Real-time prediction method and system for milling deformation of thin-wall part and electronic equipment

Abstract

The invention provides a real-time prediction method and system for milling deformation of a thin-wall part and electronic equipment. The method comprises the steps that a partial differential equation set used for reflecting the deflection and stress coupling relation in the milling machining process of the thin-walled workpiece is constructed based on a von Karman control equation set; constructing a deformation prediction model based on a depth operator network; according to the partial differential equation set, constructing a loss function comprising a partial differential equation residual term and a boundary condition term; training the deformation prediction model according to the loss function and two-dimensional coordinate points and load function sampling points used for training the deformation prediction model; and deploying the trained model in a machine tool system, inputting real-time load data and real-time coordinate data of the thin-wall part in the milling process, predicting deformation response of the thin-wall part in the milling process, and obtaining a target prediction result. The method can quickly and accurately predict the deformation response of the thin-wall part under the complex load working condition, and improves the machining quality and efficiency of the thin-wall part.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of numerical control machining, in particular to a thin-walled part milling deformation real-time prediction method and system and electronic equipment. BACKGROUND

[0002] In the machining process of thin-walled parts in the fields of aerospace and precision manufacturing, due to the low rigidity of thin plate structure, the cutting load, clamping load and other loads generated in the machining process will often cause large elastic or elastic-plastic deformation, and severe deformation will lead to decreased machining precision or even scrap of the parts. Therefore, in the milling process of thin-walled parts, high-precision and high-efficiency deformation prediction is crucial to ensure the machining quality.

[0003] At present, the technologies for predicting the milling deformation of thin-walled parts mainly include finite element simulation method and data-driven method. Although the finite element simulation has high precision, it is complex to model and has low calculation efficiency, which is difficult to meet the application requirements of real-time prediction and online compensation; the data-driven method relies on a large number of cutting experiments to obtain samples for training the model, but the experimental process is time-consuming and laborious, and the cost is high, which is difficult to cover all possible combinations of machining parameters and has limited generalization ability.

[0004] Therefore, there is an urgent need for a new method that has physical constraint ability, learning ability and generalization ability to quickly predict the deformation behavior of thin-walled parts under complex loads, so as to solve the above-mentioned deficiencies of the prior art in strong data dependency, poor real-time performance and weak adaptability. SUMMARY

[0005] Therefore, the present application provides a thin-walled part milling deformation real-time prediction method and system and electronic equipment, aiming to reduce the dependence on experiments, improve the prediction accuracy, and realize online deformation prediction and compensation with low cost and high efficiency.

[0006] According to a first aspect of an embodiment of the present application, a thin-walled part milling deformation real-time prediction method is provided, comprising: constructing a partial differential equation set for reflecting the coupling relationship between deflection and stress in the thin-walled part milling process based on the von Karman control equation set, the partial differential equation set including the size, material properties, load distribution and geometric boundary of the thin plate; constructing a deformation prediction model based on a depth operator network, the deformation prediction model adopting a double-network structure including a trunk network and a branch network; constructing a loss function including a partial differential equation residual term and a boundary condition term according to the partial differential equation set; training the deformation prediction model according to the loss function and two-dimensional coordinate points and load function sampling points used for training the deformation prediction model; deploying the trained deformation prediction model to a machine tool system and inputting real-time load data and real-time coordinate data generated by the thin-walled part in the milling process to predict the deformation response of the thin-walled part in the milling process and obtain a target prediction result, the target prediction result including deflection function prediction values and stress function prediction values.

[0007] Optionally, the method further comprises pre-adjusting or error compensating a tool path of the milling process according to the target prediction result.

[0008] Optionally, the trunk network encodes the center coordinates, length-width dimensions and load amplitude of the input load function sampling points, the input dimension is fixed as 5, and the output feature dimension is 2×D; the branch network receives the two-dimensional coordinate points as input, the input dimension is fixed as 2, and the output feature dimension is 2×D; the output features of the trunk network and the branch network are fused through Einstein summation rule, and the deflection function prediction values and the stress function prediction values are obtained through point-by-point multiplication and summation.

[0009] Optionally, the two-dimensional coordinate points include internal points and boundary points, the internal points are generated by uniform random distribution sampling within a specified rectangular region, and the boundary points are generated by uniform sampling along the four edges of the specified rectangular region.

