Milling cutting force prediction method and system of particle collision type neural network

Abstract

The invention discloses a milling cutting force prediction method and system based on a particle collision type neural network, and the method comprises the steps: taking obtained historical cutting data of a milling process as a training set, and training an inference model based on the particle collision type neural network; the inference model comprises an input layer taking the cutting speed, the feed rate and the cutting depth as inputs, an output layer taking the cutting force and the surface roughness as outputs, and a middle layer for realizing particle collision; the middle layer takes the cutting speed, the feeding amount and the cutting depth as nodes, calculates order degree values of particle collision among the nodes, and fuses the order degree values of different nodes to obtain an output value; and adopting the trained inference model to obtain a cutting force predicted value and corresponding surface roughness according to the to-be-measured cutting data. The problem that an artificial neural network model is difficult to explain is solved, and high-precision cutting force prediction is realized while the number of parameters in an inference model is reduced, so that high-quality and high-efficiency cutting machining is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence reasoning technology, and in particular to a method and system for predicting cutting force in milling processing using a particle collision type neural network. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] High-speed cutting can generate high-frequency vibration cutting forces, which further cause tool system vibrations and exacerbate tool wear. The magnitude of the cutting force also affects surface roughness; generally, the greater the cutting force, the greater the surface roughness. Cutting forces during machining have a significant impact on tool wear and workpiece quality, and are a crucial factor in selecting cutting parameters, necessitating high-precision prediction.

[0004] Particle Swarm Optimization (PSO) is a type of evolutionary algorithm, similar to simulated annealing. It starts with a random solution and iteratively searches for the optimal solution, evaluating the quality of the solution using fitness. It is easy to implement, highly accurate, and converges quickly. Currently, improvements have been made to the PSO algorithm, which has excellent global convergence capabilities and can be used to optimize model parameters in artificial neural network models and intelligent reasoning models.

[0005] Predicting cutting forces during milling is typically done using empirical formulas, support vector regression machines, Gaussian process regression models, or artificial neural network models. While these models offer a certain level of accuracy, they cannot fully explain the transformation patterns of cutting forces under varying milling parameters. Furthermore, existing prediction methods offer limited precision in predicting and optimizing cutting forces, resulting in poor prediction accuracy and program accuracy. This makes it difficult to guarantee workpiece quality and hinders their application in intelligent workshop construction. Summary of the Invention

[0006] In order to solve the above problems, the present invention proposes a method and system for predicting cutting force in milling processing based on a particle collision type neural network. The inference model of the particle collision type neural network is inspired by the particle collision pattern, which solves the problems that are difficult to explain by the artificial neural network model. While reducing the number of parameters in the inference model, it achieves high-precision cutting force prediction, thereby realizing high-quality and efficient cutting processing.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a method for predicting cutting force in milling using a particle collision neural network, comprising:

[0009] The historical cutting data of the milling process is used as a training set to train an inference model based on a particle collision neural network;

[0010] The inference model includes an input layer that takes cutting speed, feed rate, and cutting depth as input, an output layer that takes cutting force and surface roughness as output, and an intermediate layer for implementing particle collisions. The intermediate layer uses cutting speed, feed rate, and cutting depth as nodes, calculates the order value of particle collisions between nodes, and fuses the order values of different nodes to obtain the output value.

[0011] The trained inference model is used to obtain the predicted cutting force and the corresponding surface roughness according to the cutting data to be measured.

[0012] As an optional implementation, the intermediate layer includes a space conversion layer, an order layer, a rule generation layer and a rule amplification layer.

[0013] As an optional implementation, the spatial transformation layer is used to divide each input node of the input layer into at least two transformation nodes, and perform weight exchange transformation between the transformation nodes of different input nodes;

[0014] The order layer is used to generate an order value based on the result of weight exchange transformation of each transformation node using a Gaussian function and after particle collision;

[0015] The rule generation layer is used to generate a rule value based on the order value of each transformation node;

[0016] The rule amplification layer is used to sum the generated rule values after amplification processing to obtain the output result of the output layer.

