described below with reference to the accompanying drawings.

[0078] Figure 1 This is a framework diagram of a CNC machining simulation system based on artificial intelligence. DETAILED DESCRIPTION

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0080] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0081] Example 1

[0082] The embodiment of the present invention provides a numerical control machining simulation system based on artificial intelligence. Figure 1 , Figure 1 A framework diagram of an artificial intelligence-based CNC machining simulation system provided by an embodiment of the present invention. The system includes a requirements analysis module, a machining modeling module, a data perception module, a knowledge modeling module, a parameter recommendation module, a path generation module, an error prediction module, a digital simulation module, a dynamic correction module, a risk detection module, and a decision support module.

[0083] The demand analysis module is used to receive the product's geometric design and quality requirements and analyze the part processing technology requirements.

[0084] Specifically, a feature extraction method based on boundary representation and topological rules is used to decompose the CAD model into a set of topological surfaces, and the connection relationship between the surfaces is extracted. The basic geometric units of the part surface are identified from the decomposed topological surfaces, including but not limited to holes, slots, steps and contours. The identified basic geometric units are classified as processing features, and the degree of influence of each processing feature on product performance is analyzed. If the degree of influence exceeds a preset threshold, the processing feature is marked as "critical" and is used as a key point to control the processing accuracy in subsequent processing steps, and the corresponding control is strengthened. The required processing level is derived based on the roughness parameters specified in the product drawings or quality specifications, and the corresponding tools, cutting parameters and processing methods are selected based on the obtained processing level. The tolerance range is then converted into the processing error limit of the corresponding product, and an initial relationship with the processing parameters is established. Finally, a reasonable processing sequence is inferred based on the product processing characteristics and clamping conditions.

[0085] In this embodiment, the specific calculation formula for the impact degree is as follows:

[0086]

[0087] Where, Representative Dimensional sensitivity of each machined feature; represents the performance indicator function; Representative Dimensional parameters of each machining feature; Representative The weight of each processing feature in the structure;

[0088] The specific calculation formula for processing level is as follows:

[0089]

[0090] Where, Representative The root mean square value of the roughness of the machined surface; represents the total number of sampling points; Representative Surface height at each sampling point; Represents the average value of the height of all sampling points;

[0091] The specific calculation formula for the machining error limit is as follows:

[0092]

[0093] Where, Representative The allowable variation range of process parameters; Representative Tolerance bandwidth of each process parameter; Representative The participation coefficient of each process parameter in the current dimensional control;

[0094] The specific reasoning formula for the processing sequence is as follows:

[0095]

[0096] Where, Representative The processing priority of each processing feature; Representative Feature complexity evaluation of each processing feature; Representative The fixture accessibility score of each machining feature; Representative The distance between a machining feature and the datum; 、 as well as Represent the experience weight coefficients respectively.

[0097] The processing modeling module builds the processing model of the part based on the geometric design and process requirements.

[0098] Specifically, based on the processing features obtained by analyzing the processing requirements of the parts by the demand analysis module, standard processing modeling is performed on the product to be processed, and its geometric shape, size, tolerance attributes and processing method are defined. According to the product shape, equipment capacity and clamping method after processing modeling, the processing process stages are divided, and the adaptability of each product feature to the processing process of each stage is calculated. Then, all candidate combinations are sorted, and the combination with the highest adaptability is selected as the optimal allocation scheme. After that, an ideal processing angle is specified for each processing feature, and the deviation between the ideal processing angle and the actual clamping is calculated. If the deviation exceeds the preset tolerance range, the clamping posture is reset. The processing modeling module establishes the dependency order and spatial Boolean relationship between the processing paths according to the spatial inclusion relationship, processing allowance dependency and clamping dependency of each processing feature, and generates a corresponding processing path topology map. Based on the processing path topology map, the envelope area of the processing path corresponding to each processing feature is constructed.

[0099] It should be further explained that the specific calculation formula for standard machining modeling is as follows:

[0100]

[0101] Where, Representative The volume of a feature; Representative The three-dimensional space area of the processing feature; Representative voxel density function of the three-dimensional spatial region of the processed feature;

[0102] The specific calculation formula for the adaptability described in S1.2 is as follows:

[0103]

[0104] Where, Representative machining features on the machine tool Arranged to the process The adaptability; Representative The processing difficulty coefficient of each processing feature; Representative machine tools For the first Processing capability score of each processing feature; Representative processes For the first The adaptation factor of each processing feature;

[0105] The specific calculation formula for the deviation between the ideal machining angle and the actual clamping angle mentioned in S1.2 is as follows:

[0106]

[0107] Where, Representative The angle between the ideal machining angle of a machining feature and the current clamping direction; Representative Normal vector of each machined feature; Represents the preset processing direction vector.

