A physical dominant double branch modeling method for milling surface roughness prediction

Abstract

The application belongs to the technical field of intelligent manufacturing, and specifically discloses a physical leading double-branch modeling method for milling surface roughness prediction. According to the application, the cutting speed and the feed per tooth are calculated according to the machining input parameters; in the physical leading branch, the equivalent internal state is estimated, and the milling surface roughness is decomposed and aggregated; in the residual branch, the residual correction amount is calculated; the physical predicted roughness is corrected according to the residual correction amount, and the physical leading branch and the residual branch are trained and updated according to the corrected physical predicted roughness based on the stage training and the few-sample adaptation strategy. Through the above-mentioned mode, the stage training strategy is adopted, the physical leading branch is trained first, and then the residual branch is trained, under unseen working conditions, the physical leading branch is updated first using a small amount of samples, and then the residual branch is fine-tuned, so that the adaptability, interpretability and precision of modeling can be effectively improved under the condition of small sample manufacturing data.

Description

Technical Field

[0001] This application belongs to the field of intelligent manufacturing technology, and more specifically, relates to a physical-dominated bi-branch modeling method for predicting the roughness of milled surfaces. Background Technology

[0002] With the development of intelligent manufacturing and adaptive machining technologies, machining quality prediction models are playing an increasingly important role in process optimization, quality assurance, and process decision-making. Surface roughness is a crucial indicator for evaluating milling quality, affecting not only the surface integrity and service performance of parts but also subsequent machining and inspection costs. Reliable prediction of surface roughness before or during machining can reduce reliance on post-inspection and provide a basis for adjusting process parameters.

[0003] Commonly used surface roughness prediction methods can be broadly categorized into physical model methods and data-driven methods. Physical models typically establish the relationship between machining inputs and surface generation based on geometric relationships, empirical formulas, or mechanistic analysis, offering good interpretability. However, in actual milling processes, surface roughness is influenced by multiple factors, including tool geometry, cutting forces, vibration, thermal effects, tool wear, and material conditions. Single or fixed-form physical models often struggle to balance accuracy and adaptability. Data-driven methods can utilize experimental data to learn the nonlinear mapping from machining parameters to surface roughness, achieving high prediction accuracy when data is abundant and uniformly distributed. However, purely data-driven models often lack physical interpretability, and when working conditions, tool conditions, or the number of available samples changes, the model is prone to degraded extrapolation performance or instability during small-sample fine-tuning. Hybrid modeling methods such as physically-informed neural networks attempt to incorporate physical knowledge as loss constraints or regularization terms into the learning process. However, in many scenarios, the main prediction path is still dominated by the black-box neural network, with physical knowledge serving more as training guidance. This makes it difficult to guarantee that the model structure itself is physically dominant, especially in milling surface roughness prediction tasks. Therefore, the above modeling method has poor adaptability, weak interpretability, and low accuracy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application aims to provide a physically-dominated dual-branch modeling method for predicting milling surface roughness. This method addresses the problems of existing technologies, which suffer from poor adaptability, weak interpretability, and low accuracy due to the coupling effect of multiple factors such as tool geometry, cutting force, vibration, thermal effects, tool wear, and material state on surface roughness. These problems include decreased extrapolation performance, instability in fine-tuning with few samples, difficulty in ensuring the physical dominance of the model structure itself, and the resulting instability.

[0005] To achieve the above objectives, in a first aspect, this application provides a physically-dominated bi-branch modeling method for predicting the roughness of milled surfaces, comprising: Obtain the machining input parameters during the milling process, and calculate the cutting speed and feed per tooth based on the machining input parameters; In the physics-dominated branch, the equivalent internal state is estimated based on the cutting speed, the feed per tooth, and the machining input parameters, and the milling surface roughness is decomposed and aggregated to obtain the physical predicted roughness. In the residual branch, the residual correction is calculated based on the cutting speed, the feed per tooth, the machining input parameters, the equivalent internal state, and the physical predicted roughness. The physical prediction roughness is corrected based on the residual correction amount, and the physical dominant branch and the residual branch are trained and updated based on the corrected physical prediction roughness, based on a staged training and few-sample adaptation strategy, so as to realize the modeling of the physical dominant dual branch.

[0006] In one embodiment, the step of acquiring machining input parameters during the milling process and calculating the cutting speed and feed per tooth based on the machining input parameters includes: Obtain the machining input parameters during the milling process, and extract the spindle speed and feed rate from the machining input parameters; Calculate the cutting speed based on the spindle speed and tool diameter; The feed per tooth is calculated based on the feed rate, the spindle speed, and the number of cutter teeth.

