A method and system for constructing twin machine tool models for cutting parameter optimization
By constructing a multi-level component relationship tree and improving the kernel extreme learning machine model, combined with XML process knowledge ontology, the problems of insufficient dynamic motion simulation accuracy and heterogeneous and scattered process data in twin machine tool modeling were solved, and efficient optimization and closed-loop control of cutting parameters were achieved.
Patent Information
- Application Number
- CN202611628879.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing twin machine tool modeling ignores the hierarchical topology and kinematic chain transmission logic of machine tool components when replicating the geometric shape, resulting in insufficient accuracy of dynamic motion simulation, heterogeneous and scattered cutting process data with missing semantic associations, and inability to achieve highly reliable optimization of cutting parameters.
A topological model of a multi-level component relationship tree is constructed, which integrates multi-dimensional perception data of tool wear state and machining surface morphology. An improved kernel limit learning machine is used to train the cutting process prediction model, and multi-objective cutting parameter optimization is performed by combining CNC process knowledge ontology with XML description language.
It realizes a highly reliable digital carrier of twin machine tool models, improves the accuracy of virtual simulation iteration and dynamic optimization capability of cutting parameters, solves the problems of insufficient accuracy of traditional twin modeling and heterogeneous and scattered process data, and forms a closed-loop optimization system for cutting parameters.
Smart Images

Figure CN123113408A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of CNC machine tool digital twin and cutting process optimization technology, and particularly relates to a method and system for constructing a twin machine tool model for cutting parameter optimization. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] As CNC machining develops towards higher precision, higher efficiency, and greater intelligence, cutting parameter optimization has become a core means to improve machining quality, extend tool life, and reduce production costs. Digital twin technology, by constructing a digital mirror image of the physical machine tool, can realize virtual simulation and parameter iteration of the cutting process, providing a new technical path for cutting parameter optimization.
[0004] However, existing twin machine tool modeling technologies often focus on geometric replication, neglecting the hierarchical topology of machine tool components and the logic of kinematic chain transmission. This only achieves static geometric matching, making it difficult to guarantee the accuracy of dynamic motion simulation and failing to provide a highly reliable digital experimental platform for virtual optimization of cutting parameters. Secondly, cutting process data encompasses multi-source heterogeneous information, including static baseline parameters, dynamic process characteristics, and implicit process knowledge. Traditional distributed storage methods suffer from data silos, missing semantic relationships, and poor reusability, failing to provide systematic data support for cutting process modeling. Furthermore, cutting process modeling relies heavily on theoretical cutting formulas and empirical models, making it difficult to adapt to the nonlinear effects of time-varying tool wear characteristics on the surface morphology of the machined surface. This results in poor adaptability to operating conditions, insufficient prediction accuracy, and an inability to support dynamic real-time parameter optimization. Summary of the Invention
[0005] To overcome the shortcomings of the existing technologies, this invention proposes a twin machine tool model construction method and system for cutting parameter optimization. This addresses the problems of insufficient accuracy in traditional twin modeling, heterogeneous and scattered process data, reliance on experience in cutting process modeling, and insufficient depth of virtual-real fusion. It constructs a closed-loop optimization system for cutting parameters based on "perception-modeling-optimization-feedback", which can be applied to virtual iteration of process parameters, online prediction of cutting status, and closed-loop control of machining quality in CNC machining processes.
[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, this invention discloses a method for constructing a twin machine tool model for cutting parameter optimization, comprising: A topological structure model of a multi-level component relationship tree is constructed, and a twin machine tool model is obtained by static digital model comparison calibration and dynamic motion simulation verification of parts at each level. Construct a static process parameter library, a dynamic cutting feature set, a comprehensive process knowledge graph, and a modular indexing mechanism; By integrating multi-dimensional sensing data on tool wear state and machined surface morphology, a dynamic mapping matrix between the two is constructed, and a cutting process prediction model is obtained by training with an improved kernel limit learning machine. A CNC process knowledge ontology based on XML description language is constructed to quantify the credibility of the simulation results of the twin machine tool model. Based on the twin machine tool model and the cutting process prediction model, multi-objective cutting parameters are optimized to obtain the optimal cutting parameters.
[0007] Secondly, this invention discloses a twin machine tool model construction system for cutting parameter optimization, comprising: The twin model construction module is configured to: construct a topological structure model of a multi-level component relationship tree, and perform static digital model comparison calibration and dynamic motion simulation verification on the parts at each level to obtain a twin machine tool model; The data system construction module is configured to: build a static process parameter library, a dynamic cutting feature set, a comprehensive process knowledge graph, and a modular indexing mechanism; The cutting process modeling module is configured to: fuse multi-dimensional perception data of tool wear state and machined surface morphology, construct a dynamic mapping matrix between the two, and train an improved kernel extreme learning machine to obtain a cutting process prediction model; The cutting parameter optimization module is configured to: construct a CNC process knowledge ontology based on XML description language, quantify the credibility of the simulation results of the twin machine tool model, and perform multi-objective cutting parameter optimization based on the twin machine tool model and the cutting process prediction model to obtain the optimal cutting parameters.