[0017] As an optional implementation method, the process of weighted exchange transformation is: after multiplying each transformation node corresponding to one of the input nodes by the weight, each weighted transformation node obtained is input into each transformation node corresponding to the remaining input nodes in a certain proportion and accumulated, and the sum of the proportions is 1, thereby updating the transformation nodes corresponding to the remaining input nodes, and at the same time, the transformation node corresponding to the current input node is updated to the difference between the original transformation node and the weighted transformation node, and then each updated transformation node value is input into the order layer; wherein the weight is a set value greater than 0 and less than 1.

[0018] As an optional implementation, the rule generation layer includes: for each input node of the input layer, obtaining the order value generated by each transformation node corresponding to the input node in the spatial transformation layer in the order layer, and multiplying the order value of each transformation node corresponding to the input node with the order value generated by each transformation node corresponding to the remaining input nodes in the spatial transformation layer, thereby generating a rule value; wherein the order value nn is multiplied, and n is the number of input nodes.

[0019] As an optional implementation, the rule generation layer includes: for each input node of the input layer, obtaining the order value generated by each transformation node corresponding to the input node in the spatial transformation layer in the order layer, using the order value of each transformation node corresponding to each input node as a rule value, adding and averaging the order value nn of one transformation node of all input nodes to obtain a rule value, where n is the number of input nodes;

[0020] Alternatively, for each input node of the input layer, the order value generated by each transformation node corresponding to the input node in the spatial transformation layer is obtained, and the order value of each transformation node corresponding to the input node is added to the order value generated by each transformation node corresponding to the remaining input nodes in the spatial transformation layer by nn addition and averaging to generate a rule value.

[0021] In a second aspect, the present invention provides a milling cutting force prediction system based on a particle collision neural network, comprising:

[0022] The training model is configured to use the acquired historical cutting data of the milling process as a training set to train an inference model based on a particle collision neural network;

[0023] The inference model includes an input layer that takes cutting speed, feed rate, and cutting depth as input, an output layer that takes cutting force and surface roughness as output, and an intermediate layer for implementing particle collisions. The intermediate layer uses cutting speed, feed rate, and cutting depth as nodes, calculates the order value of particle collisions between nodes, and fuses the order values of different nodes to obtain the output value.

[0024] The prediction model is configured to use the trained inference model to obtain a predicted cutting force value and a corresponding surface roughness according to the cutting data to be measured.

[0025] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0026] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0027] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which implements the method described in the first aspect when executed by a processor.

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

[0029] The present invention proposes a method and system for predicting cutting force in milling processing using a particle collision neural network. The inference model of the particle collision neural network includes an input layer, a spatial transformation layer, an order layer, a rule generation layer, a rule amplification layer, and an output layer, and is used for data-based reasoning, prediction, optimization, and problem diagnosis. Inspired by the particle collision pattern, the model establishes an inference model based on the order of particle collisions, which makes it possible to reason, predict, and optimize small sample data, and to a certain extent solves the problem that artificial neural network models are difficult to explain. The inference model is then applied to cutting force prediction, and a high-speed cutting training data set and a particle swarm optimization algorithm are used to train and learn the inference model. The model's prediction ability is verified using a high-speed cutting verification data set. The inference model can achieve high-precision cutting force prediction while reducing the number of parameters in the inference model.

[0030] This paper considers the collision mechanisms of some microscopic particles in nature and the construction model of an adaptive neural-fuzzy inference system to establish a neural network inference model based on the particle collision pattern. This model is able to obtain a neural network inference model with a certain degree of interpretability, accurately representing the logical causal relationship between the problem and the solution in the form of data. The particle collision neural network inference model is applied to the prediction of cutting forces in turning and milling processes. It has a certain degree of interpretability and can predict cutting data such as cutting forces with high accuracy, further optimizing cutting parameters and achieving high-quality and efficient cutting processes.