[0010] Optionally, the loss function includes a partial differential equation loss and a boundary condition loss, wherein:

[0011] For the partial differential equation loss:

[0012] loss pde1 =D·Δ 2 (w)-q+h·L(w,Φ)

[0013]

[0014] loss pde =loss pde1 2 +losspde2 2

[0015] For the boundary condition loss, it is necessary to distinguish the x and y directions:

[0016] For the boundary of x = 0, a:

[0017]

[0018] For the boundary of y = 0, b:

[0019]

[0020] loss bd =loss bd1 2 +loss bd 2

[0021] Then the loss function is calculated as:

[0022] loss total =w p *loss pde +w b *loss bd

[0023] Wherein, w p is the weight of the partial differential equation loss, w b is the weight of the boundary condition loss.

[0024] Optionally, the training of the deformation prediction model further comprises: setting the training round number, batch size and device; creating a data iterator of internal points and boundary points; setting a learning rate scheduler and binding it with an optimizer.

[0025] According to a second aspect of the embodiment of the present application, a thin-walled part milling deformation real-time prediction system is provided, comprising: an equation set construction module, configured to construct a partial differential equation set for reflecting the coupling relationship between deflection and stress in the thin-walled part milling process based on a von Karman control equation set, wherein the partial differential equation set comprises the size, material properties, load distribution and geometric boundary of the thin plate; a model construction module, configured to construct a deformation prediction model based on a deep operator network, wherein the deformation prediction model adopts a double-network structure comprising a trunk network and a branch network; a loss function construction module, configured to construct a loss function comprising a partial differential equation residual term and a boundary condition term according to the partial differential equation set; a training module, configured to train the deformation prediction model according to the loss function and two-dimensional coordinate points and load function sampling points used for training the deformation prediction model; and a prediction module, configured to deploy the trained deformation prediction model to a machine tool system, input real-time load data and real-time coordinate data generated by the thin-walled part in the milling process, predict the deformation response of the thin-walled part in the milling process, and obtain a target prediction result, wherein the target prediction result comprises a deflection function prediction value and a stress function prediction value.

[0026] According to a third aspect of the embodiment of the present application, an electronic device is provided, comprising a processor and a memory storing a program. The program comprises instructions which, when executed by the processor, cause the processor to perform the steps of the method of the first aspect.

[0027] According to a fourth aspect of the embodiment of the present application, a computer storage medium is provided, which stores a computer program. The program is executed by a processor to implement the method of the first aspect.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] (1) The method of the present application combines physical mechanisms with small sample data, significantly reducing the dependence on a large number of cutting experiments or simulation modeling.

[0030] (2) The present application can quickly predict the deformation response under various load conditions, meeting the real-time requirement.

[0031] (3) The present application supports end-to-end modeling, has strong structure universality, and is suitable for thin-walled parts of different sizes, boundary conditions or load positions.

[0032] (4) The prediction result of the present application can be directly used for error compensation and tool path optimization, improving the processing quality and efficiency of thin-walled parts. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art based on these drawings.

[0034] Figure 1 A flow chart of the steps of the method of the present application.

[0035] Figure 2 A schematic diagram of the equivalent surface load distribution of the tool acting on the plate surface.

[0036] Figure 3 A flow chart of the algorithm of the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present application.

[0038] In order to facilitate understanding, before the detailed description of the specific embodiments of the present application, the prior art of the present application is exemplarily described. At present, there are several research attempts to realize the real-time prediction of the milling deformation of thin-walled parts, which can be mainly divided into the following categories:

[0039] 1. Modeling method based on cutting experiment or finite element analysis:

[0040] Document one: "Bai Wanjing. Prediction theory and method of precision milling deformation of aviation thin-walled parts[D]. Zhejiang University, 2009."

[0041] For high-speed milling of aluminum alloy material 7075-T7451, the overall deformation and local machining deformation of thin-walled parts in milling are studied. A finite element model and analysis process suitable for analyzing the overall machining deformation of thin-walled parts are established. However, although the traditional cutting experiment research method can provide real data, it is time-consuming and laborious, and it is difficult to cover all possible combinations of machining parameters, which is costly. Although finite element simulation can simulate complex situations, the calculation cost is high. For dynamic machining process, it is difficult to predict and adjust in real time.