[0008] Thirdly, the present invention discloses an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when run by the processor, complete the steps of the above-mentioned twin machine tool model construction method for cutting parameter optimization.
[0009] Fourthly, the present invention discloses a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described method for constructing a twin machine tool model for cutting parameter optimization.
[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention employs a two-stage modeling method combining digital model comparison and motion simulation. First, geometric dimension deviations are corrected through static calibration, and then motion transmission consistency is verified through dynamic simulation. This approach balances the static geometric accuracy and dynamic motion characteristics of the twin model, providing a highly reliable digital carrier for the virtual simulation iteration of cutting parameters.
[0011] This invention achieves structured and integrated management of three types of heterogeneous process data—static baseline, dynamic state, and process knowledge—through a three-dimensional classification system and modular indexing mechanism, providing standardized and reusable data support for subsequent cutting process modeling.
[0012] This invention employs a data-driven cutting process modeling method, which breaks away from the dependence of traditional theoretical models on prior cutting formulas and empirical parameters. It can dynamically update the model based on real-time sensing data, thereby achieving accurate dynamic representation of the cutting process.
[0013] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0015] Figure 1 This is a flowchart of the twin machine tool model construction method for cutting parameter optimization as described in Embodiment 1 of the present invention. Detailed Implementation
[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0017] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.
[0018] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.
[0019] Example 1 In one or more embodiments, a method for constructing a twin machine tool model for cutting parameter optimization is disclosed, comprising four core steps: multi-level twin machine tool model construction and two-stage calibration, construction and modular management of a three-dimensional heterogeneous process data system, dynamic mapping of tool wear and surface morphology and improved kernel limit learning machine cutting process modeling, and construction of an XML process knowledge ontology and simultaneous virtual and real cutting parameter optimization. Figure 1 As shown, the specific steps include the following: Step S101: Construct a multi-level component relationship tree for the machine tool, disassemble the machine tool structure from the system level, clarify the component topology and motion logic, and provide a structured model skeleton for subsequent calibration and simulation; then, through static digital model comparison and dynamic motion simulation dual-stage calibration, gradually correct the model deviation, and finally establish a full-dimensional accurate mapping relationship between the physical machine tool and the digital image to obtain a high-fidelity twin machine tool model.
[0020] To standardize the description of the hierarchical topology and relationships of machine tool components, the entire machine tool is decomposed into multi-level nodes from top to bottom, and the constraints and transitive relationships between nodes are defined to construct a topological structure model of a multi-level component relationship tree, as shown in Formula 1: (1) in, A multi-level component relationship tree for machine tools; This is the root node of the entire machine, representing the overall machine tool system; It is a collection of functional component nodes, including core functional components such as the spindle system, feed system, bed, and tool magazine; This is a set of core component nodes, corresponding to the key components under each functional unit; It is a set of kinematic joint nodes, including kinematic connection units such as revolute joints and prismatic joints; This is a set of assembly constraint relationships and motion transmission relationships between nodes at each level.
[0021] Based on a clear understanding of the machine tool component hierarchy and motion logic, dimensional calibration of parts at each level is necessary to correct geometric dimensional deviations between the digital model and the physical entity and ensure the static accuracy of the twin model. Since different levels of parts have varying degrees of influence on the final machining accuracy, the deviation propagation effect is more significant for parts closer to the execution end. Therefore, a hierarchical calibration coefficient is introduced to weightedly correct the deviations, and a static calibration model is constructed as follows: (2) in, For the first time after calibration The digital model dimensions of each part; For the first in the initial digital model The nominal dimensions of each part; For physical machine tools The deviation between the measured dimensions and the initial nominal dimensions of each part; The calibration coefficient is determined based on the hierarchical weight of the part in the component relationship tree and the degree of assembly correlation. The closer the part is to the end of the movement, the higher the calibration coefficient.
[0022] After completing static geometric dimension calibration, the twin model only achieves external dimensional matching and cannot guarantee dynamic consistency during motion. Cutting parameter optimization relies on accurate simulation of the machine tool's motion process; therefore, further modeling and verification of the machine tool's kinematic chain transmission characteristics are required. First, the transformation relationships of each kinematic pair in the kinematic chain and the impact of assembly coupling are quantified using the motion transfer matrix. The overall machine motion transfer matrix is obtained by sequentially coupling the matrices of each component: (3) in, This is the motion transfer matrix of the whole machine, which represents the transfer relationship from the input motion to the end effector; This represents the total number of moving links in a kinetic chain. For the first The motion transformation matrix of a kinematic pair describes the displacement and velocity transformation relationship of that kinematic pair; For the first The assembly constraint coupling matrix of each link characterizes the coupling effect of assembly errors and gaps on motion transmission.