[0031] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0033] Figure 1 Flowchart of the method for predicting cutting force in milling using a particle collision neural network provided in Example 1 of the present invention;

[0034] Figure 2 The structural framework of the particle collision neural network inference model provided in Example 1 of the present invention Figure 1 ;

[0035] Figure 3 The structural framework of the particle collision neural network inference model provided in Example 1 of the present invention Figure 2 . DETAILED DESCRIPTION

[0036] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0037] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0038] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "comprise" and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0039] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0040] In order to improve the interpretability of intelligent models based on artificial neural networks, this paper is inspired by the collision mechanism of microscopic particles and proposes a particle collision type neural network reasoning model. Its purpose is to correctly handle the logical causal relationship between problems and solutions, achieve high prediction performance and high optimization capabilities, and apply it to the cutting force prediction of milling processes.

[0041] Example 1

[0042] This embodiment provides a method for predicting cutting force in milling using a particle collision neural network. Figure 1 As shown, including:

[0043] The historical cutting data of the milling process is used as a training set to train an inference model based on a particle collision neural network;

[0044] The inference model includes an input layer that takes cutting speed, feed rate, and cutting depth as input, an output layer that takes cutting force and surface roughness as output, and an intermediate layer for implementing particle collisions. The intermediate layer uses cutting speed, feed rate, and cutting depth as nodes, calculates the order value of particle collisions between nodes, and fuses the order values of different nodes to obtain the output value.

[0045] The trained inference model is used to obtain the predicted cutting force and the corresponding surface roughness according to the cutting data to be measured.

[0046] In this embodiment, historical cutting data of the milling process is obtained and used as a training set; the historical cutting data includes cutting speed, feed rate, cutting depth, cutting force and corresponding surface roughness, etc. The cutting force and surface roughness are predicted using cutting speed, feed rate and cutting depth as input;

[0047] The historical cutting data is used as the training set to train the constructed inference model based on the particle collision neural network. The trained inference model is then used to obtain the cutting force prediction value and the corresponding surface roughness according to the cutting data to be tested.

[0048] In this embodiment, the inference model based on the particle collision neural network includes an input layer, a space conversion layer, an order layer, a rule generation layer, a rule amplification layer, and an output layer. Each layer contains one or more nodes. The meaning of each node is as follows:

[0049] (1) Input layer: The number of input nodes in the input layer is defined according to the size of the problem dimension and is consistent with the size of the problem dimension. For example, cutting speed, feed rate, and cutting depth are used as inputs. Cutting speed, feed rate, cutting depth, cutting force, and surface roughness all need to be normalized before they can be used for model training.

[0050] (2) Spatial transformation layer: Its nodes are variable nodes. The function of the nodes in the spatial transformation layer is to divide each input node of the input layer into at least two transformation nodes. The number of transformation nodes that each input layer node is divided into the spatial transformation layer is determined by the number of types of information contained in the problem dimension.

[0051] The transformation nodes of different input nodes are exchanged and transformed with weights; that is, each transformation node corresponding to one of the input nodes is multiplied by a weight (the weight is a set value greater than 0 and less than 1, usually a weight of 0-0.3), and each weighted transformation node is input into each transformation node corresponding to the remaining input nodes in a certain proportion and accumulated, thereby updating the transformation nodes corresponding to the remaining input nodes. At the same time, the transformation node corresponding to the current input node is updated to the difference between the original transformation node and the weighted transformation node, and then each updated transformation node value is input into the order layer.

[0052] For example, there are input nodes X, Y, and Z. After input node X is mapped to the spatial transformation layer, transformation nodes A1 and B1 are generated. After input node Y is mapped to the spatial transformation layer, transformation nodes C1 and D1 are generated. After input node Z is mapped to the spatial transformation layer, transformation nodes E1 and F1 are generated. After A1 and B1 are weighted respectively, the weighted A1 is input to C1 and E1 in a certain proportion (the sum of the proportions is 1), and the weighted B1 is input to D1 and F1 in a certain proportion (the sum of the proportions is 1). After accumulation, the updated C1, D1, E1, and F1 are obtained. Finally, the difference between A1 and the weighted A1, the difference between B1 and the weighted B1, and the updated C1, D1, E1, and F1 are input to the order layer.

[0053] (3) Order layer: Each transformation node of the spatial transformation layer is input into the order layer after weight exchange transformation. The function function of the order layer node is a Gaussian function, or other function functions with the type of first rising and then falling. The order value is generated after particle collision.

