Milling parameter optimization method based on combination of offline and online monitoring

Abstract

The application discloses a kind of milling parameter optimization methods based on offline and online monitoring combination, real-time consider the dynamic characteristics in machining process, including tool wear state in machining process, the surface roughness of the workpiece being processed and the cutting signal change of machining process, and then establish tool state and machining surface quality prediction module, cutting parameter optimization module and intelligent decision module.The present application is practical and feasible, solves the lack of consideration of the physical characteristics of the entire machining system in the existing cutting parameter optimization method, makes up for the shortcomings of the experience-based parameter selection method, improves the machining quality and machining efficiency of the part, and provides theoretical support for the selection and optimization of machining parameters in the milling process of the part.

Description

Technical Field

[0001] This invention belongs to the field of machining technology, specifically a milling parameter optimization method based on a combination of offline and online monitoring. Background Technology

[0002] In aerospace, energy equipment, and other fields, there are numerous complex parts characterized by high material removal rates, difficult-to-machine materials, harsh service environments, and high machining accuracy requirements. To ensure machining quality, conservative cutting parameters are often used in actual machining, leading to underutilization of machine tools and cutting tools and low machining efficiency. To improve both product quality and machining efficiency, some companies have adopted manual control methods. For example, operators adjust parameters based on extensive experience; however, differences in experience and knowledge among technicians result in inconsistent product quality. Furthermore, cutting processes are time-varying and non-linear, making it difficult to accurately judge the entire machining process based on experience alone. Therefore, there is an urgent need for scientific guidance methods to ensure the stability and reliability of the machining process and improve machining efficiency.

[0003] Optimizing cutting parameters in milling is a complex nonlinear, multi-constraint problem, but it is an effective way to improve machining quality and efficiency. Optimization of cutting parameters mainly involves adjusting parameters based on responses such as tool life, material removal rate, power consumption, and surface roughness during machining. Cutting parameter optimization primarily includes single-objective optimization and multi-objective optimization. Due to the complexity of the machining process, single-objective methods have limited value in determining optimal cutting conditions. Multi-objective optimization simultaneously optimizes several different and contradictory objectives, thus being closer to practical engineering. However, the cutting process is dynamic and unstable, and the specific values ​​of objectives and constraints are not easily obtained. Although some scholars have proposed theoretical models to describe the cutting process, due to simplification, they still differ significantly from reality. Furthermore, tool wear during machining is also a major challenge affecting cutting parameter optimization, especially for difficult-to-machine materials, where force-thermal coupling accelerates tool wear, leading to a sharp decrease in tool life.

[0004] Existing parameter optimization methods lack a comprehensive consideration of the physical characteristics of the entire machining system, such as cutting force, cutting torque, and power, when optimizing process parameters. Surface roughness and tool wear are key issues that need to be considered when machining difficult-to-machine materials. Surface roughness deteriorates under tool wear, so the optimization model should update parameters in real time according to changing machining conditions. However, the applicability of existing cutting parameter optimization methods to these conditions lacks in-depth and systematic research. Summary of the Invention

[0005] The present invention aims to provide a milling parameter optimization method based on a combination of offline and online monitoring. It takes into account tool wear and surface roughness during the machining process, solves the problem that existing cutting parameter optimization methods lack consideration of the physical characteristics of the entire machining system, and makes up for the shortcomings of experience-based parameter selection methods. This improves the quality and efficiency of part machining and provides an effective way for the selection and optimization of cutting parameters in the milling process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A milling parameter optimization method based on combined offline and online monitoring includes establishing a tool condition and machining surface quality prediction module, a cutting parameter optimization module, and an intelligent decision-making module, wherein:

[0008] The tool status and machining surface quality prediction module, on the one hand, establishes a prediction model based on the collected cutting signal data from the historical machining process, and realizes a nonlinear mapping from cutting signals to machining surface quality and tool wear through the prediction model; on the other hand, it obtains changes in cutting signals through sensing technology, and dynamically updates the established prediction model parameters and the boundary values ​​of the constraint conditions used for cutting parameter optimization in real time. It should be noted that the sensing technology in this invention refers to a solution that can convert the changing physical quantities in the cutting process into processable digital signals through a series of sensors.