[0023] The motion pose of the end effector can be simulated in a twin model based on the motion transfer matrix. To quantify the dynamic motion consistency between the twin model and the physical machine tool, it is necessary to calculate the deviation between the physically measured pose and the virtual simulated pose under the same motion command. The L2 norm is used to characterize the magnitude of the pose deviation, and a dynamic consistency verification function is constructed by averaging the deviation from multiple sampling points. This function serves as the evaluation basis for dynamic calibration in the two-stage modeling process. (4) in, This represents the dynamic consistency error value. This represents the total number of sampling points during the motion process; For the first The measured pose vector of the physical machine tool end effector at each sampling time; This is the pose simulation vector of the end effector in the twin model at the corresponding time point; It is the vector L2 norm, used to characterize the magnitude of pose deviation.
[0024] When the dynamic consistency error value is less than the preset accuracy threshold, the static geometric accuracy and dynamic motion accuracy of the twin model are considered to meet the requirements, thus completing the construction of the high-fidelity twin machine tool model and providing a reliable digital carrier for subsequent virtual simulation and parameter optimization of the cutting process. The accuracy threshold is preset according to the machine tool's machining accuracy level and machining conditions. For example, for a typical vertical CNC milling machining scenario, the accuracy threshold can be set to 0.02 mm.
[0025] This embodiment constructs a multi-level component relationship tree architecture of "complete machine - functional components - core parts - kinematic pairs", which clarifies the assembly constraints and motion transmission logic of each level of machine tool components, realizes the complete digital expression of machine tool structure topology and kinematic chain, and makes up for the defects of traditional geometric modeling lack of hierarchy and unclear topological logic; By constructing a motion transfer matrix and a dynamic consistency verification function, the transmission law and simulation error of the motion chain are quantified, enabling quantitative evaluation and iterative calibration of the twin model's accuracy and improving the reliability of motion simulation.
[0026] Step S102: Construct a static process parameter library, a dynamic cutting feature set, a comprehensive process knowledge graph, and a modular indexing mechanism.
[0027] In response to the heterogeneous nature of cutting process information, this step constructs a classification system from three dimensions: static benchmark, dynamic state, and process knowledge. This enables structured and modular management of process data, providing systematic data support for cutting process modeling.
[0028] First, a static process parameter library is constructed to solidify the baseline static information during the processing, serving as the foundational layer of the entire process data system. Its structured representation is shown in Formula 5: (5) in, This is a static process parameter library; This is a subset of basic machine tool parameters, including information such as machine tool model, motion axis parameters, and accuracy class. This is a subset of tool parameters, containing information such as tool type, geometric parameters, and material properties. This is a subset of workpiece parameters, including information such as workpiece material, dimensions, and blank condition. It is a subset of the reference cutting parameters, including the recommended cutting speed, feed rate, depth of cut, and other reference parameters under different working conditions.
[0029] The static process parameter library stores the baseline static information of the machining process and forms the basic data layer of the process system. However, the cutting process is dynamic and time-varying, and static parameters alone cannot characterize the real-time state of the process. Therefore, it is necessary to construct a dynamic cutting feature set to extract features and perform structured representation of the real-time state signals during the machining process. At each sampling time, the multi-source sensor signals are extracted into a unified feature vector and stored sequentially on a time-time basis. (6) in, for The dynamic cutting feature vector at any given time; for The characteristic value of the cutting force at any given time is the effective value of the main cutting force component; for The vibration characteristic value at time t is the root mean square value of the vibration acceleration; for The characteristic value of the cutting zone temperature at any given time; for The characteristic value of the acoustic emission signal at time t.
[0030] Static parameters and dynamic features cover baseline information and process states respectively, but they lack semantic-level logical connections, making it difficult to support the reasoning and reuse of process knowledge. Therefore, it is necessary to construct a comprehensive process knowledge graph, defining process entities, inter-entity relationships, and entity attributes using a triplet structure, to uncover the causal relationships and semantic connections between process parameters, cutting states, and machining quality. (7) in, For the synthesis process knowledge graph; It is a collection of entities related to the technological process, including various entities such as process parameters, cutting conditions, and machining quality; It is a set of semantic relationships between entities, including causal relationships, influence relationships, correspondence relationships, etc. It is a set of attributes for an entity, describing the numerical characteristics and category attributes of each entity.