[0054] Taking the Gaussian function as an example, through the Gaussian function Get the order value of particle collision Its value is between (0, 1); a and σ are adjustable parameters, and their values are determined by the learning algorithm; μ Mi is the order value; x is the input node value.

[0055] (4) Rule generation layer:

[0056] like Figure 2 As shown, for each input node in the input layer, the order value generated by each transformation node corresponding to the input node in the spatial transformation layer is obtained, and the order value of each transformation node corresponding to the input node is multiplied with the order value generated by each transformation node corresponding to the remaining input nodes in the spatial transformation layer, thereby generating a certain number of rules. The order value can be multiplied in pairs, in groups of three, or in groups of n according to the number of dimensions of the problem.

[0057] Therefore, the role of the nodes in the rule generation layer is to multiply the values generated by the order layer to generate rules.

[0058]

[0059] Where w i The rule strength of the rule generation layer; is the order value of the Zth input under the i-th rule; is the order value of the U-th input under the i-th rule; k is the k-th group of data in the training data set; i is the i-th rule.

[0060] Or: Figure 3 As shown, for each input node of the input layer, the order value generated by each transformation node corresponding to the input node in the spatial transformation layer in the order layer is obtained, and the order value of each transformation node corresponding to each input node is used as a rule value. The order values of one of the transformation nodes of all input nodes are added (two by two, three by three, n by n, etc.) and divided by n to obtain the average, and a rule value is obtained, thereby generating a certain number of rules.

[0061] Alternatively, for each input node of the input layer, obtain the order value generated by each transformation node corresponding to the input node in the spatial transformation layer in the order layer, and add the order value of each transformation node corresponding to the input node and the order value generated by each transformation node corresponding to the remaining input nodes in the spatial transformation layer (which can be added in pairs, threes, n / n, etc.), and divide by n to obtain the average, to generate a certain number of rules.

[0062] for example Figure 3 In the example, there are input nodes X and Y. After input node X is mapped to the spatial conversion layer, transformation nodes A1 and B1 are generated, and the order values generated in the order layer are A2 and B2. After input node Y is mapped to the spatial conversion layer, transformation nodes C1 and D1 are generated, and the order values generated in the order layer are C2 and D2. Then, the sum of A2 and C2 is divided by 2 as the first rule value, and the sum of B2 and D2 is divided by 2 as the second rule value. A2, B2, C2, and D2 are each used as a rule value, so there are a total of 6 rule values.

[0063] Alternatively, the sum of A2 and C2 divided by 2 is used as the first rule value; the sum of A2 and D2 divided by 2 is used as the second rule value; the sum of B2 and C2 divided by 2 is used as the third rule value; and the sum of B2 and D2 divided by 2 is used as the fourth rule value. There are a total of 4 rule values.

[0064] Therefore, the order values generated by the previous layer are added together to generate rule i. For example, when adding two by two, the formula is:

[0065]

[0066] Where w i is the rule strength of the layer; is the order value of the Zth input under the i-th rule; is the order value of the U-th input under the i-th rule; k is the k-th group of data in the training data set; i is the i-th rule.

[0067] (5) The rule amplification layer is used to amplify the rule values of the rule generation layer by multiplying them by weights to form the nodes of the rule amplification layer. The processed nodes are then summed to form the output layer.

[0068] (6) Output layer: The number of nodes in the output layer can be manually adjusted. This output layer serves as the output of the problem result, such as cutting force and surface roughness. Secondly, the number of nodes in the output layer can also be used as the input value of another problem to conduct further reasoning of the particle collision neural network inference model.

[0069] In this embodiment, the inference model can be coded using Matlab or Python.

[0070] Firstly, a particle collision neural network inference model and a particle swarm optimization algorithm are written in the Matlab compiler. The particle collision neural network inference model is used as the fitness function of the particle swarm optimization algorithm, and the output of the fitness function is the difference between the actual output of the inference model and the target output.

[0071] Secondly, high-speed cutting force data is collected and used as training and testing datasets. The training dataset is deployed in the particle swarm optimization algorithm, making it easy for the particle collision neural network inference model to call the training dataset at any time.