[0108] The data sensing module is used to collect sensor data including tool vibration, spindle temperature, stress and strain, and cutting force;

[0109] The knowledge modeling module integrates historical processing data, equipment performance, material mechanical properties and tool geometry information to build a cross-material-machine-tool map.

[0110] Specifically, the text data in machine tool operation logs, historical processing technology documents, material technical manuals, CAD / CAM templates and tool manuals are extracted, and a manually annotated sample set is constructed. The NER technology is then used to identify each entity in each text data in the manually annotated sample set, and the semantic matching confidence between the identified entity and the known processing semantic words is calculated. If the semantic matching confidence is higher than the preset threshold, it is considered to be successfully aligned. Otherwise, the entity is marked as a new entity, and the successfully aligned entity is normalized into the data format used within the system. The processing adaptation strength between different entities is calculated, and the corresponding processing adaptation strength of each entity is established according to the processing adaptation strength of each entity. The relationship edges between entities are constructed, and then each group of entities is used as a node, and different sets of nodes are connected through the established relationship edges. At the same time, the processing adaptation strength of each relationship edge is used as the weight value of each relationship edge to construct the corresponding cross-material-machine tool-tool graph. After the graph is constructed, the semantic similarity of different entities in the cross-material-machine tool-tool graph is calculated, and the corresponding similarity matrix is established. Then, the entities and relationships in the cross-material-machine tool-tool graph are embedded in the vector space, and the marked new entities are added to the cross-material-machine tool-tool graph to expand the graph and update the processing adaptation strength of the relationship edge, the graph structure and the embedding vector.

[0111] Example 2

[0112] The embodiment of the present invention provides a numerical control machining simulation system based on artificial intelligence. Figure 1 , Figure 1 A framework diagram of an artificial intelligence-based CNC machining simulation system provided by an embodiment of the present invention. The system includes a requirements analysis module, a machining modeling module, a data perception module, a knowledge modeling module, a parameter recommendation module, a path generation module, an error prediction module, a digital simulation module, a dynamic correction module, a risk detection module, and a decision support module.

[0113] The parameter recommendation module matches processing equipment, materials and tools based on process requirements, processing models and material-machine-tool-tool maps, and recommends initial processing parameters.

[0114] Specifically, based on the processing model, material properties and process requirements, an initial process state set is set, and the similarity between all historical combinations and the current process state is calculated using a cross-material-machine-tool-tool map. The historical combinations with a combination similarity higher than a preset threshold are used as candidate combinations of "equipment-material-tool", i.e., migration sources, to establish an initial action space. A corresponding migration weight is assigned to each migration source in the action space according to the similarity, and multiple groups of migration sources are weightedly fused according to the migration weight to generate an initial processing strategy for the current processing task, and the initial processing strategy is migrated to the current processing task. The instant reward of the initial processing strategy is calculated based on the processing accuracy, efficiency and energy consumption, and the reward after executing any processing parameter combination in the action space according to the processing state of the processing model module is calculated. State changes, and obtain the corresponding state transition probability. According to the processing state, action space, state transition probability and immediate reward, the corresponding parameter recommendation set is established, and the policy function is initialized based on the current parameter recommendation set. The current processing strategy is input into the processing model for simulation interaction to obtain the state transition trajectory. The initial processing strategy is used for interaction during the first simulation interaction. Based on the obtained state transition trajectory, the current policy parameters are optimized through the policy gradient function to maximize the expected reward. The processing strategy update, simulation interaction and policy parameter optimization are repeated many times until the policy gradient function converges to the preset range after multiple rounds of iteration, and the optimal processing parameter combination and equipment tool matching recommendation under the current process state are output.