[0031] The three-dimensional classification system enables the classified storage of heterogeneous data, but data from different dimensions is scattered across different databases, requiring cross-database matching for retrieval, which is inefficient. To achieve correlated retrieval and rapid access to multi-dimensional process data, a unified modular index function needs to be constructed to weightedly fuse the classification indexes of the three types of data, obtaining a comprehensive index value for each data entry. This will enable rapid data location and correlated access. (8) in, For the first The comprehensive index value of the process data; This is the category index value of the data in the static process parameter library; This is the classification index value of the data in the dynamic cutting feature set; This is the classification index value of the data in the knowledge graph of the integration process; These are the weighting coefficients for the three types of indexes, which are dynamically adjusted according to the needs of the process scenario.
[0032] This embodiment constructs a three-dimensional classification system consisting of a static process parameter library, a dynamic cutting feature set, and a comprehensive process knowledge graph. This system enables the classified storage and structured management of multi-source heterogeneous process information, breaks down the data silos of traditional distributed storage, and improves the organization of process data. By using modular indexing functions for process data, the system enables the retrieval and rapid access of static parameters, dynamic features, and process knowledge, thereby improving the reusability and access efficiency of process data and providing systematic data support for cutting process modeling. The integrated process knowledge graph establishes semantic relationships and causal logic between process parameters, cutting conditions and machining quality, and uncovers implicit knowledge in the process, providing knowledge-level guidance for parameter optimization.
[0033] Step S103: Integrate multi-dimensional perception data of tool wear state and machining surface morphology, construct a dynamic mapping matrix between the two, and use an improved kernel limit learning machine to train and obtain a cutting process prediction model.
[0034] First, tool wear is a core time-varying factor affecting the morphology of machined surfaces. To quantify the relationship between the two, tool wear state is vectorized, integrating the two core indicators of wear amount and remaining life into a unified state vector: (9) in, for The tool wear state vector at any given time; for The amount of wear on the flank face of the cutting tool at any given time; for The remaining tool life at any given time is calculated from the wear amount and wear rate.
[0035] After completing the vectorized representation of tool wear state, it is also necessary to construct a unified evaluation vector for the quality index of the machined surface, and establish a correspondence with the wear state vector, so as to provide a standardized output representation for subsequent mapping modeling: (10) in, for The evaluation vector of the processed surface morphology at any given time; The arithmetic mean deviation of the profile of the machined surface; is the macroscopic defect rate of the processed surface, and is the proportion of the defect area to the total processed area.
[0036] Based on the vectorized representation of the two types of features, since the impact of tool wear on surface quality during cutting dynamically changes with working conditions and time, the fixed-coefficient model cannot accurately describe its time-varying nonlinear characteristics. Therefore, a dynamic matrix mapping relationship is constructed. This time-varying mapping matrix quantifies the influence of each wear feature on each morphology index at different times, uncovering the time-varying nonlinear correlation between tool wear and the machined surface morphology. This provides a mechanistic prior for improving the feature selection of the kernel limit learning machine model. Simultaneously, this wear-morphology mapping relationship is included as a sub-relation in the subsequent element mapping relationship set of the XML process knowledge ontology. Furthermore, a random error vector is introduced to represent the interference of other unquantified factors. (11) in, for The dynamic mapping matrix at any time is a 2×2 matrix. The matrix elements represent the influence coefficients of each wear feature on each morphological index, and are dynamically updated with cutting conditions and time. This is a random error vector, representing the interference of unquantified factors.
[0037] Dynamic matrix mapping clarifies the correlation between tool wear and surface quality. However, the cutting process involves multiple input sources and multiple target outputs, exhibiting strong nonlinearity and time-varying characteristics. Matrix mapping alone is insufficient for achieving high-precision process prediction. Kernel extreme learning machines (KEMs) offer advantages such as fast training speed and strong generalization ability, but traditional single kernel functions struggle to simultaneously balance local fitting accuracy and global generalization capability. Therefore, this invention proposes an improved hybrid kernel function that integrates the local fitting characteristics of radial basis kernels with the global generalization characteristics of polynomial kernels. The weighting coefficients adjust the proportion of these two kernels to adapt to the modeling needs of different cutting conditions. (12) in, To improve the hybrid kernel function in samples and The value at; The kernel function weight coefficients have a value range of [0,1] and are used to balance the proportions of local and global kernels. It is a radial basis kernel function, possessing strong local fitting ability; It is a polynomial kernel function, possessing strong global generalization ability; The input sample vector.
[0038] Based on an improved hybrid kernel function, training samples and hidden layer nodes are respectively substituted into the kernel function for calculation, constructing a hidden layer kernel output matrix. Each row of this matrix corresponds to a training sample, each column corresponds to a hidden layer node, and the matrix elements are the kernel function calculation results for the corresponding sample and node, thereby achieving a non-linear mapping of input features. (13) in, This is the output matrix of the hidden layer kernel; The total number of training samples; is the number of hidden layer nodes; the in the matrix is the . Line 1 The column element is the first The training sample and the first The kernel function calculation value corresponding to each hidden layer node.