[0072] The number of particles, weights, and iteration times in the particle swarm optimization algorithm need to be set in advance according to the convergence ability of the particles. After the setting is completed, the particle collision neural network inference model is trained to obtain a trained particle collision neural network inference model. The trained particle collision neural network inference model needs to use a test data set to analyze the model's prediction ability. After successful training, the inference model can obtain the corresponding problem prediction results by inputting the corresponding problem parameters.

[0073] Because the number of particles, weights, and number of iterations in the particle swarm optimization algorithm can affect its global convergence ability, and even cause the particle swarm optimization algorithm to converge to a local optimum during some training processes, this can affect the determination of unknown parameters in the particle collision neural network inference model and its prediction ability. In this embodiment, the particle collision neural network inference model is trained multiple times, and the inference model with the smallest error value of the fitness function is selected as the final particle collision neural network inference model.

[0074] It can be understood that the node parameters in the inference model can be trained and learned by the particle swarm optimization algorithm, or by other swarm intelligence optimization algorithms.

[0075] In order to verify the reasoning and prediction function of the above inference model, the following verification example is set up.

[0076] Verification Example 1:

[0077] For the parameter settings in the particle swarm optimization algorithm, the number of iterations is 500, the number of particles is 50, and the acceleration constant is 2. The detailed process is as follows:

[0078] The velocity and position of each particle are:

[0079]

[0080]

[0081] Where, X i (t) is the position of the i-th particle at the t-th iteration, V i (t) is the velocity of the i-th particle at the t-th iteration, and D is the dimension of the problem domain; is the position of the i-th particle in dimension j at the t-th iteration; is the velocity of the i-th particle in dimension j at the t-th iteration; T is the maximum number of iterations.

[0082] The good and bad positions of each particle are determined by the fitness function. There is a local historical optimal position Pbest of the particle in the particle community. i (t) and the global optimal position of the particle Gbest i (t).

[0083] When the conventional PSO algorithm iterates to the t+1th time, the position of each particle in each dimension and speed They are updated according to the information of the tth iteration, and the iterative formula is:

[0084]

[0085] Where, is the velocity of the i-th particle in dimension j at the t+1th iteration; is the local historical optimal position of the i-th particle in dimension j at the t+1th iteration; gbest j (t) is the global optimal position of the i-th particle in dimension j at the t+1th iteration; is the position of the i-th particle in dimension j at the t+1-th iteration; c1 and c2 are acceleration factors, c1 is used to adjust the moving step of a single particle to the historical optimal position, and c2 is the moving step of the particle group to the global optimal position; r1 and r2 are both random numbers in (0, 1); w is the inertia coefficient, which can enhance the exploration ability of the conventional PSO algorithm space and the control of the search range. The inertia coefficient w decreases linearly from 0.9 to 0.25, that is:

[0086]

[0087] Where w max With w min are the maximum inertia weight and the minimum inertia weight respectively; T is the maximum number of iterations.

[0088] In order to avoid the disadvantage of inaccurate late-stage convergence of particles in the conventional PSO algorithm, this embodiment proposes a vibration mechanism for the global optimal particle, which is divided into local micro-vibration of the global optimal particle and local strong vibration of the global optimal particle, namely:

[0089]

[0090] Where, is the particle at the global optimal position at the t+1th iteration, is the particle at the global optimal position at the tth iteration, r is a random number in (0, 1); V1 and V2 are the micro-vibration factor and strong vibration factor, respectively, which can be customized.

[0091] In this verification example, the performance of the model is verified and analyzed based on the particle collision neural network inference model trained above. The training algorithm is the particle swarm optimization algorithm, and the rule formation method is multiplication processing. The model structure is as follows: Figure 2 As shown in Figure 2, a cutting force dataset was obtained through high-speed cutting experiments. The dataset was divided into a training dataset and a test dataset. The test dataset was used to verify the prediction performance of the particle collision neural network inference model in terms of cutting force, and its prediction accuracy reached 91%.