[0042] 2. Data-driven method based on neural network:

[0043] Document two: Li X., Ouyang J., Jiang T., Yang B. Integration modified wavelet neural networks for solving thin plate bending problem. Applied Mathematical Modelling, 2013, 37(5): 2983-2994. DOI: 10.1016 / j.apm.2012.07.036.

[0044] The paper adopts the traditional feedforward neural network idea of "improved wavelet neural network (MWNN) + particle swarm optimization", directly approximates the unknown field in the thin plate bending control equation, and obtains the displacement / curvature quantity through the integral relationship, so as to solve the classic thin plate bending problem (Kirchhoff type) in the non-grid and non-FEM framework, and verify the accuracy and convergence under various loads and boundary conditions. However, this method is only suitable for global uniform load field, and has limited modeling capability for spatially varying loads, especially for loads with boundary discontinuity or transition zone variation.

[0045] In summary, the existing technology faces the following key problems: on the one hand, the deformation prediction based on the physical model (such as finite element method) needs complex and time-consuming iterative calculation for different load conditions, which cannot meet the real-time prediction requirements; on the other hand, the traditional data-driven model (such as general neural network) lacks sufficient physical constraints, and has insufficient prediction accuracy and weak generalization ability for local non-uniform load (such as milling force) under geometric nonlinear large deflection deformation, resulting in distorted prediction results.

[0046] The present application provides a method for accurately predicting the full-field deformation of thin-walled parts under any milling load distribution under strict time constraints, to support real-time process optimization and improve surface quality and processing efficiency.

[0047] Specifically, referring to Figure 1 The thin-walled part milling deformation real-time prediction method of the present application comprises:

[0048] Step S101, based on the von Karman control equation set, a partial differential equation set for reflecting the coupling relationship between deflection and stress in the milling process of thin-walled parts is constructed, and the partial differential equation set includes the size, material properties, load distribution and geometric boundary of the thin plate;

[0049] Step S102, a deformation prediction model based on deep operator network is constructed, and the deformation prediction model adopts a double-network structure including a main network and a branch network;

[0050] Step S103, constructing a loss function containing a partial differential equation residual term and a boundary condition term according to the partial differential equation set;

[0051] Step S104, training the deformation prediction model according to the loss function and two-dimensional coordinate points and load function sampling points used for training the deformation prediction model;

[0052] Step S105, deploying the trained deformation prediction model to a machine tool system, and inputting real-time load data and real-time coordinate data generated in the milling process of the thin-walled part to predict the deformation response of the thin-walled part in the milling process, and obtaining a target prediction result, the target prediction result including a deflection function prediction value and a stress function prediction value.

[0053] It should be understood that the partial differential equation residual is calculated by automatic differentiation without obtaining an explicit solution.

[0054] It should also be understood that the constructed loss function can contain an experimental data term in addition to the partial differential equation residual term and the boundary condition term, and the experimental data term is not necessarily included.

[0055] Optionally, the method further comprises pre-adjusting or error compensating a tool path of the milling process according to the target prediction result.

[0056] Optionally, the center coordinates, length and width dimensions, and load amplitude of the load function sampling points input by the main network are encoded, the input dimension is fixed to 5, and the output feature dimension is 2xD; the branch network receives the two-dimensional coordinate points as input, the input dimension is fixed to 2, and the output feature dimension is 2xD; the output features of the main network and the branch network are fused by Einstein summation rule, and the deflection function prediction value and the stress function prediction value are obtained by point-by-point multiplication and summation.

[0057] Optionally, the two-dimensional coordinate points include internal points and boundary points, the internal points are generated by uniform random distribution sampling within a specified rectangular region, and the boundary points are generated by uniform sampling along the four sides of the specified rectangular region.

[0058] Optionally, the loss function includes a partial differential equation loss and a boundary condition loss, wherein:

[0059] For the partial differential equation loss:

[0060] loss pde1 =D·Δ 2 (w)-q+h·L(w,Φ)

[0061]

[0062] losspde = loss pde1 2 + loss pde2 2

[0063] For the boundary condition loss, the x and y directions need to be distinguished:

[0064] For the boundary of x = 0, a:

[0065]

[0066] For the boundary of y = 0, b:

[0067]

[0068] loss bd = loss bd1 2 + loss bd 2

[0069] Then the loss function is calculated as:

[0070] loss total = w p * loss pde + w b * loss bd

[0071] where w p is the weight of the partial differential equation loss, and w b is the weight of the boundary condition loss.