[0009] The cutting parameter optimization module first establishes a Kriging model based on different inputs and outputs of objective function, design variables, and constraints using cutting experiment sample points. The design variables are the cutting parameters to be optimized, and their values ​​are selected from the cutting signals collected by the tool state and machining surface quality prediction module.

[0010] Next, the NSGA-II algorithm is used to optimize the cutting parameters of the established Kriging model to obtain the Pareto solution set after the first optimization. Then, the cutting signal obtained by the sensing technology in the tool state and machining surface quality prediction module and dynamically updated is used as the design variable, and the boundary value of the constraint condition is used as the constraint condition to obtain the new Pareto solution set after the second optimization.

[0011] The intelligent decision-making module first determines whether the milling process is in the roughing or finishing stage, and then determines the selection of the optimal solution in the Pareto solution set based on the different weights of the objective function in the roughing and finishing stages.

[0012] Furthermore, in the tool state and machining surface quality prediction module, a prediction model is established through a convolutional neural network to realize the nonlinear mapping from cutting signals to machining surface quality and tool wear.

[0013] Furthermore, the cutting parameter optimization module uses a central composite experimental design method to obtain sample points.

[0014] Furthermore, in the tool status and machining surface quality prediction module, the cutting signal includes, but is not limited to, the cutting force signal, the spindle three-phase current signal and the acceleration signal, the machining surface quality includes, but is not limited to, surface roughness, and the tool wear refers to the tool's service life.

[0015] Furthermore, in the cutting parameter optimization module, the objective function includes, but is not limited to, material removal rate, surface roughness, cutting force and cutting power; the design variables include, but are not limited to, cutting speed, cutting depth and feed per tooth; and the constraints include, but are not limited to, cutting force constraints, surface roughness constraints, maximum feed constraints of the machine tool spindle, maximum torque constraints of the machine tool spindle and maximum speed constraints of the machine tool spindle.

[0016] Compared with existing technologies, the optimization method of this invention takes into account tool wear and surface roughness during the machining process, and proposes different optimization strategies for roughing and finishing. The proposed optimization method is more effective and advantageous than other mainstream methods when optimizing the objective function, making up for the shortcomings of the experience-based parameter selection method, improving the machining quality and efficiency of parts, and providing an effective solution for the selection and optimization of machining parameters in the milling process of typical parts.

[0017] This invention has the following characteristics:

[0018] (1) The milling parameter optimization scheme proposed in this invention involves comprehensive constraints, covering cutting signals, power signals, and surface roughness of the machined parts, which better meets the requirements of actual production. More importantly, the optimization method proposed in this invention is not limited to these types of constraints. When the number of features to be optimized increases, the optimization strategy of this invention remains effective.

[0019] (2) The milling parameter optimization scheme proposed in this invention is supported by a database and prediction model established by historical cutting data. It is updated online in real time through sensing technology and offline data is obtained through experiments, providing a closed-loop optimization function. The optimization can be done by combining offline and online optimization and is applicable to different working conditions.

[0020] (3) The tool wear and surface quality are obtained by prediction through deep learning and multi-sensor fusion algorithm. The convolutional neural network algorithm proposed in this invention has high accuracy. The service life of the tool after wear will affect the boundary value changes of other constraints in the model.

[0021] (4) The present invention proposes different optimization strategies for the roughing stage and the finishing stage, which meet the actual production requirements.

[0022] (5) The optimization model proposed in this invention is based on the Kriging model, and its advantage is that it can provide the optimal linear unbiased estimate.

[0023] (6) In this invention, the Pareto solution optimization of the Kriging model is solved by the NSGA-II multi-objective genetic algorithm. Compared with traditional mathematical methods, its advantages are fast speed and good solution set convergence. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating the principle of the milling parameter optimization scheme of the present invention;

[0025] Figure 2 This is a flowchart of the cutting parameter optimization method of the present invention;

[0026] Figure 3 This is the response of the Kriging model to the output under different cutting parameters in this invention;

[0027] Figure 4 This is a schematic diagram of the Pareto solution during roughing in this invention;

[0028] Figure 5 This is a schematic diagram of the Pareto solution under the three objective functions for roughing in this invention;

[0029] Figure 6 This is a schematic diagram illustrating the principle of the tool condition and machining surface quality prediction module of the present invention, which uses a convolutional neural network to establish a prediction model. Detailed Implementation

[0030] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. However, it should not be construed that the scope of the subject matter of the present invention is limited to the following embodiments. All modifications, substitutions and alterations made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention are included within the scope of the present invention.