[0092] Verification Example 2:

[0093] For the parameter settings in the particle swarm optimization algorithm, the number of iterations is 500, the number of particles is 50, and the acceleration constant is 2. The detailed process is as follows:

[0094] The velocity and position of each particle are:

[0095]

[0096]

[0097] Where, X i (t) is the position of the i-th particle at the t-th iteration, V i (t) is the velocity of the i-th particle at the t-th iteration, and D is the dimension of the problem domain; is the position of the i-th particle in dimension j at the t-th iteration; is the velocity of the i-th particle in dimension j at the t-th iteration.

[0098] The good and bad positions of each particle are determined by the fitness function. There is a local historical optimal position Gbest of the particle in the particle community. i (t) and the global optimal position of the particle Gbest i (t).

[0099] When the conventional PSO algorithm iterates to the t+1th time, the position of each particle in each dimension and speed They are updated according to the information of the tth iteration, and the iterative formula is:

[0100]

[0101]

[0102] Where, is the velocity of the i-th particle in dimension j at the t+1th iteration; is the local historical optimal position of the i-th particle in dimension j at the t+1th iteration; gbest j (t) is the global optimal position of the i-th particle in dimension j at the t+1th iteration; is the position of the i-th particle in dimension j at the t+1-th iteration; c1 and c2 are acceleration factors, c1 is used to adjust the moving step of a single particle to the historical optimal position, and c2 is the moving step of the particle group to the global optimal position; r1 and r2 are both random numbers in (0, 1); w is the inertia coefficient, which can enhance the exploration ability of the conventional PSO algorithm space and the control of the search range. The inertia coefficient w decreases linearly from 0.9 to 0.25, that is:

[0103]

[0104] Where w max With wmin are the maximum inertia weight and the minimum inertia weight respectively; T is the maximum number of iterations.

[0105] In order to avoid the disadvantage of inaccurate particle convergence in the late stage of the conventional PSO algorithm, this embodiment proposes a vibration mechanism of the global optimal particle, which is divided into local area micro-vibration of the global optimal particle and local area strong vibration of the global optimal particle, namely:

[0106]

[0107] Where, is the particle at the global optimal position at the t+1th iteration, is the particle at the global optimal position at the tth iteration, r is a random number in (0, 1); V1 and V2 are the micro-vibration factor and strong vibration factor, respectively, which can be customized.

[0108] In this verification example, the performance of the model is verified and analyzed based on the particle collision neural network inference model trained above. The training algorithm is the particle swarm optimization algorithm, and the rule formation method is the addition process. The model structure is as follows: Figure 3 As shown in Figure 2, a cutting force dataset was obtained through high-speed cutting experiments. The dataset was divided into a training dataset and a test dataset. The test dataset was used to verify the prediction performance of the particle collision neural network inference model in terms of cutting force, and its prediction accuracy reached 90%.

[0109] Example 2

[0110] This embodiment provides a milling cutting force prediction system using a particle collision neural network, comprising:

[0111] The training model is configured to use the acquired historical cutting data of the milling process as a training set to train an inference model based on a particle collision neural network;

[0112] The inference model includes an input layer that takes cutting speed, feed rate, and cutting depth as input, an output layer that takes cutting force and surface roughness as output, and an intermediate layer for implementing particle collisions. The intermediate layer uses cutting speed, feed rate, and cutting depth as nodes, calculates the order value of particle collisions between nodes, and fuses the order values of different nodes to obtain the output value.

[0113] The prediction model is configured to use the trained inference model to obtain a predicted cutting force value and a corresponding surface roughness according to the cutting data to be measured.

[0114] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0115] In further embodiments, there is also provided:

[0116] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor, wherein when the computer instructions are executed by the processor, the method described in Example 1 is performed. For the sake of brevity, no further details are given here.

[0117] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0118] The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0119] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is performed.

[0120] The method in Example 1 can be directly implemented as a hardware processor, or can be implemented using a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, it will not be described in detail here.

[0121] A computer program product includes a computer program, which implements the method described in embodiment 1 when executed by a processor.

[0122] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions contained in program modules, which are executed in a device on a real or virtual processor of a target to perform the process / method described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided between program modules as needed. The machine-executable instructions for the program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0123] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the computer or other programmable data processing device, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0124] In the context of the present invention, computer program code or related data can be carried by any appropriate carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals include electrical, optical, radio, acoustic, or other forms of propagated signals, such as carrier waves, infrared signals, and the like.