[0007] In one embodiment, the step of estimating the equivalent internal state based on the cutting speed, the feed per tooth, and the machining input parameters in the physical-dominant branch, and decomposing and aggregating the milled surface roughness to obtain the physically predicted roughness, includes: The depth of cut is extracted from the machining input parameters, and the cutting speed, the feed per tooth, and the depth of cut are input into the physical dominant branch; In the physical dominant branch, the equivalent temperature state is estimated based on the cutting speed, the feed per tooth, the depth of cut, and the learnable temperature coefficient. In the physical dominant branch, the equivalent wear evolution response is estimated based on the cutting speed, the equivalent temperature state, the learnable wear correlation coefficient, and the scaling factor of the wear evolution response, and the equivalent wear evolution response is converted into an equivalent wear state using a recursive iterative method. In the physical dominant branch, the equivalent cutting force state is estimated based on the cutting speed, the feed per tooth, the depth of cut, the equivalent wear state, and the learnable cutting force correlation coefficient. The equivalent internal state is obtained based on the equivalent temperature state, the equivalent wear evolution response, the equivalent wear state, and the equivalent cutting force state. The milled surface roughness is then decomposed and aggregated to obtain the physical predicted roughness.

[0008] In one embodiment, the step of decomposing and aggregating the roughness of the milled surface to obtain the physically predicted roughness includes: The equivalent tool tip radius factor is determined based on the nominal tool tip radius and the equivalent wear state, and the geometric contribution is determined based on the equivalent tool tip radius factor and the feed per tooth. The wear contribution is determined based on the learnable wear correlation coefficient, and the vibration contribution is determined based on the equivalent cutting force state and the learnable vibration coefficient. The geometric contribution, the wear contribution, and the vibration contribution are used as the decomposition results of the milled surface roughness. In the physics-dominated branch, the physical predicted roughness is calculated based on the geometric contribution, the wear contribution, the vibration contribution, and the learnable bias term.

[0009] In one embodiment, the step of correcting the physical prediction roughness based on the residual correction amount, and training and updating the physical dominant branch and the residual branch based on the corrected physical prediction roughness according to a staged training and few-shot adaptation strategy, to achieve the modeling of the physical dominant dual branch, includes: The systematic bias of the learning system is determined as the training objective of the residual branch; The residual correction and the physical prediction roughness are added together according to the training objective; Based on a phased training and few-sample adaptation strategy, the physical dominant branch and the residual branch are trained and updated according to the corrected physical prediction roughness to achieve modeling of the physical dominant dual branch.

[0010] In one embodiment, the step of training and updating the physical dominant branch and the residual branch based on the corrected physical prediction roughness according to the staged training and few-shot adaptation strategy to achieve the modeling of the physical dominant dual branch includes: In the first stage, the physical dominant branch is trained based on a staged training strategy, according to the corrected physical prediction roughness. In the second stage, the parameters of the physical dominant branch after training are frozen, and the residual branch is trained based on a staged training strategy. On support samples where no working conditions have been seen, the physical dominant branch after training is updated based on a few-shot adaptation strategy; The trained physical dominant branch is frozen or largely preserved, and the trained residual branch is fine-tuned to a limited extent based on the few-shot adaptation strategy to achieve modeling of the physical dominant dual branch.

[0011] Secondly, this application provides a physically-dominated dual-branch modeling apparatus for predicting the roughness of milled surfaces, comprising: The calculation module is used to obtain the machining input parameters during the milling process, and calculate the cutting speed and feed per tooth based on the machining input parameters. The estimation module is used to estimate the equivalent internal state in the physical dominant branch based on the cutting speed, the feed per tooth, and the machining input parameters, and to decompose and aggregate the milled surface roughness to obtain the physical predicted roughness. The calculation module is also used to calculate the residual correction amount in the residual branch based on the cutting speed, the feed per tooth, the machining input parameters, the equivalent internal state, and the physical predicted roughness; The modeling module is used to correct the physical prediction roughness according to the residual correction amount, and to train and update the physical dominant branch and the residual branch according to the corrected physical prediction roughness based on the staged training and few-sample adaptation strategy, so as to realize the modeling of the physical dominant dual branch.

[0012] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0013] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0014] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0015] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0016] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: (1) After calculating the cutting speed and feed per tooth according to the machining input parameters, this application introduces the derived parameters related to the machining mechanism into the physical dominant branch, so that the subsequent equivalent state estimation is consistent with the empirical relationship. In order to make the surface roughness prediction process have a clear equivalent state modeling, wear iteration accumulation and roughness decomposition path, this application will also directly embed the structured physical dominant branch into the prediction architecture, estimate the equivalent internal state and perform roughness decomposition in the physical dominant branch, and then calculate the residual correction amount in the residual branch, that is, compensate the systematic deviation of the physical model through the light residual branch, taking into account both physical interpretability and prediction accuracy. Introducing the staged training and few-sample adaptation strategy, the physical dominant branch is trained first and then the residual branch is trained. In the absence of working conditions, the physical dominant branch is updated first with a small number of samples, and then the residual branch is fine-tuned to improve the extrapolation ability and adaptability of the model under new working conditions. Thus, it can effectively improve the adaptability, interpretability and accuracy of modeling under the condition of small sample manufacturing data, and is suitable for intelligent manufacturing quality prediction scenarios with limited machining data.