[0031] This invention takes into account the dynamic characteristics of the milling process, including the tool wear state, the surface roughness of the workpiece, and the changes in the cutting signal during the milling process, to optimize the cutting parameters.

[0032] The cutting parameter optimization scheme proposed in this invention is as follows: Figure 1 As shown, the scheme mainly consists of three parts, including a tool status and machining surface quality prediction module, a cutting parameter optimization module, and an intelligent decision-making module.

[0033] The tool condition and machining surface quality prediction module consists of two parts. First, a convolutional neural network prediction model is established based on historical data collected during the machining process. This model achieves a nonlinear mapping between cutting signals (cutting signals refer to signals collected by sensors during the cutting process, such as collected cutting force signals, collected spindle three-phase current signals, collected acceleration signals, etc., but not limited to the three signals mentioned above) and machining surface quality (such as surface roughness mentioned in this invention, and can be further extended to surface shape errors, etc.) and tool wear (mainly referring to tool life). Figure 6 As shown, Figure 6 (a) is a convolution operation, which can be described as a convolution kernel sliding across the input feature map to obtain an output feature map. In each convolution part, several convolution kernels are used to obtain different feature maps to increase feature diversity. Figure 6 (b) shows cutting signals from multiple sensors and multiple channels. Features were learned from these signals using a developed parallel convolutional network. Finally, the learned features were fused in series along the channel dimension to achieve feature fusion of the multi-sensor signals. For example, Hall current sensors were installed at the three-wire power input terminals of the CNC machine tool spindle motor to collect three-phase current signals, and cutting force signals in three directions were collected using, but not limited to, a Kistler force measurement system. Cutting parameters were determined using, but not limited to, the optimal Latin hypercube experimental design method. Basic cutting experiments were then conducted to obtain cutting force data and spindle current data under each set of parameters to build, train, and test the model.

[0034] Secondly, by using sensing technology to dynamically update the parameters of the prediction model and the constraint boundary conditions (i.e., the boundary values ​​of the constraint conditions) for the optimization of cutting parameters in real time, the established prediction model can be optimized and adjusted according to the machining state, thereby improving the prediction accuracy of the prediction model and providing constraint boundary conditions for the cutting parameter optimization module.

[0035] Cutting parameter optimization module: This module mainly optimizes cutting parameters. It provides the constraint boundary conditions for cutting parameter optimization through the tool state and machining surface quality prediction module, and uses the non-dominated sorting genetic algorithm NSGA-II to perform the initial optimization of cutting parameters to obtain the Pareto solution set. At the same time, it uses sensing technology to sense changes in the cutting process in real time, such as tool wear, cutting signals during machining, and surface roughness of the machined part. Combining the sensing results, the cutting parameters are optimized again to obtain a new Pareto solution set.

[0036] Intelligent Decision Module: After obtaining the Pareto optimal solution set based on multi-objective optimization, it is difficult to derive the optimal solution according to actual needs. Therefore, this intelligent decision analysis module first determines whether the optimization belongs to the roughing or finishing stage, then determines the selection of the optimal solution based on the importance that process engineers attach to the objective function, and finally uses the optimized optimal parameters for actual processing.

[0037] The core of this invention is the cutting parameter optimization module, and the specific optimization process of the cutting parameter optimization module is as follows: Figure 2 As shown, it includes the following steps:

[0038] Step 1: Determine the objective function, design variables, constraints, and range of feasible solutions for the cutting parameters to be optimized.

[0039] Step 2: Based on the sample points of the central composite experiment design, establish a Kriging model based on the objective function, design variables, and constraints. Use the NSGA-II algorithm to optimize the cutting parameters of the established Kriging model and obtain the optimized Pareto solution set.