[0039] In kernel extreme learning machines, output weights can be solved using the least squares method. However, the number of training samples in machining scenarios is usually limited, and direct calculation can easily lead to overfitting, resulting in decreased model generalization performance. To suppress overfitting and improve prediction stability with small samples, an L2 regularization term is introduced in the weight calculation process. The regularization coefficient controls the penalty strength. The improved formula for calculating output weights is as follows: (14) in, This is the output weight matrix of the model; This is the regularization coefficient, used to control the strength of regularization; It is the identity matrix; This is the expected output matrix of the training samples, corresponding to the cutting state and machining quality index.
[0040] After obtaining the optimal output weights using the above method, the training of the improved kernel extreme learning machine model is complete, resulting in the cutting process prediction model. For any input cutting parameters and state feature vector, substituting them and all training samples into the improved hybrid kernel function for calculation, and then performing operations with the output weight matrix, yields the corresponding cutting process prediction result. (15) in, For input vectors The corresponding cutting process prediction output vector includes predicted cutting force, vibration, temperature, surface roughness, defect rate and other indicators; For the training sample set; The kernel function is used to compute the value vector of the input vector and all training samples.
[0041] The cutting process prediction model takes cutting parameter vectors, tool wear state vectors, and dynamic cutting feature vectors as inputs, and outputs predicted surface morphology and cutting state values. The cutting parameter vectors are taken from a subset of benchmark cutting parameters in a static process parameter library. The cutting parameter vectors are input variables of the model and are not limited to the benchmark parameters already stored in the library. Optimized candidate parameters can also be used to construct the cutting parameter vector input model. The data processing flow is as follows: sample data preprocessing, constructing an improved hybrid kernel function, calculating the hidden layer kernel output matrix, solving for regularized output weights, offline model training, and online real-time prediction. This enables fast and high-precision prediction of the cutting process, providing core support for cutting parameter optimization.
[0042] This embodiment constructs a dynamic matrix mapping relationship between tool wear state and machined surface morphology, quantifies the nonlinear influence of time-varying tool wear characteristics on surface machining quality, clarifies the intrinsic correlation mechanism between the two, and provides a feature correlation basis for cutting process modeling; We propose a hybrid kernel function-improved kernel limit learning machine, which combines the local fitting accuracy of the radial basis kernel with the global generalization ability of the polynomial kernel. This can adapt to the modeling needs of different cutting conditions and improve the model's adaptability to different conditions. Introducing a regularization term into the weight calculation of the kernel limit learning machine effectively suppresses the overfitting problem and improves the stability of the prediction of cutting state and machining quality under small sample conditions.
[0043] Step S104: Construct a CNC process knowledge ontology based on XML description language, quantify the credibility of the simulation results of the twin machine tool model, and perform multi-objective cutting parameter optimization based on the twin machine tool model and the cutting process prediction model to obtain the optimal cutting parameters.
[0044] To unify the description standards of process knowledge in both virtual and physical spaces and resolve the semantic differences in knowledge interaction between heterogeneous systems, an XML process knowledge ontology is constructed. This ontology standardizes the definition of the semantics, attributes, and association rules of core process elements. The ontology structure model is shown in Equation 16. (16) in, XML process knowledge ontology; It is a set of process parameter elements, including parameters such as cutting speed, feed rate, and depth of cut; It is a set of cutting state elements, including state elements such as cutting force, vibration, and temperature; It is a set of processing quality elements, including quality elements such as roughness and defect rate; It is a set of mapping relationships between elements, describing the numerical and semantic relationships between elements; Define the iterative update logic for ontology elements to synchronize the update rule set.
[0045] The XML process knowledge ontology standardizes the description of process elements, but to achieve synchronous evolution between physical and virtual spaces, it is first necessary to quantify the mapping and matching degree between the physical and virtual spaces, using this as a criterion for synchronous updates. The overall mapping and matching degree is obtained by weighted summation of the similarity of multi-dimensional features: (17) in, The degree of mapping and matching between physical space and virtual space; The total number of dimensions of the matching features; For the first The similarity of the dimensional features is calculated from the physical measured values and the virtual simulation values; For the first The weighting coefficients of the features are determined based on the degree of influence of the features on the processing quality.
[0046] When the virtual-to-real mapping matching degree meets the preset requirements, it indicates that the simulation results of the twin model are credible. At this point, cutting parameters can be optimized based on the twin model and the cutting process prediction model. The preset requirements are set comprehensively according to the machining accuracy requirements and the sensor acquisition accuracy. For example, in the case of vertical CNC milling, the virtual-to-real mapping matching degree can be set to ≥0.90.