[0017] (2) To maintain the stability and interpretability of the physical dominant branch, this application also sets preset upper and lower bounds for the learnable coefficients in the physical dominant branch. These preset upper and lower bounds can be determined based on the classical geometric roughness relationship, the expected relative magnitudes of wear contribution and vibration contribution, and the roughness range in the training data. Before each forward calculation and backward training feedback, the corresponding coefficients are restricted to the preset range, thereby reducing the risk of parameter combinations with extreme numerical values ​​but weak physical meaning under limited data conditions.

[0018] In summary, this application obtains the machining input parameters during the milling process and calculates the cutting speed and feed per tooth based on these parameters. In the physical dominant branch, the equivalent internal state is estimated based on the cutting speed, feed per tooth, and machining input parameters, and the milled surface roughness is decomposed and aggregated to obtain the physical predicted roughness. In the residual branch, the residual correction amount is calculated based on the cutting speed, feed per tooth, machining input parameters, equivalent internal state, and physical predicted roughness. The physical predicted roughness is corrected based on the residual correction amount, and the physical dominant branch and the residual branch are trained and updated based on the corrected physical predicted roughness using a staged training and few-sample adaptation strategy to achieve physical dominant dual-branch modeling. Through this method, a staged training strategy is adopted, training the physical dominant branch first and then the residual branch. In the absence of observed working conditions, the physical dominant branch is updated first using a small number of samples, and then the residual branch is fine-tuned. This effectively improves the adaptability, interpretability, and accuracy of the modeling under conditions of small-sample manufacturing data. Attached Figure Description

[0019] Figure 1 This is one of the flowcharts of the physically-dominated bi-branch modeling method for predicting the roughness of milled surfaces provided in the embodiments of this application; Figure 2 This is a schematic diagram of the staged training and few-shot adaptation process provided in the embodiments of this application; Figure 3 This is the second flowchart of the physical-dominated dual-branch modeling method for predicting milled surface roughness provided in the embodiments of this application; Figure 4 This is a schematic diagram of the module structure of the physical-dominated dual-branch modeling device for predicting milled surface roughness provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] In this article, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The symbol " / " in this article indicates that the related objects are in an "or" relationship; for example, A / B means A or B.

[0022] The terms "first" and "second," etc., used in the specification and claims herein are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0023] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0024] Based on this, embodiments of this application provide a physically-dominated dual-branch modeling method for predicting the roughness of milled surfaces, referring to... Figure 1 , Figure 1This is one of the flowcharts illustrating a physically-dominated bi-branch modeling method for predicting milled surface roughness provided in this application. In this embodiment, the physically-dominated bi-branch modeling method for predicting milled surface roughness includes steps S10 to S40: Step S10: Obtain the machining input parameters during the milling process, and calculate the cutting speed and feed per tooth based on the machining input parameters.

[0025] It should be noted that the machining input parameters in the milling process include, but are not limited to, feed rate, spindle speed, and depth of cut. These machining input parameters can be used as basic inputs, or derived parameters related to the machining mechanism, to calculate the cutting speed and feed per tooth respectively.

[0026] Further, step S10 includes: acquiring machining input parameters during the milling process; extracting the spindle speed and feed rate from the machining input parameters; calculating the cutting speed based on the spindle speed and tool diameter; and calculating the feed per tooth based on the feed rate, the spindle speed, and the number of cutting teeth.

[0027] It should be understood that after extracting the spindle speed and feed rate from the machining input parameters during the milling process, the cutting speed can be calculated based on the spindle speed and tool diameter, specifically: V = πD×Ns / 1000.

[0028] Where V represents the cutting speed, D represents the tool diameter, and Ns represents the spindle speed.

[0029] On the other hand, the feed per tooth can also be calculated based on the feed rate, spindle speed, and number of teeth, specifically: fz = vf / (Ns×Z).

[0030] Where fz represents the feed per tooth, vf represents the feed rate, Ns represents the spindle speed, and Z represents the number of cutter teeth.

[0031] Step S20: In the physical dominant branch, the equivalent internal state is estimated based on the cutting speed, the feed per tooth, and the machining input parameters, and the milling surface roughness is decomposed and aggregated to obtain the physical predicted roughness.

[0032] Understandably, unlike directly fitting roughness with the original input, this embodiment introduces derived parameters (cutting speed, feed per tooth, etc.) related to the machining mechanism into the physical dominant branch, so that the subsequent equivalent state estimation is consistent with the empirical relationship. That is, the equivalent internal state is estimated based on the cutting speed, feed per tooth, and machining input parameters. The equivalent internal state includes the equivalent temperature state, the equivalent wear evolution response, the equivalent wear state, and the equivalent cutting force state.

