Image data driven numerical control machining cutting force prediction and standard test tool path generation method

Abstract

A numerical control machining cutting force prediction and standard test tool path generation method driven by image data is characterized by comprising the steps that firstly, a standard test tool path is generated according to a tool given by a user, discretization operation is conducted on cutting geometry in the simulation cutting process on the tool path, digital image characterization is conducted, and a gray level image is generated; taking the data as an input, testing to obtain a cutting force peak value in each cutting period, taking the cutting force peak value as an output, and training the data by using a convolutional neural network to fit a cutting force model; and finally, realizing cutting force off-line prediction under different process parameters through the cutting force model obtained through training. According to the method, the standard test tool path and the corresponding cutting geometric image can be generated according to the tool given by a user, training is carried out through cutting force data obtained through monitoring under a small amount of real cutting working conditions, and accurate prediction of the tool under complex working conditions is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of cutting force prediction, in particular to a cutting force prediction technology driven by image data, specifically to an image data driven CNC machining cutting force prediction and standard test tool path generation method. Background Art

[0002] In the manufacturing of complex aerospace parts, to meet aircraft weight reduction requirements and maintain structural performance, features such as holes and ribs are often added during the design and manufacturing process, further complicating the machining process. Drastic fluctuations in cutting forces can easily lead to significant deformation in thin-walled components, such as aircraft skins, ribs, and webs. This compromises machining accuracy and hinders automated production. Therefore, a method for predicting cutting forces during CNC machining is needed to better ensure workpiece quality and improve machining efficiency.

[0003] Traditional cutting force prediction methods also model the mechanism of cutting forces generated during machining, primarily involving modeling cutting geometry and cutting force. Cutting geometry modeling generally involves analytical geometry modeling and discrete geometry modeling. However, the definition of irregular cutting geometry, especially for the characterization of free-form workpieces, presents complex modeling processes. Furthermore, the cutting geometry continuously changes in real cutting conditions, making it less practical. Data-driven learning models exist for cutting force prediction models, but these models are unable to capture the input-output mapping relationship under real-world conditions during training. In particular, when cutting geometry undergoes irregular changes, the training set can easily overfit, impacting prediction accuracy.

[0004] In response to these problems mentioned above, the present invention proposes a method for CNC machining cutting force prediction and standard test tool path generation based on image data drive. It takes advantage of digital images to characterize the cutting geometry, reduces the difficulty of obtaining the cutting geometry while ensuring modeling accuracy, introduces convolutional neural networks, and limits the output of the model to the corresponding cutting force data, making it possible to accurately predict the cutting force. Summary of the Invention

[0005] This invention addresses the difficulties of existing cutting geometry modeling and low prediction accuracy by developing an image-driven method for predicting cutting forces and generating standard test toolpaths in CNC machining. Users can input machining process parameters such as tool depth of cut, width of cut, tool speed, and feed rate to generate a standard test toolpath. The cutting process is then simulated, and a neural network is trained using the generated grayscale image of the workpiece to produce a network model capable of predicting cutting forces.

[0006] The technical solution of the present invention is:

[0007] An image data-driven method for predicting cutting forces and generating standard test toolpaths for CNC machining is characterized by: first, generating a standard test toolpath based on a user-given tool, discretizing the cutting geometry during the simulated cutting process on this toolpath, performing digital image representation, and generating a grayscale image; using this as input, and testing the peak cutting force within each cutting cycle as output, a convolutional neural network is used to train the data and fit a cutting force model; finally, the trained cutting force model is used to achieve offline prediction of cutting forces under different process parameters.

[0008] The specific steps include:

[0009] First, the user inputs the tool radius and machining process parameters to generate a standard test tool path, and simulates the height grayscale image of the workpiece to be machined during the tool cutting process based on the parameters.

[0010] Second, the cutting force and its three cutting components of the actual cutting process are collected and processed to obtain the maximum cutting force in each cutting cycle.

[0011] Third, the grayscale image is used to correspond to the screened cutting force to construct a convolutional neural network and generate a neural network model.