[0040] Step three: Based on actual engineering requirements, the Pareto optimal solution set is sorted to obtain the optimal combination of cutting parameters for actual machining, ultimately achieving efficiency improvement and surface quality control of the machined parts. In NSGA-II, the cutting parameters in the design variables are binary encoded and initialized. Then, crossover, mutation, and selection operations are performed on the generated initial population, thus generating the first generation of cutting parameter subgroups. In the second generation evolution process, the cutting parameter subgroups and the initial population are integrated, and crowding calculation and fast non-dominated sorting are performed to generate a new parent group. After crossover, selection, and mutation operations are performed again, the next generation of iterative calculations begins until the set number of evolution generations is reached. After iteration, the population generated in the last generation can usually be taken as the optimal solution to the optimization problem. After decoding, the optimal set of process parameter solutions can be obtained. After optimization in NSGA II, considering different optimization objectives, multiple sets of Pareto solutions for process parameters are formed.

[0041] In step one, this embodiment uses the cutting speed, depth of cut, and feed per tooth as design variables during the cutting process, i.e.: x = [v c ,a p ,f z ] TThe objective function is divided into two stages. During roughing, the main objective is to limit the cutting load within a certain range, maximize material removal rate, improve tool life, and reduce energy consumption. Therefore, the optimization objective function for this stage is to maximize material removal rate, minimize cutting force and cutting power, and ensure the quality of the machined surface. During finishing, the optimization objectives are mainly the surface quality of the machined part, tool life, and cutting power. Therefore, the objective function for this stage is determined to be the minimum material removal rate, minimum surface roughness, minimum cutting force, and minimum cutting power. The objective function is defined as follows:

[0042]

[0043] The constraints for roughing and finishing are basically the same. The difference is that in the roughing stage, surface roughness is used as a constraint, meaning that one of the constraints in roughing is simply to meet general surface roughness requirements. In the finishing stage, the constraint is to ensure that the roughness is minimized. The roughness is obtained through a convolutional neural network prediction model. Other major constraints include machine tool constraints, which stipulate that the maximum allowable spindle feed, maximum spindle torque, and maximum speed during machining should not exceed the machine tool's allowable values.

[0044] In step two, a cutting parameter optimization model based on Kriging is established. First, the research object is determined, then experimental design is conducted, and Kriging models with different inputs and outputs are established based on the sample points obtained from the experimental design. Finally, an optimization algorithm is used to optimize the established Kriging model to find the optimal design variable values, which are then used in actual production. In this embodiment, the Kriging model is first established based on the sample points obtained from the cutting experiments conducted using the central composite design method. The input variables are cutting speed, depth of cut, and feed per tooth, and the output responses are material removal rate and surface roughness, respectively. Material removal rate and surface roughness are chosen as the output responses because in actual machining, the cost of parts is often much higher than the cost of tools (up to millions or even tens of millions of times), such as large gear rings, guide vanes, and blades. Therefore, in actual production, the primary goal is to machine parts with high quality without damage. After all, tool wear / scrap is much cheaper than part scrap. Therefore, when optimizing milling parameters, tool wear is not directly used as one of the output responses; instead, surface roughness and material removal rate are used as the primary responses because the primary goal is to machine parts that meet surface quality requirements. However, after tool wear, the cutting load increases. Therefore, the cutting force or spindle current signal collected by the online sensing technology in the tool condition and machining surface quality prediction module will change, for example, its amplitude will increase. This indirectly reflects the impact of tool wear, and consequently changes the boundary values ​​of the constraints (demonstrating the influence of tool wear on milling parameter optimization in this invention). Of course, the response can also be extended to cutting force or cutting power, etc., as needed; this embodiment only provides an example. Kriging models from the above cutting parameters to the response are established respectively, and the established models are as follows: Figure 3 As shown. Figure 3 The paper demonstrates the response of cutting parameters to material removal rate and surface roughness.

[0045] Based on the above process and the central composite experimental design method to obtain sample points, after establishing an optimization model based on Kriging, taking roughing as an example, the NSGA-II optimization algorithm is used to optimize and solve the Kriging model in the roughing stage.

[0046] During roughing, the goal is to limit the cutting load within a certain range, maximize material removal rate, improve tool life, and reduce energy consumption. Therefore, the objective function for this stage is to maximize material removal rate, minimize cutting force and cutting power, and ensure the quality of the machined surface.

[0047] In the NSGA-II algorithm, the population size is set to 20, the crossover probability to 0.98, the number of generations to 20, and the number of iterations to 1000, for multi-objective optimization. Since this optimization problem is multi-objective, the results need to be converted into a Pareto optimal solution set problem to construct a non-dominated solution set.