[0033] It should be understood that, for milled surface roughness, this embodiment will use the physical dominant branch to decompose it into geometric contribution, wear contribution and vibration contribution, and finally adopt an aggregation method to obtain the physical predicted roughness based on the geometric contribution, wear contribution, vibration contribution and learnable bias term.

[0034] Step S30: In the residual branch, the residual correction amount is calculated based on the cutting speed, the feed per tooth, the machining input parameters, the equivalent internal state, and the physical predicted roughness.

[0035] It should be understood that in this embodiment, the residual branch is used to learn the residual deviation between the physical predicted roughness and the actual measured roughness. The residual branch adopts a lightweight multilayer perceptron, whose inputs are cutting speed, feed per tooth, machining input parameters, equivalent internal state and physical predicted roughness, and the input is the residual correction amount, which is used to correct the physical predicted roughness.

[0036] Step S40: Correct the physical prediction roughness according to the residual correction amount, and based on the staged training and few-sample adaptation strategy, train and update the physical dominant branch and the residual branch according to the corrected physical prediction roughness to realize the modeling of the physical dominant dual branch.

[0037] It should be noted that after correcting the physical predicted roughness based on the residual correction amount, a staged training and few-shot adaptation strategy needs to be introduced. That is, the physical dominant branch and the residual branch are trained separately based on staged training, and the physical dominant branch and the residual branch are updated based on the few-shot adaptation strategy, so as to realize the modeling of the physical dominant dual branch, that is, to achieve efficient, stable and interpretable prediction of milled surface roughness under the condition of small sample manufacturing data.

[0038] Further, step S40 includes: determining the systematic bias of the learning system as the training objective of the residual branch; adding the residual correction amount and the physical prediction roughness according to the training objective; and training and updating the physical dominant branch and the residual branch according to the corrected physical prediction roughness based on a staged training and few-shot adaptation strategy, so as to realize the modeling of the physical dominant dual branch.

[0039] It should be understood that the residual branch can be a lightweight multilayer perceptron, whose training objective can be to learn the systematic deviation between the actual surface roughness and the physically predicted roughness. In this case, the residual correction and the physically predicted roughness can be added together according to the training objective, specifically: Ra_final = Ra_phys + ΔRa.

[0040] Where Ra_final represents the corrected physical prediction roughness, Ra_phys represents the original physical prediction roughness, and ΔRa represents the residual correction amount.

[0041] It should be noted that, through the above structure, the residual branch does not directly replace the physical dominant branch for complete mapping learning, but rather performs limited compensation based on the physical prediction, thereby maintaining the physical dominance of the model.

[0042] Furthermore, the step of training and updating the physical dominant branch and the residual branch based on the corrected physical prediction roughness according to the staged training and few-shot adaptation strategy to achieve the modeling of the physical dominant dual branch includes: in the first stage, training the physical dominant branch according to the corrected physical prediction roughness based on the staged training strategy; in the second stage, freezing the parameters of the trained physical dominant branch and training the residual branch based on the staged training strategy; updating the trained physical dominant branch based on the few-shot adaptation strategy on support samples of unseen working conditions; freezing or basically maintaining the trained physical dominant branch and making limited fine-tuning of the trained residual branch based on the few-shot adaptation strategy to achieve the modeling of the physical dominant dual branch.

[0043] It should be noted that the reference Figure 2 , Figure 2 This diagram illustrates the phased training and few-shot adaptation process. Specifically, for standard training, it is divided into two stages. In the first stage, only the physical dominant branch is trained, and the physical predicted roughness is optimized using the surface roughness labels of the training samples. This allows the branch to capture the dominant changes in surface roughness through a structured physical path, prioritizing the learning of the dominant processing mechanism relationships. In the second stage, the parameters of the trained physical dominant branch are frozen, and the residual branch is trained using the same or corresponding training samples. The residual branch learns the difference between the true surface roughness and the pre-corrected physical predicted roughness Ra_phys. For few-shot adaptation updates, when the model migrates to unseen conditions, a few-shot adaptation strategy is introduced. On the support samples of unseen conditions, the trained physical dominant branch is updated based on the few-shot adaptation strategy, enabling the equivalent internal state estimation and physical prediction path to adapt to the new conditions. Subsequently, the trained physical dominant branch is frozen or largely maintained, and the trained residual branch is fine-tuned to compensate only for the remaining bias under the new conditions. This adaptation sequence reduces overfitting caused by directly fine-tuning parameters driven by a large amount of data.

[0044] It is understood that this embodiment can also utilize milling experimental data to train and verify the above-mentioned technical solution. The milling samples include surface roughness measurements obtained under different spindle speeds, feed rates, and depths of cut. During training, standard working condition samples are used for training the physical dominant branch and the residual branch, while a small number of samples from unseen or boundary conditions are used as support samples for few-sample adaptation. Since the physical dominant branch directly uses original and derived physical quantities with machining significance, the model output includes not only the final roughness prediction value but also intermediate responses such as equivalent temperature state, equivalent wear state, and equivalent cutting force state, which can be used to assist in analyzing the influence of different process parameters on roughness.