[0072] Optionally, the training of the deformation prediction model further includes: setting the number of training rounds, batch size and device; creating a data iterator of internal points and boundary points; setting a learning rate scheduler and binding it with an optimizer.

[0073] In summary, the present application uses a training strategy combining limited cutting experiment data and physical equation constraints to significantly reduce the dependence on large-scale experimental data, realizing real-time deformation prediction and compensation in the thin-walled plate machining process. The present application is a thin-walled plate milling deformation prediction method based on PINN and DeepONet, which realizes the joint solution of deflection and stress functions in combination with von Karman large deflection plate equations. The present application encodes and fuses the machining load parameters and spatial coordinate information respectively through the separated trunk network and branch network structure, effectively improving the generalization prediction ability under different loads and boundary conditions.

[0074] Specifically, the scheme of the present application is further described according to the following examples:

[0075] Referring toFigure 2 、 Figure 3 The present application comprises the following key steps:

[0076] Step 1: The present application first establishes a partial differential equation set for describing the deformation behavior of thin-walled parts in the milling process based on the von Karman equation in the large deflection plate theory as a physical control model. This physical control model is suitable for thin plate structures with moderate to large deflection and can describe the deflection and stress state of the plate simultaneously after being loaded.

[0077] It should be understood that the Physics-Informed Neural Networks (PINN) is a method of embedding physical laws into the neural network training process. The Physics-Informed Neural Networks can embed control equations (such as large deflection equations of plates) in neural networks through automatic differentiation and other means, avoiding a large dependence on data.

[0078] At the same time, Deep Operator Neural Networks (DeepONet) proposes an idea based on operator learning, which directly learns the mapping relationship from the input function to the target function, realizing efficient prediction of thin plate deflection and stress field under arbitrary load space distribution. Compared with traditional PINN, DeepONet has stronger advantages in handling high-dimensional input, complex load distribution and cross-scene generalization.

[0079] The present application combines the physical constraint ability of PINN with the operator generalization ability of DeepONet, which can significantly improve the prediction accuracy and adaptability of the model under complex processing conditions while ensuring physical consistency.

[0080] 1-1: Let the thin plate structure be a rectangular plane in the initial state, with length and width dimensions a and b, and thickness h, where a, v >> h. In the two-dimensional coordinate system, let any point in the thin plate be (x, y), the external force load be q(x, y), and the structure be deflected after being loaded, with the deflection function represented as w(x, y), and the corresponding stress function as F(x, y). According to the large deflection plate theory, the deflection function and the stress function satisfy the following von Karman control equation set:

[0081]

[0082] where the biharmonic operator is:

[0083]

[0084] Linear differential operator:

[0085]

[0086] Φ is the Airy stress function:

[0087]

[0088] q(x, y) represents the equivalent surface load distribution on the plate surface by the tool during processing, such as Figure 2 .

[0089] 1-2: When the four edges of the thin plate are fixed, the boundary conditions are as follows:

[0090]

[0091]

[0092] For the large deflection deformation problem of the thin plate, the partial differential equation group is solved under the corresponding boundary conditions and loads to obtain the unknown functions w(x, y) and Φ(x, y).

[0093] Step 2: Construct a double-output neural network structure based on a deep operator network (DeepONet), that is, construct a deformation prediction model based on a deep operator network, for simultaneously predicting the deflection distribution of the large deflection thin plate in space and the stress function related output under the given two-dimensional coordinate points and load function sampling points.

[0094] Specifically, the network structure of the deformation prediction model includes two parts: a backbone network and a branch network, wherein the backbone network is responsible for encoding five parameters of the load function sampling points (the center coordinates (x c ,y c ) of the load region and the length and width dimensions (w, h) and the load amplitude), and the input dimension is fixed at 5; the branch network receives the two-dimensional coordinate points (x, y) as input, and the input dimension is fixed at 2. The feature dimension output by the backbone network is 2xD, wherein the first D dimensions are used for the deflection branch w, and the last D dimensions are used for the stress branch Φ. The branch network outputs are also 2xD, and the corresponding features of the branch network are output at each coordinate point. Finally, the outputs are fused by Einstein summation rule: the branch outputs and the backbone outputs are multiplied point by point and summed to obtain the final deflection function prediction value w(x, y) and the stress function prediction value Φ(x, y), that is, the response quantity is predicted for each group of load functions corresponding to multiple space points. The model structure can adapt to rectangular load changes of any position and size, while ensuring the universality of the structure, facilitating the parameterization of the nonlinear response under the variable load condition.