[0048] Pareto solution optimized for the roughing stage is as follows Figure 4 As shown, Figure 4 Each point in the set represents a Pareto solution that meets the requirements after optimization, and the entire set constitutes the Pareto solution set. Figure 4 The Pareto solutions can be divided into three regions. Region A has relatively low cutting force and material removal rate. Region B offers a suitable balance between material removal rate and cutting force. Region C has a high material removal rate but also a high cutting force. To balance these two factors, in practice, solutions from region B can be used as the optimized solution set for actual production, depending on requirements.

[0049] The Pareto solution considering the three objectives of cutting power, cutting force, and material removal rate is as follows: Figure 5 As shown in the figure, all Pareto solutions can be decomposed into three regions: A, B, and C. Region A has relatively low cutting force and cutting power, but also relatively low material removal rate. Furthermore, a positive correlation can be observed between cutting force and cutting power in this region. Therefore, to simplify computation time, these two objectives can be considered as one in multi-objective optimization. The Pareto solution in region B is the ideal solution that simultaneously considers low cutting force and cutting power while maximizing material removal rate. Region C represents the solution that prioritizes material removal rate.

[0050] In actual machining, changes in cutting parameters and tool condition affect the final machining result. Traditional cutting parameter optimization methods only establish optimization models for the machine tool and process at a certain moment, without considering the time-varying characteristics of tools and machine tools, thus affecting the accuracy of optimization results and failing to fully utilize the tools and machine tools. This invention proposes a milling parameter optimization method based on a combination of offline and online monitoring. The optimization method considers tool wear and surface roughness during machining and proposes different optimization strategies for roughing and finishing, making up for the shortcomings of experience-based parameter selection methods, improving part machining quality and efficiency, and providing theoretical support for the selection and optimization of machining parameters in the part milling process. Contents not described in detail in this specification are prior art known to those skilled in the art. Although the invention has been described in specific embodiments to facilitate understanding by those skilled in the art, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the claims, and all inventions utilizing the concept of this invention are within the protection scope of this invention.

Claims

1. A milling parameter optimization method based on combined offline and online monitoring, characterized in that: This includes a tool condition and machining surface quality prediction module, a cutting parameter optimization module, and an intelligent decision-making module, among which: The tool status and machining surface quality prediction module, on the one hand, establishes a prediction model based on the collected cutting signal data from the historical machining process, and realizes a nonlinear mapping from cutting signals to machining surface quality and tool wear through the prediction model; on the other hand, it obtains changes in cutting signals through sensing technology and dynamically updates the established prediction model parameters and the boundary values ​​of the constraint conditions used for cutting parameter optimization in real time. The cutting parameter optimization module first establishes a Kriging model based on different inputs and outputs of objective function, design variables, and constraints using cutting experiment sample points. The design variables are the cutting parameters to be optimized, and their values ​​are selected from the cutting signals collected by the tool state and machining surface quality prediction module. Next, the NSGA-II algorithm is used to optimize the cutting parameters of the established Kriging model to obtain the Pareto solution set after the first optimization. Then, the cutting signal obtained by the sensing technology in the tool state and machining surface quality prediction module and dynamically updated is used as the design variable, and the boundary value of the constraint condition is used as the constraint condition to obtain the new Pareto solution set after the second optimization. The intelligent decision-making module first determines whether the milling process is in the roughing or finishing stage, and then determines the selection of the optimal solution in the Pareto solution set based on the different weights of the objective function in the roughing and finishing stages. In the tool condition and machining surface quality prediction module, a prediction model is established through a convolutional neural network to realize the nonlinear mapping from cutting signals to machining surface quality and tool wear; In the cutting parameter optimization module, the central composite experimental design method is used to obtain sample points; In the tool status and machining surface quality prediction module, the cutting signal includes the cutting force signal, the three-phase current signal of the spindle and the acceleration signal, the machining surface quality includes the surface roughness, and the tool wear refers to the tool's service life. In the cutting parameter optimization module, the objective function includes material removal rate, surface roughness, cutting force and cutting power, the design variables include cutting speed, cutting depth and feed per tooth, and the constraints include cutting force constraint, surface roughness constraint, maximum feed constraint of machine tool spindle, maximum torque constraint of machine tool spindle and maximum speed constraint of machine tool spindle.

Images