[0047] Optimizing cutting parameters requires simultaneously considering multiple objectives such as machining quality and tool life. Therefore, a multi-objective optimization function is constructed, which balances the importance of each sub-objective through weighting coefficients, and seeks the comprehensive optimal solution using the cutting parameter vector as the optimization variable. (18) in, The vector of cutting parameters to be optimized includes cutting speed, feed rate, and depth of cut. To comprehensively optimize the objective function; These are the weighting coefficients for each sub-objective, set according to process requirements. The sum of the values of all weighting coefficients is 1. As an example, under conventional vertical milling machining conditions, these can be set as follows: This achieves a balanced optimization of the machining surface roughness, surface macro defect rate, and remaining tool life. For the predicted surface roughness; The predicted surface macroscopic defect rate; The remaining tool life is predicted. A multi-objective cutting parameter optimization function is constructed, which can comprehensively balance multiple process objectives such as machining quality and tool life, thereby improving the overall process performance of the optimization scheme.
[0048] After obtaining the optimal cutting parameters, the results are fed back to the physical machine tool for machining, and the optimization results and machining effect data are updated to the XML process knowledge ontology to drive the iterative calibration of the twin model, forming a closed-loop optimization mechanism of "perception-modeling-optimization-feedback" to achieve synchronous evolution of virtual and real space and continuous optimization of cutting parameters.
[0049] Existing technologies for virtual-real fusion often remain at the level of one-way mapping at the data level, lacking a unified standard for describing process knowledge. This makes it difficult to achieve knowledge exchange between virtual and real spaces and continuous iteration of twin models, thus failing to form a closed-loop mechanism for optimizing cutting parameters. Therefore, this step constructs a CNC process knowledge ontology based on XML description language to achieve standardized knowledge interaction between virtual and real spaces; establishes a virtual-real synchronous evolution mechanism, constructs a multi-objective optimization model, and achieves closed-loop optimization of cutting parameters.
[0050] This embodiment constructs a CNC process knowledge ontology based on XML description language, standardizes the semantics, attributes and association rules of core process elements, realizes the standardized expression and cross-platform interaction of process knowledge, and solves the problem of inconsistent description of knowledge in virtual and real space. Furthermore, by using the virtual-real space mapping matching degree function, the degree of matching between physical perception data and virtual simulation data can be quantified, ensuring a one-to-one correspondence between virtual and real spaces and providing a basis for judgment on synchronous evolution. A virtual-real bidirectional synchronous evolution mechanism is established, in which the perception data of physical space can drive the iterative update of virtual models and knowledge ontology, and the optimization parameters of virtual space can be fed back to the physical machine tool for execution, forming a closed-loop optimization system of cutting parameters of "perception-modeling-optimization-feedback".
[0051] As an example, taking the optimization of cutting parameters on a vertical CNC milling machine as an application scenario, the specific implementation process is as follows: Step S201: Construct a high-fidelity twin machine tool model.
[0052] First, we analyze the component hierarchy of the vertical CNC milling machine and construct a multi-level component relationship tree according to Formula 1: root node This is a complete vertical milling machine. It comprises five functional components: spindle system, X / Y / Z feed system, bed, worktable, and tool magazine; each functional component is further subdivided into its core parts. For example, a spindle system includes components such as a spindle, bearings, and tool holders; each motion connection unit constitutes... Examples include X-axis prismatic joints and spindle rotary joints; simultaneously, assembly constraints and motion transmission relationships between different levels are defined. .
[0053] Next, the first stage of static calibration was carried out: the key dimensions of the core components were measured using a coordinate measuring machine, and the deviation between the measured dimensions and the initial digital model was calculated. Set calibration coefficients according to part level The calibrated part dimensions are calculated using Formula 2, the digital model is updated, and the static geometric accuracy calibration is completed.
[0054] Then, the second phase of dynamic simulation verification was carried out: the motion chain of the machine tool from the servo motor to the end of the tool was analyzed, and the motion transmission matrix of the whole machine was constructed using Equation 3. By inputting standard motion commands into the twin model, the pose sequence of the end-effector is obtained through simulation. Simultaneously, the measured sequence of the end-effector pose of the physical machine tool is acquired using a machine tool grating ruler and a laser tracker. The dynamic consistency error is calculated using Formula 4. If the error meets the preset threshold, the twin machine tool model construction is completed; if not, the coupling parameters in the motion transfer matrix are corrected until the error meets the standard.
[0055] Step S202: Construct a modular system for three-dimensional process data.
[0056] Construct a static process parameter library according to Formula 5. Enter machine tool parameters such as machine tool model, axis travel, and positioning accuracy; enter tool parameters such as diameter, number of teeth, cutting edge length, and tool material of the end mill; enter workpiece parameters such as material grade, size, and blank condition of the workpiece; enter reference parameters such as reference cutting speed and feed rate corresponding to different materials.