[0115] It should be noted that the specific manifestations of the initial process state are as follows:

[0116]

[0117] Where, represents the initial process state vector, represents the material density, represents the geometric complexity coefficient, represents the quality requirement weight, Represents the current tool cutting capability coefficient;

[0118] The specific calculation formula for the similarity described in S3.1 is as follows:

[0119]

[0120] Where, Representative historical combination Similarity to the current process state, where Representative (equipment, materials, tools); Representative Dynamic stiffness rating of each device; Representative The machinability index of a material; Representative The sharpness rating of each knife; Represents the target material density of the current process; Representative The standard density of a material;

[0121] The specific calculation formula for the migration weight described in S3.2 is as follows:

[0122]

[0123] Where, Representative and historical combination The corresponding migration weight; represents the similarity sensitivity factor; Represents a set of candidate historical tasks; Represents the current processing status Combined with history similarity; Represents the current processing status Combined with history similarity; as well as Obtained respectively through the above similarity calculation formula;

[0124] The specific calculation formula for the instant reward mentioned in S3.3 is as follows:

[0125]

[0126] Where, Representative Moment Total rewards; Representative Moment Satisfaction score of machining accuracy; Representative Moment processing time; Representative Moment Energy consumption index value; 、 as well as Respectively represent the user's preference coefficients for the current machining strategy's machining accuracy, efficiency, and energy consumption;

[0127] The specific form of the parameter recommendation set described in S3.4 is as follows:

[0128]

[0129] In the formula Representative The processing state space of the round iteration, Representative The action space of round iteration, Representative The state transition probability function of the round iteration, Representative The immediate reward function of the round iteration, Represents the discount factor.

[0130] The path generation module combines the processing model and initial processing parameters to generate the initial processing path and plans the path in layers; the error prediction module predicts the errors caused by various factors based on the sensor data and the processing path, and generates a high-precision error distribution map.

[0131] Specifically, the data perception module synchronously collects multi-source sensor data from multiple sensors, and unifies the timestamps, frequencies and scales of various sensor data. The modal features of each group of sensor data are extracted through the corresponding encoding network. Then, all modal features are projected into a common latent space using the alignment network. The outputs of different modal features under the same input conditions are unified through the modal loss consistency function. The aligned modal features are weighted summed to generate the final cross-modal shared features. The fused cross-modal shared features are then input into the generator. The generator outputs the prediction error map corresponding to each cross-modal shared feature through the forward propagation algorithm, collects the true error map, and combines the prediction error map and the true error map. The actual error map, current processing status and cross-modal shared features are input into the discriminator. The discriminator processes each set of input data through a fully connected layer or a convolutional layer, and outputs the authenticity probability of the error image. The error value of the discriminator and the generator is then calculated using a binary adversarial loss function, and the discriminator and generator parameters are updated through the back-propagation algorithm. The prediction error map generation and parameter update are repeated until the error value of the discriminator and the generator converges to a preset range. The iteration is stopped, and the latest cross-modal shared features are input into the generator, and the error map is generated through forward propagation of the generator. A high-precision error distribution map is generated through spatial smoothing and abnormal area hot zone amplification.

[0132] The digital simulation module simulates the machining process in real time based on various sensor data, machining paths, error information, and machining parameters, verifies the rationality of the path, the validity of the parameters, and the conformity of the machining results, and outputs a simulation report; the dynamic correction module fine-tunes and replans the machining path based on real-time simulation data and error prediction results; the risk detection module analyzes the real-time machining path, generates a collision heat map, and performs visual monitoring and retrospective analysis of potential collision risks; the auxiliary decision-making module assists in optimizing the current machining parameters based on simulation data.

[0133] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A numerical control machining simulation system based on artificial intelligence, characterized in that: include: Demand analysis module, processing modeling module, data perception module, knowledge modeling module, parameter recommendation module, path generation module, error prediction module, digital simulation module, dynamic correction module, risk detection module and decision support module; The demand analysis module is used to receive the geometric design and quality requirements of the product and analyze the part processing technology requirements; The processing modeling module constructs a processing model of the part based on the geometric design and process requirements; The data sensing module is used to collect sensor data including tool vibration, spindle temperature, stress and strain, and cutting force; The knowledge modeling module integrates historical processing data, equipment performance, material mechanical properties and tool geometry information to build a cross-material-machine-tool map; The parameter recommendation module matches processing equipment, materials and tools according to process requirements, processing models and material-machine-tool-tool maps, and recommends initial processing parameters; The path generation module generates an initial processing path by combining the processing model and the initial processing parameters, and plans the path in layers; The error prediction module predicts the errors caused by various factors based on various sensor data and processing paths, and generates a high-precision error distribution map; The digital simulation module simulates the machining process in real time based on the sensor data, machining path, error information and machining parameters, verifies the rationality of the path, the validity of the parameters and the conformity of the machining results, and outputs a simulation report; The dynamic correction module fine-tunes and replans the machining path based on real-time simulation data and error prediction results; The risk detection module analyzes the real-time processing path, generates a collision heat map, and performs visual monitoring and retrospective analysis of potential collision risks; The auxiliary decision module assists in optimizing the current processing parameters based on the simulation data.