[0045] This embodiment acquires the machining input parameters during the milling process and calculates the cutting speed and feed per tooth based on these parameters. In the physical dominant branch, the equivalent internal state is estimated based on the cutting speed, feed per tooth, and machining input parameters, and the milled surface roughness is decomposed and aggregated to obtain the physical predicted roughness. In the residual branch, the residual correction amount is calculated based on the cutting speed, feed per tooth, machining input parameters, equivalent internal state, and physical predicted roughness. The physical predicted roughness is corrected based on the residual correction amount, and the physical dominant branch and the residual branch are trained and updated based on the corrected physical predicted roughness using a staged training and few-sample adaptation strategy to achieve physical dominant dual-branch modeling. Through this method, a staged training strategy is adopted, training the physical dominant branch first and then the residual branch. In the absence of observed working conditions, the physical dominant branch is updated first using a small number of samples, and then the residual branch is fine-tuned. This effectively improves the adaptability, interpretability, and accuracy of the modeling under conditions of small-sample manufacturing data.

[0046] In one specific embodiment, this application provides steps for estimating the equivalent internal state and determining the physical predicted roughness. Please refer to... Figure 3 , Figure 3 This is the second flowchart illustrating the physically-dominated dual-branch modeling method for predicting milled surface roughness provided in this application embodiment. Step S20 includes steps S201 to S205: Step S201: Extract the cutting depth from the machining input parameters, and input the cutting speed, the feed per tooth, and the cutting depth into the physical dominant branch.

[0047] It should be noted that, unlike directly fitting the roughness with the original input, after extracting the cutting depth from the machining input parameters, this embodiment introduces the derived parameters (cutting speed, feed per tooth, etc.) related to the machining mechanism into the physical dominant branch, so that the subsequent equivalent state estimation is consistent with the empirical relationship.

[0048] Step S202: In the physical dominant branch, the equivalent temperature state is estimated based on the cutting speed, the feed per tooth, the depth of cut, and the learnable temperature coefficient.

[0049] Understandably, the equivalent temperature state is used to reflect the influence of milling conditions on the thermally related internal response, and is not required to be equivalent to the temperature directly measured by a sensor. For the physics-driven branch, after obtaining the cutting speed, feed per tooth, and depth of cut, the equivalent temperature state is estimated as follows: T_eq = T_base + C_temp×V^α_temp×fz^β_temp×ap^γ_temp.

[0050] Where T_eq represents the equivalent temperature state, T_base represents the baseline term, C_temp, α_temp, β_temp and γ_temp all represent learnable temperature coefficients, V represents the cutting speed, fz represents the feed per tooth, and ap represents the depth of cut.

[0051] Step S203: In the physical dominant branch, the equivalent wear evolution response is estimated based on the cutting speed, the equivalent temperature state, the learnable wear correlation coefficient, and the scaling factor of the wear evolution response, and the equivalent wear evolution response is converted into an equivalent wear state using a recursive iteration method.

[0052] It should be understood that in the physics-dominated branch, in addition to estimating the equivalent temperature state mentioned above, it is also necessary to estimate the equivalent wear evolution response based on the cutting speed, equivalent temperature state, and learnable wear correlation coefficient, specifically: .

[0053] Where dVB_eq / dt represents the equivalent wear evolution response, C_taylor represents the learnable wear correlation coefficient, and v_scale, t_scale, and wear_scale represent the scaling factors for cutting speed, equivalent temperature state, and wear evolution response, respectively.

[0054] It should be noted that after estimating the equivalent wear evolution response, a recursive iterative approach is used to convert the equivalent wear evolution response into an equivalent wear state. Specifically, the initial equivalent wear state is set to a preset value. In each iteration step, the current equivalent wear evolution response is updated using the equivalent wear evolution response from the previous iteration step and the target learnable scaling factor. The equivalent wear state is also updated cumulatively according to the time step. .

[0055] Where dVB_i / dt represents the equivalent wear state, output_scale represents the target learnable scaling factor, Δt represents the time step, and n represents the number of iterations. After the iteration is complete, VB_eq = VB_n. In this way, the physical dominant branch can accumulate the influence of processing conditions on the wear trend into an equivalent wear state that can be used for roughness prediction.

[0056] Step S204: In the physical dominant branch, the equivalent cutting force state is estimated based on the cutting speed, the feed per tooth, the depth of cut, the equivalent wear state, and the learnable cutting force correlation coefficient.

[0057] Understandably, in the physics-dominated branch, in addition to estimating the equivalent temperature state and equivalent wear state mentioned above, it is also necessary to estimate the equivalent cutting force state based on the cutting speed, feed per tooth, depth of cut, equivalent wear state, and learnable cutting force correlation coefficient. Specifically: F_eq = k_force×ap^α_ap×fz^α_fz×V^α_V×(1+0.1VB_eq).