[0095] The present application trains the neural network with small sample cutting experiment data or a small amount of simulation data, and simultaneously improves the adaptability to new loads or new structures through weight sharing or transfer learning.

[0096] Step 3: Construct the loss function according to the von Karman control equation set.

[0097] For the loss of partial differential equations:

[0098] loss pde1 = D * Delta 2 (w) - q + h * L(w, Phi)

[0099]

[0100] loss pde = loss pde1 2 + loss pde2 2

[0101] For the loss of boundary conditions, we need to distinguish between x and y directions.

[0102] For the boundary of x = 0, a:

[0103]

[0104] For the boundary of y = 0, b:

[0105]

[0106] loss bd = loss bd1 2 + loss bd 2

[0107] Then the total loss can be calculated as:

[0108] loss total = w p * loss pde + w b * loss bd

[0109] where w p is the weight of the loss of partial differential equations, and w b is the weight of the loss of boundary conditions.

[0110] Step 4: Set a random seed to ensure the reproducibility of the results.

[0111] Step 5: Generate the two-dimensional coordinate points (x, y) required for the deformation prediction model training, which are divided into boundary points and internal points according to the actual physical meaning.

[0112] 5-1: Internal point sampling. In a given rectangular region range (such as a plate size of width width and length length), N_i coordinate points are generated by uniformly random distribution, and the output coordinate shape is a tensor of (N_i, 2), where N_i is the number of internal points, and each row represents a two-dimensional point (x i ,y i ).

[0113] 5-2: Boundary point sampling, according to the four edges x=0, x=width, y=0, y=length, respectively, to ensure that the coordinate points are located on the boundary, and a certain number of points are uniformly sampled on each edge, and a total of N_b boundary points are generated, and finally a tensor of shape (N_b, 2) is output, which is consistent with the format of internal points.

[0114] Step 6: Generate load values on the sampling coordinate points using function mapping method, which corresponds to the coordinate points one by one.

[0115] 6-1: Internal point load sampling, given the center coordinates (x c ,y c ) and the length and width size (w, h) of the rectangular load region, determine the load region range, and for each coordinate point (x i ,y i ), the corresponding load value is calculated using the following formula:

[0116] q(x i ,y i ) = q0·σ(α(x i -x1))·(1-σ(α(x i -x2)))·σ(α(y i -y1))·(1-σ(α(y i -y2)))

[0117] Where σ(·) is the Sigmoid function, α is the parameter controlling the boundary smoothness, and [x1, x2], [y1, y2] are the load region boundaries.

[0118]

[0119] Construct multiple samples by predefining the center position list and size list, and finally generate a load tensor of shape (k, N_i), where each row corresponds to a sample function q (i) (x, y).

[0120] 6-2: Boundary point load assignment, boundary points do not participate in load input, and the load values corresponding to all boundary points are uniformly set to 0, that is, a shape (N_b, 1) is generated to ensure that the load only acts on the internal region of the plate.

[0121] Step 7: Instantiate the DeepONet model, i.e., the deformation prediction model, set the number of training rounds, batch size, and device (central processing unit (CPU) or graphics processing unit (GPU)); create data iterators for internal points and boundary points; set the learning rate scheduler (cosine annealing scheduler (CosineAnnealingLR)) and bind it to the optimizer.

[0122] Step 8: Input the generated coordinate data (x, y) and load function q (i) (x, y) into the deformation prediction model to obtain the predicted deflection function w(x, y) and stress function Φ(x, y), and calculate the derivative terms required in step 5 based on automatic differentiation.

[0123] Step 9: Calculate the partial differential equation loss loss pde and the boundary condition loss loss bd according to step 5, and combine them into the total loss according to the weights:

[0124] loss total = w p *loss pde + w b *loss bd

[0125] Step 10: Perform backpropagation on loss total to calculate the gradient; call the optimizer (AdamW, adaptive moment estimation weight decay optimizer) to update the network parameters; then, gradually reduce the learning rate according to the preset rules.

[0126] Step 11: Repeat steps 8 to 10 until the preset number of training rounds is reached or the convergence criterion is met, and save the model file.