2. The artificial intelligence-based numerical control machining simulation system according to claim 1, characterized in that: The specific steps of the processing modeling module to construct the processing model of the part are as follows: S1.1: Based on the requirements analysis module, the processing characteristics obtained by analyzing the part processing requirements are analyzed, and standard processing modeling is performed on the product to be processed. Its geometry, dimensions, tolerance attributes and processing methods are defined. The processing process stages are divided according to the product form, equipment capabilities and clamping method after processing modeling. S1.2: Calculate the compatibility of each product feature with each stage of the processing process, then sort all candidate combinations and select the combination with the highest compatibility as the optimal allocation solution. Then, assign an ideal processing angle to each processing feature and calculate the deviation between the ideal processing angle and the actual clamping. If the deviation exceeds the preset tolerance range, reset the clamping posture. S1.3: The processing modeling module establishes the dependency order and spatial Boolean relationship between processing paths according to the spatial inclusion relationship, processing allowance dependency relationship and clamping dependency relationship of each processing feature, and generates a corresponding processing path topology map. Based on this processing path topology map, the envelope area of the processing path corresponding to each processing feature is constructed.

3. The artificial intelligence-based numerical control machining simulation system according to claim 2, characterized in that: The specific calculation formula for the standard processing modeling described in S1.1 is as follows: Where, Representative The volume of a feature; Representative The three-dimensional space area of the processing feature; Representative voxel density function of the three-dimensional spatial region of the processed feature; The specific calculation formula for the adaptability described in S1.2 is as follows: Where, Representative machining features on the machine tool Arranged to the next process The adaptability; Representative The processing difficulty coefficient of each processing feature; Representative machine tools For the first Processing capability score of each processing feature; Representative processes For the first The adaptation factor of each processing feature; The specific calculation formula for the deviation between the ideal machining angle and the actual clamping angle mentioned in S1.2 is as follows: Where, Representative The angle between the ideal machining angle of a machining feature and the current clamping direction; Representative Normal vector of each machined feature; Represents the preset processing direction vector.

4. The artificial intelligence-based numerical control machining simulation system according to claim 2, characterized in that: The specific steps of constructing the cross-material-machine-tool graph by the knowledge modeling module are as follows: S2.1: Extract text data from machine tool operation logs, historical processing documents, material technical manuals, CAD / CAM templates, and tool manuals, and construct a manually annotated sample set. Then, use NER technology to identify entities in each text data in the manually annotated sample set. S2.2: Calculate the semantic matching confidence between the identified entity and the known processed semantic words. If the semantic matching confidence is higher than the preset threshold, the alignment is considered successful. Otherwise, the entity is marked as a new entity. The successfully aligned entity is then normalized into the data format used internally by the system. S2.3: Calculate the machining compatibility between different entities. Establish relationship edges between corresponding entities based on the machining compatibility of each entity. Then, treat each group of entities as a node, and connect different sets of nodes through the established relationship edges. At the same time, the machining compatibility of each relationship edge is used as the weight value of each relationship edge to construct the corresponding cross-material-machine-tool-tool graph. S2.4: After the graph is constructed, the semantic similarity of different entities in the cross-material-machine-tool-tool graph is calculated, and the corresponding similarity matrix is established. The entities and relationships in the cross-material-machine-tool-tool graph are then embedded into the vector space. The marked new entities are added to the cross-material-machine-tool-tool graph to expand the graph and update the processing adaptation strength, graph structure and embedding vector of the relationship edge.