[0058] Where F_eq represents the equivalent cutting force state, k_force, α_ap, α_fz and α_V all represent learnable cutting force correlation coefficients, and the term (1+0.1VB_eq) is used to reflect the influence of the equivalent wear state on the internal response related to the cutting force.

[0059] Step S205: Based on the equivalent temperature state, the equivalent wear evolution response, the equivalent wear state, and the equivalent cutting force state, the equivalent internal state is obtained, and the milling surface roughness is decomposed and aggregated to obtain the physical predicted roughness.

[0060] Further, the step of decomposing and aggregating the milled surface roughness to obtain the physically predicted roughness includes: determining the equivalent tool tip radius factor based on the nominal tool tip radius and the equivalent wear state, and determining the geometric contribution based on the equivalent tool tip radius factor and the feed per tooth; determining the wear contribution based on the learnable wear correlation coefficient, and determining the vibration contribution based on the equivalent cutting force state and the learnable vibration coefficient, and using the geometric contribution, the wear contribution, and the vibration contribution as the decomposition result of the milled surface roughness; in the physically dominant branch, calculating the physically predicted roughness based on the geometric contribution, the wear contribution, the vibration contribution, and the learnable bias term.

[0061] It should be noted that, for milling surface roughness, this embodiment utilizes the dominant physical branch to decompose it into geometric contribution, wear contribution, and vibration contribution. On one hand, the equivalent tool tip radius factor can be determined based on the nominal tool tip radius and the equivalent wear state, and the geometric contribution can be determined based on the equivalent tool tip radius factor and the feed per tooth, specifically: Ra_geo = k_geo×fz^α_geo / (32r_eff)×(1+β_geo×fz^2 / r_eff)×1000 r_eff = r0 / (1+c_vb,geo×VB_eq).

[0062] Where Ra_geo represents the geometric contribution, r_eff represents the equivalent tool tip radius factor, r0 represents the nominal tool tip radius, c_vb and geo are both learnable geometric coefficients, and VB_eq represents the equivalent wear state. By setting the equivalent tool tip radius factor r_eff, the influence of the equivalent wear state on the geometric roughness contribution can be incorporated into the prediction path.

[0063] On the other hand, the wear contribution can also be determined based on the learnable wear correlation coefficient, specifically: Ra_wear = k_wear×VB_eq.

[0064] Where Ra_wear represents the wear contribution, VB_eq represents the equivalent wear state, and k_wear represents the learnable wear correlation coefficient.

[0065] On the other hand, the vibration contribution can also be determined based on the equivalent cutting force state and the learnable vibration coefficient, specifically: Ra_vib = k_vib×(sqrt(F_eq / 1000)+γ_vib×F_eq / 1000).

[0066] Where Ra_vib represents the vibration contribution, k_vib and γ_vib both represent learnable vibration coefficients, and F_eq represents the equivalent cutting force state.

[0067] It should be noted that after obtaining the geometric contribution, wear contribution, and wear contribution as decomposition results of the milled surface roughness, the physically predicted roughness can be calculated in the physics-dominant branch based on the geometric contribution, wear contribution, vibration contribution, and learnable bias term, specifically: Ra_phys = Ra_geo + Ra_wear + Ra_vib + bias.

[0068] Where Ra_phys represents the physical predicted roughness, Ra_geo represents the geometric contribution, Ra_wear represents the wear contribution, Ra_vib represents the vibration contribution, and bias represents the learnable bias term.

[0069] It should be emphasized that, as can be seen from the above method, the physical dominant branch in this embodiment can express three types of contributions—geometric, wear, and vibration—within the same structure and output the predicted physical roughness Ra_phys. Furthermore, to maintain the stability and interpretability of the physical dominant branch, this application also sets preset upper and lower bounds for the learnable coefficients in the physical dominant branch. These preset upper and lower bounds can be determined based on classical geometric roughness relationships, the expected relative magnitudes of wear and vibration contributions, and the roughness range in the training data. Before each forward calculation and backward training feedback, the corresponding coefficients are limited to the preset range, thereby reducing the risk of parameter combinations with extreme values ​​but weak physical meaning under limited data conditions.