[0057] A dynamic cutting feature set is constructed according to Formula 6: real-time signals of the machining process are collected by cutting force sensors, vibration sensors, temperature sensors, and acoustic emission sensors. The effective value of cutting force, root mean square value of vibration, cutting temperature, and acoustic emission feature value at each moment are extracted to form a dynamic cutting feature vector. And store.
[0058] Construct a comprehensive process knowledge graph according to Formula 7: extract three types of entities: process parameters, cutting state, and machining quality, define semantic relationships such as "influence", "correspondence", and "cause and effect", associate the attribute values of the entities, and form a structured process knowledge network.
[0059] Finally, the comprehensive index value of each process data point is calculated using Formula 8. It enables the association and modular management of static, dynamic, and knowledge-based data, and supports fast retrieval by working condition, tool, workpiece, and other dimensions.
[0060] Step S203: Construct a cutting process prediction model.
[0061] The amount of flank wear is collected using a tool wear detection device. Calculate the remaining life based on the wear rate. The tool wear state vector is constructed according to Formula 9. The roughness of the machined surface was measured using a surface profilometer. With defect rate The surface morphology evaluation vector is constructed according to Formula 10. .
[0062] Based on experimental data from multiple sets of different wear stages, the dynamic mapping matrix was identified using Equation 11. The influence coefficient of quantified wear on roughness and defect rate.
[0063] Subsequently, an improved kernel limit learning model was trained: cutting parameters, tool wear state, and dynamic cutting features were used as input samples, and the machined surface morphology and cutting state were used as output samples; kernel function weight coefficients were then set. To balance local and global kernel performance, the improved hybrid kernel function value is calculated according to formula 12, and the hidden layer kernel output matrix is constructed using formula 13. Set the regularization coefficient. The output weight matrix is solved using Equation 14. Finally, the cutting process prediction model is obtained through Formula 15, which can realize the rapid prediction of machining quality and cutting state under any input.
[0064] Step S204: Simultaneous synchronization of virtual and real cutting parameters and optimization of cutting parameters.
[0065] Construct an XML process knowledge ontology according to Formula 16 The XML tag defines three types of elements: process parameters, cutting status, and machining quality. It defines the attributes and data types of the elements, configures the mapping relationship and synchronization update rules between the elements, and forms a standardized process knowledge description file.
[0066] Real-time acquisition of sensing data from the physical machine tool is used to synchronously drive the twin model simulation, and the virtual-real mapping matching degree is calculated using formula 17. When the matching degree meets the standard, initiate cutting parameter optimization: set weighting coefficients. , , A multi-objective optimization function is constructed according to Formula 18, with cutting speed, feed rate, and depth of cut as optimization variables, and the optimal cutting parameters are obtained by solving the problem through an intelligent optimization algorithm.
[0067] The optimal parameters are fed back to the physical machine tool for machining, and the machining results are updated to the process database and knowledge ontology to drive the iterative calibration of the twin model, thereby achieving synchronous evolution of the virtual and real spaces and continuous optimization of cutting parameters.
[0068] Example 2 In one or more embodiments, a twin machine tool model construction system for cutting parameter optimization is disclosed, specifically including: The twin model construction module is configured to: construct a topological structure model of a multi-level component relationship tree, and perform static digital model comparison calibration and dynamic motion simulation verification on the parts at each level to obtain a twin machine tool model; The data system construction module is configured to: build a static process parameter library, a dynamic cutting feature set, a comprehensive process knowledge graph, and a modular indexing mechanism; The cutting process modeling module is configured to: fuse multi-dimensional perception data of tool wear state and machined surface morphology, construct a dynamic mapping matrix between the two, and train an improved kernel extreme learning machine to obtain a cutting process prediction model; The cutting parameter optimization module is configured to: construct a CNC process knowledge ontology based on XML description language, quantify the credibility of the simulation results of the twin machine tool model, and perform multi-objective cutting parameter optimization based on the twin machine tool model and the cutting process prediction model to obtain the optimal cutting parameters.
[0069] Example 3 This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-described method for constructing a twin machine tool model for cutting parameter optimization.
[0070] Example 4 This embodiment provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described method for constructing a twin machine tool model oriented towards cutting parameter optimization.
[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods and apparatus (systems) according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer instructions. These computer instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0072] These computer instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0073] These computer instructions can also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0074] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for constructing a twin machine tool model for cutting parameter optimization, characterized in that, include: A topological structure model of a multi-level component relationship tree is constructed, and a twin machine tool model is obtained by static digital model comparison calibration and dynamic motion simulation verification of parts at each level. Construct a static process parameter library, a dynamic cutting feature set, a comprehensive process knowledge graph, and a modular indexing mechanism; By integrating multi-dimensional sensing data on tool wear state and machined surface morphology, a dynamic mapping matrix between the two is constructed, and a cutting process prediction model is obtained by training with an improved kernel limit learning machine. A CNC process knowledge ontology based on XML description language is constructed to quantify the credibility of the simulation results of the twin machine tool model. Based on the twin machine tool model and the cutting process prediction model, multi-objective cutting parameters are optimized to obtain the optimal cutting parameters.