[0127] Step 12: Deploy the trained deformation prediction model to the machine tool system, input real-time load data and real-time coordinate data, and predict the deformation of the thin plate.

[0128] In addition, according to the target prediction results, the tool path is pre-adjusted or error compensated to improve the machining precision.

[0129] In summary, the application adopts a joint loss training strategy of limited experimental data and physical equation residual, greatly reduces the dependence on large-scale simulation or experimental data, improves the generalization ability to new load distribution and different boundary conditions, and realizes real-time deformation prediction and compensation in the thin-walled milling process; The application introduces physical information constraints in the deep neural operator network structure, directly embeds the von Karman large deflection plate equation group into the network model training process, realizes the joint prediction of the thin plate deflection and stress function, avoids the convergence difficulty problem of the existing method in the strong nonlinear coupled system, and more importantly, gives the model a clear physical meaning, realizes the physically interpretable joint prediction of the thin plate deflection and stress function, and is significantly different from the black box characteristics of the traditional neural network; The application realizes stable prediction accuracy under any local load distribution, boundary discontinuity or transition change condition through double network design, the main network encodes the local geometry and amplitude parameters of the processing load, the branch network encodes the spatial coordinates, and the load and spatial information are decoupled in the fusion stage, so that the network model, i.e. the deformation prediction model, can maintain stable prediction accuracy under any local load distribution, boundary discontinuity or transition change condition. The design idea of the application can be applied to other theoretical systems constructed by partial differential equations, i.e. different deep operator neural networks and different loss functions are constructed according to different partial differential equations, so as to obtain different prediction relationships of physical quantities.

[0130] The embodiment of the application also provides a thin-walled part milling deformation real-time prediction system, which comprises:

[0131] An equation group construction module is configured to construct a partial differential equation group for reflecting the coupling relationship between deflection and stress in the thin-walled part milling process based on the von Karman control equation group, wherein the partial differential equation group comprises the size, material properties, load distribution and geometric boundary of the thin plate;

[0132] A model construction module is configured to construct a deformation prediction model based on a deep operator network, wherein the deformation prediction model adopts a double-network structure comprising a main network and a branch network;

[0133] A loss function construction module is configured to construct a loss function comprising a partial differential equation residual term and a boundary condition term according to the partial differential equation group;

[0134] A training module is configured to train the deformation prediction model according to the loss function and two-dimensional coordinate points and load function sampling points used for training the deformation prediction model;

[0135] A prediction module is configured to deploy the trained deformation prediction model to a machine tool system, input real-time load data and real-time coordinate data generated in the milling process of the thin-walled part, predict the deformation response of the thin-walled part in the milling process, and obtain a target prediction result, wherein the target prediction result comprises a deflection function prediction value and a stress function prediction value.

[0136] It should be understood that the system of the embodiment is used to implement the corresponding method in the foregoing plurality of method embodiments, and has the beneficial effects of the corresponding method embodiments.

[0137] As another example, the present application also provides an electronic device, which will now be described as an example of a hardware device that can be applied to aspects of the present application, which is an electronic device that can be used as a server or a client of the present application. The electronic device is intended to represent a wide variety of digital electronic computing devices, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computing devices. The electronic device can also represent a wide variety of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown in the figures, their connections, and their functions, as well as their implementation, are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0138] The electronic device can include a processor, a communications interface, a memory, and a communications bus.

[0139] The processor, the communications interface, and the memory communicate with each other through the communications bus. The communications interface is used to communicate with other electronic devices or servers.

[0140] The processor is used to execute a program, which can specifically execute the related steps in the above method embodiments.

[0141] Specifically, the program can include program code, which includes computer operation instructions.

[0142] The processor can be a processor CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present application. The one or more processors included in the smart device can be the same type of processor, such as one or more CPUs; or can be different types of processors, such as one or more CPUs and one or more ASICs.

[0143] The memory is used to store the program. The memory can include a high-speed RAM memory, and can also include a non-volatile memory, such as at least one disk memory.

[0144] The program, when executed by the processor, is used to make the electronic device execute a thin-walled part milling deformation real-time prediction method of the present application.

[0145] Furthermore, the specific implementation of each step in the program can be found in the corresponding descriptions of the steps and units in the above method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0146] An exemplary embodiment of the present invention also provides a computer storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the methods of the various embodiments of the present invention. The corresponding process descriptions in the foregoing method embodiments can be referred to, and will not be repeated here.