[0070] This embodiment extracts the cutting depth from the machining input parameters and inputs the cutting speed, feed per tooth, and cutting depth into the physical dominant branch. In the physical dominant branch, the equivalent temperature state is estimated based on the cutting speed, feed per tooth, cutting depth, and a learnable temperature coefficient. In the physical dominant branch, the equivalent wear evolution response is estimated based on the cutting speed, the equivalent temperature state, a learnable wear correlation coefficient, and a scaling factor for the wear evolution response, and the equivalent wear evolution response is converted into an equivalent wear state using a recursive iteration method. In the physical dominant branch, the equivalent cutting force state is estimated based on the cutting speed, feed per tooth, cutting depth, equivalent wear state, and a learnable cutting force correlation coefficient. The equivalent internal state is obtained based on the equivalent temperature state, the equivalent wear evolution response, the equivalent wear state, and the equivalent cutting force state, and the milled surface roughness is decomposed and aggregated to obtain the physically predicted roughness. By introducing derived parameters related to the machining mechanism into the physical dominant branch in the above manner, the subsequent equivalent state estimation is consistent with the empirical relationship. In the physical dominant branch, the equivalent temperature state, equivalent wear evolution response, equivalent wear state, and equivalent cutting force state are estimated respectively. This can express the three types of contributions of geometry, wear, and vibration in the same structure. Furthermore, the physical predicted roughness is calculated by decomposition and aggregation, thereby improving the accuracy of the physical predicted roughness calculation.

[0071] The following describes the physically-dominated bi-branch modeling apparatus for predicting milled surface roughness provided in this application. The physically-dominated bi-branch modeling apparatus for predicting milled surface roughness described below corresponds to the physically-dominated bi-branch modeling method for predicting milled surface roughness described above. Please refer to... Figure 4 , Figure 4 This is a schematic diagram of the module structure of the physical-dominated dual-branch modeling device for predicting milled surface roughness provided in this application embodiment, including: The calculation module T10 is used to obtain the machining input parameters during the milling process, and to calculate the cutting speed and feed per tooth based on the machining input parameters.

[0072] The estimation module T20 is used to estimate the equivalent internal state in the physical dominant branch based on the cutting speed, the feed per tooth, and the machining input parameters, and to decompose and aggregate the milled surface roughness to obtain the physical predicted roughness.

[0073] The calculation module T10 is also used to calculate the residual correction amount in the residual branch based on the cutting speed, the feed per tooth, the machining input parameters, the equivalent internal state, and the physical predicted roughness.

[0074] The modeling module T30 is used to correct the physical prediction roughness according to the residual correction amount, and to train and update the physical dominant branch and the residual branch according to the corrected physical prediction roughness based on the staged training and few-sample adaptation strategy, so as to realize the modeling of the physical dominant dual branch.

[0075] This embodiment acquires the machining input parameters during the milling process and calculates the cutting speed and feed per tooth based on these parameters. In the physical dominant branch, the equivalent internal state is estimated based on the cutting speed, feed per tooth, and machining input parameters, and the milled surface roughness is decomposed and aggregated to obtain the physical predicted roughness. In the residual branch, the residual correction amount is calculated based on the cutting speed, feed per tooth, machining input parameters, equivalent internal state, and physical predicted roughness. The physical predicted roughness is corrected based on the residual correction amount, and the physical dominant branch and the residual branch are trained and updated based on the corrected physical predicted roughness using a staged training and few-sample adaptation strategy to achieve physical dominant dual-branch modeling. Through this method, a staged training strategy is adopted, training the physical dominant branch first and then the residual branch. In the absence of observed working conditions, the physical dominant branch is updated first using a small number of samples, and then the residual branch is fine-tuned. This effectively improves the adaptability, interpretability, and accuracy of the modeling under conditions of small-sample manufacturing data.

[0076] It is understood that the detailed functional implementation of each of the above modules can be found in the description of the aforementioned method embodiments, and will not be repeated here.

[0077] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0078] Based on the methods in the above embodiments, this application provides an electronic device, please refer to... Figure 5 , Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application.

[0079] It should be noted that the system may include: a processor 10, a communications interface 20, a memory 30, and a communication bus 40. The processor 10, communications interface 20, and memory 30 communicate with each other via the communication bus 40. The processor 10 can invoke logical instructions stored in the memory 30 to execute the methods described in the above embodiments.

[0080] Furthermore, the logical instructions in the aforementioned memory 30 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0081] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0082] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0083] It is understood that the processor in the embodiments of this application can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0084] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, portable hard disks, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor.

[0085] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. Those skilled in the art will readily understand that the above descriptions are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A physically-dominated bi-branch modeling method for predicting the surface roughness of milled surfaces, characterized in that, include: Obtain the machining input parameters during the milling process, and calculate the cutting speed and feed per tooth based on the machining input parameters; In the physics-dominated branch, the equivalent internal state is estimated based on the cutting speed, the feed per tooth, and the machining input parameters, and the milling surface roughness is decomposed and aggregated to obtain the physical predicted roughness. In the residual branch, the residual correction is calculated based on the cutting speed, the feed per tooth, the machining input parameters, the equivalent internal state, and the physical predicted roughness. The physical prediction roughness is corrected based on the residual correction amount, and the physical dominant branch and the residual branch are trained and updated based on the corrected physical prediction roughness, based on a staged training and few-sample adaptation strategy, so as to realize the modeling of the physical dominant dual branch.