[0012] Finally, based on the obtained neural network model, the user can independently input the required process parameters and predict the x-axis, y-axis and z-axis cutting force components respectively.

[0013] The height grayscale image of the workpiece processed during the simulation of tool cutting according to parameters refers to modeling the tool and obtaining the height value of its envelope surface. Then, when the tool envelope surface moves on the surface of the workpiece to be processed, it is continuously determined whether the tool envelope surface and the surface to be processed intersect based on the height value. When it is determined that they intersect, if the height of the surface of the workpiece to be processed is higher than the tool envelope surface, it is considered that the tool is cutting at this tool position, the height value of the surface to be processed is updated, and a grayscale image is generated.

[0014] The machining process parameters mainly include radial cutting depth, tool radius, feed rate, tool speed, etc.

[0015] The collection of the cutting force and its three cutting components during the actual cutting process and processing thereof means screening out the x-axis, y-axis and z-axis cutting force components corresponding to the time when the cutting force is maximum in a cutting cycle, for use in the next step of constructing a convolutional neural network.

[0016] The construction of the convolutional neural network refers to convolving the grayscale image generated above and training it using a fully connected neural network. The generated neural network model is generally a *.mat format file.

[0017] The beneficial effects of the present invention are:

[0018] The present invention provides an image data-driven method for predicting cutting force in CNC machining and generating a standard test tool path. The method can generate a neural network model using digital images under the condition of a small amount of sample cutting force data, and predict the cutting force under different machining parameters. The method is simple and efficient, and enriches the method for predicting cutting force in the CNC machining process, thereby making the machining process more controllable. The method can improve the machining process in a targeted manner for parts with excessive cutting force, thereby improving machining efficiency while ensuring workpiece quality.

[0019] The present invention can generate a standard test tool path and a corresponding cutting geometry image based on a given tool by the user, and train the cutting force data obtained by monitoring a small amount of real cutting conditions to achieve accurate prediction of the tool under complex working conditions.

[0020] The present invention avoids the problems of low prediction accuracy and poor generalization in traditional cutting force prediction methods based on mechanism models due to incomplete input cutting geometry information. By introducing grayscale images to model and represent the cutting geometry, and using this as the input of a convolutional neural network to fit the cutting force, the accurate prediction of the cutting force under different process parameters can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] FIG1( a ) is a standard test tool path in Example 1 of the present invention.

[0022] FIG1( b ) is a schematic diagram of different cutting depths during a cutting process in Example 1 of the present invention.

[0023] Figure 2 This is an illustration of generating a single grayscale image in Example 1 of the present invention.

[0024] Figure 3 Schematic diagram of the convolutional neural network structure in Example 1 of the present invention.

[0025] Figure 4 Schematic diagram of the comparison between the predicted values ​​and the actual values ​​of the x-axis and y-axis components of the cutting force in Example 1 of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described below with reference to the accompanying drawings and examples.

[0027] As shown in Figure 1-4.

[0028] A method for predicting cutting forces and generating standard test tool paths for numerical control machining based on image data drive, the method comprising the following steps:

[0029] S1: The user inputs the tool radius and machining process parameters to generate a standard test tool path, and simulates the height grayscale image of the workpiece to be machined during the tool cutting process according to the parameters.

[0030] S2: The cutting force and its three cutting components (x-axis, y-axis, and z-axis) of the actual cutting process are collected and processed to obtain the maximum cutting force in each cutting cycle.

[0031] S3: Using the obtained grayscale image and corresponding to the screened cutting force, a convolutional neural network is constructed to generate a neural network model.

[0032] S4: Based on the obtained neural network model, users can independently input the required process parameters and predict the three-axis cutting force components.

[0033] Details are as follows:

[0034] Example 1: Take the prediction of the x-axis and y-axis components of the cutting force of a ball-end milling cutter as an example to illustrate.

[0035] User needs: The cutting force of a specific machining process needs to be predicted.