2. The twin machine tool model construction method for cutting parameter optimization as described in claim 1, characterized in that, The modular indexing mechanism specifically involves constructing a unified modular indexing function to perform weighted fusion of the classification indexes of the three types of data: the static process parameter library, the dynamic cutting feature set, and the comprehensive process knowledge graph, to obtain a comprehensive index value for each data item, thereby enabling rapid data location and associated retrieval. in, For the first The comprehensive index value of the process data; This is the category index value of the data in the static process parameter library; This is the classification index value of the data in the dynamic cutting feature set; This is the classification index value of the data in the knowledge graph of the integration process; These are the weighting coefficients for the three types of indexes.
3. The twin machine tool model construction method for cutting parameter optimization as described in claim 1, characterized in that, The construction of the dynamic mapping matrix between the two is specifically as follows: The influence of various wear characteristics on various morphological indicators at different times is quantified by a time-varying mapping matrix, while a random error vector is introduced to represent the interference of other unquantified factors: in, for The dynamic mapping matrix at time points; It is a random error vector; for The evaluation vector of the processed surface morphology at any given time; for The tool wear state vector at any given time.
4. The twin machine tool model construction method for cutting parameter optimization as described in claim 1, characterized in that, The cutting process prediction model obtained by training with an improved kernel extreme learning machine is specifically as follows: An improved hybrid kernel function is constructed, which combines the local fitting properties of the radial basis kernel with the global generalization properties of the polynomial kernel, and the proportion of the two is adjusted by weighting coefficients; Based on the improved hybrid kernel function, the training samples and hidden layer nodes are substituted into the kernel function for calculation to construct the hidden layer kernel output matrix; The improved output weights are solved using L2 regularization terms to obtain the cutting process prediction model.
5. The twin machine tool model construction method for cutting parameter optimization as described in claim 4, characterized in that, For any input cutting parameters and state feature vector, substitute them and all training samples into the improved hybrid kernel function for calculation, and then perform operations with the output weight matrix to obtain the corresponding cutting process prediction result: in, For input vectors The corresponding cutting process prediction output vector; For the training sample set; The kernel function is used to compute the value vector of the input vector and all training samples.
6. The twin machine tool model construction method for cutting parameter optimization as described in claim 1, characterized in that, The reliability of the simulation results of the quantitative twin machine tool model is specifically determined by: quantifying the degree of mapping and matching between the physical space and the virtual space through a weighted summation of the similarity of multi-dimensional features. in, The degree of mapping and matching between physical space and virtual space; The total number of dimensions of the matching features; For the first Similarity of dimensional features; For the first Weight coefficients of dimensional features.
7. The twin machine tool model construction method for cutting parameter optimization as described in claim 1, characterized in that, The process involves multi-objective cutting parameter optimization based on the twin machine tool model and the cutting process prediction model to obtain the optimal cutting parameters. Specifically, a multi-objective optimization function is constructed, and the importance of each sub-objective is balanced by weighting coefficients. The cutting parameter vector is used as the optimization variable to seek the comprehensive optimal solution. in, The vector of cutting parameters to be optimized includes cutting speed, feed rate, and depth of cut. To comprehensively optimize the objective function; These are the weighting coefficients for each sub-objective; For the predicted surface roughness; The predicted surface macroscopic defect rate; This represents the predicted remaining tool life.
8. A twin machine tool model construction system for cutting parameter optimization, characterized in that, include: The twin model construction module is configured to: construct a topological structure model of a multi-level component relationship tree, and perform static digital model comparison calibration and dynamic motion simulation verification on the parts at each level to obtain a twin machine tool model; The data system construction module is configured to: build a static process parameter library, a dynamic cutting feature set, a comprehensive process knowledge graph, and a modular indexing mechanism; The cutting process modeling module is configured to: fuse multi-dimensional perception data of tool wear state and machined surface morphology, construct a dynamic mapping matrix between the two, and train an improved kernel extreme learning machine to obtain a cutting process prediction model; The cutting parameter optimization module is configured to: construct a CNC process knowledge ontology based on XML description language, quantify the credibility of the simulation results of the twin machine tool model, and perform multi-objective cutting parameter optimization based on the twin machine tool model and the cutting process prediction model to obtain the optimal cutting parameters.
9. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, complete the twin machine tool model construction method for cutting parameter optimization as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the twin machine tool model construction method for cutting parameter optimization as described in any one of claims 1-7.