2. The physics-dominated bi-branch modeling method as described in claim 1, characterized in that, The steps of acquiring machining input parameters during the milling process and calculating the cutting speed and feed per tooth based on the machining input parameters include: Obtain the machining input parameters during the milling process, and extract the spindle speed and feed rate from the machining input parameters; Calculate the cutting speed based on the spindle speed and tool diameter; The feed per tooth is calculated based on the feed rate, the spindle speed, and the number of cutter teeth.

3. The physics-dominated bi-branch modeling method as described in claim 1, characterized in that, In the physically-dominated branch, the step of estimating the equivalent internal state based on the cutting speed, the feed per tooth, and the machining input parameters, and decomposing and aggregating the milling surface roughness to obtain the physically predicted roughness, includes: The depth of cut is extracted from the machining input parameters, and the cutting speed, the feed per tooth, and the depth of cut are input into the physical dominant branch; In the physical dominant branch, the equivalent temperature state is estimated based on the cutting speed, the feed per tooth, the depth of cut, and the learnable temperature coefficient. In the physical dominant branch, the equivalent wear evolution response is estimated based on the cutting speed, the equivalent temperature state, the learnable wear correlation coefficient, and the scaling factor of the wear evolution response, and the equivalent wear evolution response is converted into an equivalent wear state using a recursive iterative method. In the physical dominant branch, the equivalent cutting force state is estimated based on the cutting speed, the feed per tooth, the depth of cut, the equivalent wear state, and the learnable cutting force correlation coefficient. The equivalent internal state is obtained based on the equivalent temperature state, the equivalent wear evolution response, the equivalent wear state, and the equivalent cutting force state. The milled surface roughness is then decomposed and aggregated to obtain the physical predicted roughness.

4. The physics-dominated bi-branch modeling method as described in claim 3, characterized in that, The step of decomposing and aggregating the surface roughness of the milled surface to obtain the physically predicted roughness includes: The equivalent tool tip radius factor is determined based on the nominal tool tip radius and the equivalent wear state, and the geometric contribution is determined based on the equivalent tool tip radius factor and the feed per tooth. The wear contribution is determined based on the learnable wear correlation coefficient, and the vibration contribution is determined based on the equivalent cutting force state and the learnable vibration coefficient. The geometric contribution, the wear contribution, and the vibration contribution are used as the decomposition results of the milled surface roughness. In the physics-dominated branch, the physical predicted roughness is calculated based on the geometric contribution, the wear contribution, the vibration contribution, and the learnable bias term.

5. The physics-dominated bi-branch modeling method as described in any one of claims 1 to 4, characterized in that, The steps of correcting the physical prediction roughness based on the residual correction amount, and training and updating the physical dominant branch and the residual branch based on the corrected physical prediction roughness according to a staged training and few-shot adaptation strategy, to achieve the modeling of the physical dominant dual branch, include: The systematic bias of the learning system is determined as the training objective of the residual branch; The residual correction and the physical prediction roughness are added together according to the training objective; Based on a phased training and few-sample adaptation strategy, the physical dominant branch and the residual branch are trained and updated according to the corrected physical prediction roughness to achieve modeling of the physical dominant dual branch.

6. The physics-dominated bi-branch modeling method as described in claim 5, characterized in that, The steps of training and updating the physical dominant branch and the residual branch based on the corrected physical prediction roughness, according to the staged training and few-shot adaptation strategy, to achieve the modeling of the physical dominant dual branch, include: In the first stage, the physical dominant branch is trained based on a staged training strategy, according to the corrected physical prediction roughness. In the second stage, the parameters of the physical dominant branch after training are frozen, and the residual branch is trained based on a staged training strategy. On support samples where no working conditions have been seen, the physical dominant branch after training is updated based on a few-sample adaptation strategy; The trained physical dominant branch is frozen or largely preserved, and the trained residual branch is fine-tuned to a limited extent based on the few-shot adaptation strategy to achieve modeling of the physical dominant dual branch.

7. A physically-dominated dual-branch modeling apparatus for predicting the roughness of milled surfaces, characterized in that, include: The calculation module is used to obtain the machining input parameters during the milling process, and calculate the cutting speed and feed per tooth based on the machining input parameters. The estimation module is used to estimate the equivalent internal state in the physical dominant branch based on the cutting speed, the feed per tooth, and the machining input parameters, and to decompose and aggregate the milled surface roughness to obtain the physical predicted roughness. The calculation module is also used to calculate the residual correction amount in the residual branch based on the cutting speed, the feed per tooth, the machining input parameters, the equivalent internal state, and the physical predicted roughness; The modeling module is used to correct the physical prediction roughness according to the residual correction amount, and to train and update the physical dominant branch and the residual branch according to the corrected physical prediction roughness based on the staged training and few-sample adaptation strategy, so as to realize the modeling of the physical dominant dual branch.

8. An electronic device, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run on the processor, it causes the processor to perform the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, When the computer program product is run on a processor, the processor causes the processor to perform the method as described in any one of claims 1-6.

Images