[0036] In order to prove the effectiveness and versatility of the method described in this invention, a φ12 ball end milling cutter is selected for verification. The radius of the ball end milling cutter and the process parameters are input to generate the standard test tool path shown in Figure 1(a). Figure 1(b) is a schematic diagram of different cutting depths during a cutting process. Based on Figure 1, a series of grayscale images of the workpiece to be processed are generated. Figure 2 As shown in the figure, part of the image and the two cutting components of the x-axis and y-axis corresponding to the maximum cutting force in the cutting cycle are used as the training set input Figure 3 The convolutional neural network is trained. Then, based on the obtained neural network model, the user can independently input the required tool speed, radial cutting depth and other process parameters to verify the accuracy of the model, and obtain the predicted cutting force x-axis and y-axis components and compare them with the actual cutting force. Figure 4 shown.

[0037] The image data-driven CNC machining cutting force prediction and standard test tool path generation method described in the present invention innovatively uses tool parameters to generate standard test tool paths and introduces digital images to generate a neural network model to predict cutting forces under different machining parameters. The method is simple, effective and universal.

[0038] The parts not involved in the present invention are the same as the existing technology or can be implemented by using the existing technology.

Claims

1. An image data driven CNC machining cutting force prediction and standard test tool path generation method, characterized in that: Firstly, a standard test tool path is generated according to the tool given by the user. The cutting geometry is discretized in the simulated cutting process on this tool path, and digital image representation is performed to generate a grayscale image. This is used as input, and the peak cutting force in each cutting cycle is tested as output. The convolutional neural network is used to train the data to fit the cutting force model. Finally, the cutting force model obtained through training is used to realize offline prediction of cutting force under different process parameters.

2. The method according to claim 1, characterized in that: The generation of standard test tool paths refers to the autonomous generation of cutting paths containing different process parameters according to the tool radius and maximum cutting depth given by the user, including defining the tool path step according to the tool radius, thereby gradually changing the tool spacing in sequence to form a tool path with a variable cutting width until a full tool spacing appears; at the same time, during the tool movement process, the cutting depth is gradually increased according to the proportion of the tool radius, and after reaching the rated maximum cutting depth of the tool, the next set of tool path spacing cutting is carried out, and the above steps are repeated until the tool path is exited; the process parameters include cutting depth and cutting width.

3. The method according to claim 1, characterized in that: The cutting geometry refers to the geometry of the workpiece removed by the tool during one cutting cycle. A cutting cycle is defined as the time taken for the tool to rotate one circle. The cutting geometry is the geometric element of the workpiece removed by the tool during one rotation of the tool.

4. The method according to claim 1, characterized in that: The geometric discretization operation refers to the grid generation of the cutting geometric surface. The direction of the straight line where the tool axis is located pointing to the outside of the workpiece is set as the normal direction of the discretized grid. The axial coordinate values ​​of the tool on the upper and lower surfaces at each grid position are subtracted to obtain the axial height value of the cutting geometric element; wherein the axial height value of the tool of the geometric element is represented by the height grayscale value of the corresponding position in the digital image.

5. The method according to claim 1, characterized in that: The digital image representation means that in the simulation process, the tool is first modeled and its envelope surface height value is obtained, and then the height value of the tool envelope surface when it moves on the surface of the workpiece to be processed is used to continuously determine whether the tool envelope surface and the surface to be processed intersect, and the two surfaces are discretely meshed at the same scale; when the two meshed surfaces are determined to intersect, if the height of the surface of the workpiece to be processed is higher than the tool envelope surface, it is considered that the tool is cutting at this tool position, the axial height value of the surface to be processed is updated, and a height grayscale image is generated.

6. The method according to claim 1, characterized in that: The use of a convolutional neural network to train data to fit the cutting force model refers to: using the cutting geometry image generated by simulation as the input of the neural network, and taking the maximum cutting force per cycle measured by the corresponding experiment as the label value for model training, obtaining an input-output mapping relationship, and thus generating a cutting force model; the user only needs to input the tool and tool path to predict the corresponding cutting force.

Images