[0147] The methods described above according to embodiments of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be processed by software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0148] Specific embodiments of the invention have now been described. Other embodiments are within the scope of the appended claims. In some cases, the actions described in the claims can be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing can be advantageous.

[0149] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the patent protection scope of the embodiments of the present invention should be defined by the claims.

Claims

1. A method for real-time prediction of milling deformation of thin-walled parts, characterized in that, include: Based on the von Kármán control equations, a set of partial differential equations is constructed to reflect the coupling relationship between deflection and stress during the milling process of thin-walled parts. The set of partial differential equations includes the dimensions, material properties, load distribution, and geometric boundaries of the thin plate. A deformation prediction model based on deep operator networks is constructed, wherein the deformation prediction model adopts a dual-network structure including a backbone network and a branch network; Based on the aforementioned system of partial differential equations, a loss function is constructed that includes the residual terms and boundary condition terms of the partial differential equations. The deformation prediction model is trained based on the loss function and the two-dimensional coordinate points and load function sampling points used to train the deformation prediction model. The trained deformation prediction model is deployed on the machine tool system, and real-time load data and real-time coordinate data generated during the milling process of the thin-walled part are input to predict the deformation response of the thin-walled part during the milling process and obtain the target prediction result, which includes the predicted value of the deflection function and the predicted value of the stress function.

2. The method according to claim 1, characterized in that, The method further includes pre-adjusting or error compensation of the tool path for milling based on the target prediction result.

3. The method according to claim 1, characterized in that, The backbone network encodes the center coordinates, length and width dimensions, and load amplitude of the load function sampling points as input, with a fixed input dimension of 5 and an output feature dimension of 2×D. The branch network receives the two-dimensional coordinate points as input, with a fixed input dimension of 2 and an output feature dimension of 2×D. The output features of the backbone network and the branch network are fused using the Einstein summation rule, and the predicted values ​​of the deflection function and stress function are obtained by summing the points-by-point products.

4. The method according to claim 3, characterized in that, The two-dimensional coordinate points include internal points and boundary points. Within a specified rectangular area, the internal points are generated by uniformly random sampling, and the boundary points are generated by uniformly sampling along the four sides of the specified rectangular area.

5. The method according to claim 1, characterized in that, The loss function includes partial differential equation loss and boundary condition loss, wherein: For partial differential equation loss: loss pde1 =D·Δ 2 (w)-q+h·L(w,Φ) loss pde =loss pde1 2 +loss pde2 2 For boundary condition loss, it is necessary to distinguish between the x and y directions: For x = 0, the boundary of a: For y = 0, the boundary of b: loss bd =loss bd1 2 +loss bd 2 The loss function is then calculated as follows: loss total =w p *loss pde +w b *loss bd Among them, w p The weights w that constitute the loss in the partial differential equation are b These are the weights that determine the loss based on the boundary conditions.

6. The method according to claim 4, characterized in that, Training the deformation prediction model further includes: Set the number of training rounds, batch size, and device; Create a data iterator for internal and boundary points; Configure the learning rate scheduler and bind it to the optimizer.

7. A real-time prediction system for milling deformation of thin-walled parts, characterized in that, include: The equation system construction module is used to construct a set of partial differential equations based on the von Kármán control equations to reflect the coupling relationship between deflection and stress during the milling process of thin-walled parts. The set of partial differential equations includes the dimensions, material properties, load distribution and geometric boundaries of the thin plate. The model building module is used to build a deformation prediction model based on a deep operator network. The deformation prediction model adopts a dual-network structure that includes a backbone network and a branch network. The loss function construction module is used to construct a loss function containing the partial differential equation residuals and boundary condition terms based on the partial differential equation system. The training module is used to train the deformation prediction model based on the loss function and the two-dimensional coordinate points and load function sampling points used to train the deformation prediction model. The prediction module is used to deploy the trained deformation prediction model on the machine tool system and input the real-time load data and real-time coordinate data generated by the thin-walled part during the milling process to predict the deformation response of the thin-walled part during the milling process and obtain the target prediction result, which includes the predicted value of the deflection function and the predicted value of the stress function.

8. An electronic device, characterized in that, include: processor; Memory for stored programs; The program includes instructions that, when executed by the processor, cause the processor to perform the steps of the method as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-6.

Images