5. The artificial intelligence-based numerical control machining simulation system according to claim 4, characterized in that: The specific steps of the parameter recommendation module for recommending initial processing parameters are as follows: S3.1: Based on the machining model, material properties, and process requirements, an initial set of process states is set. Using a cross-material-machine-tool graph, the similarity between all historical combinations and the current process state is calculated. Historical combinations with a similarity above a preset threshold are selected as candidate "equipment-material-tool" combinations, i.e., migration sources, to establish the initial action space. S3.2: Assign a corresponding migration weight to each migration source in the action space based on the similarity, and perform weighted fusion on multiple groups of migration sources based on the migration weight to generate an initial processing strategy for the current processing task, and then migrate the initial processing strategy to the current processing task; S3.3: Calculate the immediate reward of the initial machining strategy based on machining accuracy, efficiency, and energy consumption, execute the state change of any machining parameter combination in the action space according to the machining state of the machining model module, and obtain the corresponding state transition probability; S3.4: Based on the processing state, action space, state transition probability, and immediate reward, a corresponding parameter recommendation set is established. The strategy function is initialized based on the current parameter recommendation set. The current processing strategy is input into the processing model for simulation interaction to obtain the state transition trajectory. The initial processing strategy is used for the first simulation interaction. S3.5: Based on the acquired state transition trajectory, the current policy parameters are optimized through the policy gradient function to maximize the expected reward. The processing strategy update, simulation interaction, and policy parameter optimization are repeated multiple times until the policy gradient function converges to the preset range after multiple rounds of iteration. The optimal processing parameter combination and equipment tool matching recommendations under the current process state are output.

6. The artificial intelligence-based numerical control machining simulation system according to claim 5, characterized in that: The specific manifestations of the initial process state described in S3.1 are as follows: Where, represents the initial process state vector, represents the material density, represents the geometric complexity coefficient, represents the quality requirement weight, Represents the current tool cutting capability coefficient; The specific calculation formula for the similarity described in S3.1 is as follows: Where, Representative historical combination Similarity to the current process state, where Representative (equipment, materials, tools); Representative Dynamic stiffness score of each device; Representative The machinability index of a material; Representative The sharpness rating of each knife; Represents the target material density of the current process; Representative The standard density of a material; The specific calculation formula for the migration weight described in S3.2 is as follows: Where, Representative and historical combination The corresponding migration weight; represents the similarity sensitivity factor; Represents a set of candidate historical tasks; Represents the current processing status Combined with history similarity; Represents the current processing status Combined with history similarity; as well as Obtained respectively through the above similarity calculation formula; The specific calculation formula for the instant reward mentioned in S3.3 is as follows: Where, Representative Moment Total rewards; Representative Moment Satisfaction score of machining accuracy; Representative Moment processing time; Representative Moment Energy consumption index value; 、 as well as Respectively represent the user's preference coefficients for the current machining strategy's machining accuracy, efficiency, and energy consumption; The specific form of the parameter recommendation set described in S3.4 is as follows: In the formula Representative The processing state space of the round iteration, Representative The action space of round iteration, Representative The state transition probability function of the round iteration, Representative The immediate reward function of the round iteration, Represents the discount factor.

7. The artificial intelligence-based numerical control machining simulation system according to claim 5, characterized in that: The specific steps of the error prediction module predicting the errors caused by various factors and generating a high-precision error distribution map are as follows: S4.1: The data perception module synchronously collects multi-source sensor data from multiple sensors and unifies the timestamps, frequencies, and scales of various sensor data. It extracts the modal features of each group of sensor data through the corresponding encoding network, and then uses the alignment network to project all modal features into a common latent space. S4.2: Using the modal loss consistency function, unify the outputs of different modal features under the same input conditions. The aligned modal features are weighted summed to generate the final cross-modal shared features. The fused cross-modal shared features are then input into the generator, which uses the forward propagation algorithm to output the prediction error map corresponding to each cross-modal shared feature. S4.3: Collect the true error map and input the predicted error map, true error map, current processing status, and cross-modal shared features into the discriminator. The discriminator processes each set of input data through a fully connected layer or a convolutional layer and outputs the authenticity probability of the error image. It then uses a binary adversarial loss function to calculate the error value between the discriminator and the generator, and updates the discriminator and generator parameters through the backpropagation algorithm. S4.4: Repeatedly generate the prediction error map and update the parameters until the error values of the discriminator and the generator converge to the preset range. Then stop the iteration, input the latest cross-modal shared features into the generator, and generate the error map through the forward propagation of the generator. Then, through spatial smoothing and abnormal area hot spot amplification processing, generate a high-precision error distribution map.

Images