[0125] Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0126] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for predicting cutting force in milling using a particle collision neural network, characterized in that: include: The historical cutting data of the milling process is used as a training set to train an inference model based on a particle collision neural network; The inference model includes an input layer that takes cutting speed, feed rate, and cutting depth as input, an output layer that takes cutting force and surface roughness as output, and an intermediate layer for implementing particle collisions. The intermediate layer uses cutting speed, feed rate, and cutting depth as nodes, calculates the order value of particle collisions between nodes, and fuses the order values of different nodes to obtain the output value. The trained inference model is used to obtain the predicted cutting force and the corresponding surface roughness according to the cutting data to be measured.

2. The method for predicting cutting force in milling using a particle collision neural network according to claim 1, wherein: The intermediate layer includes a space conversion layer, an order layer, a rule generation layer and a rule amplification layer.

3. The method for predicting cutting force in milling using a particle collision neural network according to claim 2, wherein: The spatial conversion layer is used to divide each input node of the input layer into at least two transformation nodes, and perform weight exchange transformation between the transformation nodes of different input nodes; The order layer is used to generate an order value based on the result of weight exchange transformation of each transformation node using a Gaussian function and after particle collision; The rule generation layer is used to generate a rule value based on the order value of each transformation node; The rule amplification layer is used to sum the generated rule values after amplification processing to obtain the output result of the output layer.

4. The method for predicting cutting force in milling using a particle collision neural network according to claim 3, wherein: The process of weighted exchange transformation is as follows: after multiplying each transformation node corresponding to one of the input nodes by the weight, each weighted transformation node obtained is input into each transformation node corresponding to the remaining input nodes in a certain proportion and accumulated, and the sum of the proportions is 1, thereby updating the transformation nodes corresponding to the remaining input nodes, and at the same time, the transformation node corresponding to the current input node is updated to the difference between the original transformation node and the weighted transformation node, and then each updated transformation node value is input into the order layer; wherein the weight is a set value greater than 0 and less than 1.

5. The method for predicting cutting force in milling using a particle collision neural network according to claim 2, wherein: The rule generation layer includes: for each input node of the input layer, obtaining the order value generated by each transformation node corresponding to the input node in the spatial transformation layer in the order layer, and multiplying the order value of each transformation node corresponding to the input node with the order value generated by each transformation node corresponding to the remaining input nodes in the spatial transformation layer, thereby generating a rule value; wherein the order value nn is multiplied, and n is the number of input nodes.

6. The method for predicting cutting force in milling using a particle collision neural network according to claim 2, wherein: The rule generation layer includes: for each input node of the input layer, obtaining the order value generated by each transformation node corresponding to the input node in the spatial transformation layer in the order layer, taking the order value of each transformation node corresponding to each input node as a rule value, adding and averaging the order value nn of one transformation node of all input nodes to obtain a rule value, where n is the number of input nodes; Alternatively, for each input node of the input layer, the order value generated by each transformation node corresponding to the input node in the spatial transformation layer is obtained, and the order value of each transformation node corresponding to the input node is added to the order value generated by each transformation node corresponding to the remaining input nodes in the spatial transformation layer by nn addition and averaging to generate a rule value.

7. A particle collision neural network milling cutting force prediction system, characterized in that: include: The training model is configured to use the acquired historical cutting data of the milling process as a training set to train an inference model based on a particle collision neural network; The inference model includes an input layer that takes cutting speed, feed rate, and cutting depth as input, an output layer that takes cutting force and surface roughness as output, and an intermediate layer for implementing particle collisions. The intermediate layer uses cutting speed, feed rate, and cutting depth as nodes, calculates the order value of particle collisions between nodes, and fuses the order values of different nodes to obtain the output value. The prediction model is configured to use the trained inference model to obtain a predicted cutting force value and a corresponding surface roughness according to the cutting data to be measured.

8. An electronic device, characterized in that: The invention comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when executed by a processor.

Images