A Method and System for Positioning Single-Point Diamond Turning Tools Based on Variational Autoencoders

Abstract

This invention provides a method and system for positioning a single-point diamond turning tool based on a variational autoencoder. The method includes: extracting features from the input force signal and learning its data distribution based on an encoder network; performing batch normalization on the signals output by the encoder network; reconstructing the force signal based on the data distribution of the force signal in the batch-normalized signal using a decoder network; calculating the reconstruction error between the reconstructed force signal and the input force signal; and comparing the reconstruction error with a pre-generated state threshold to determine whether the turning tool is in the idle tool movement stage or the workpiece contact stage. By learning the signal feature distribution of the tool during the idle tool movement stage using a neural network based on a variational autoencoder and reconstructing the input force signal, the signal error of the reconstructed model is used to determine whether the tool is in contact with the workpiece, thereby achieving precise positioning of the single-point diamond turning tool. This method improves the accuracy of tool positioning and is low-cost and easy to operate.

Description

Technical Field

[0001] This invention relates to the field of turning tool positioning technology, specifically to a single-point diamond turning tool positioning method and system based on a variational autoencoder. Background Technology

[0002] With social development and scientific progress, the domestic machine tool industry has made great strides in recent years. The industry is huge in scale, but there is still much room for improvement in high-precision core technologies.

[0003] CNC lathes are among the most widely used CNC machine tools. They are primarily used for cutting and machining the inner and outer cylindrical surfaces, inner and outer conical surfaces with arbitrary cone angles, complex rotating inner and outer curved surfaces, and cylindrical and conical threads of shaft-like or disc-like workpieces. They can also perform grooving, drilling, reaming, boring, and other machining operations. The cutting tool is the main core component of a CNC lathe used for turning operations. Tool positioning is a crucial factor affecting the machining accuracy of a CNC lathe, thus placing higher demands on tool positioning.

[0004] A Chinese patent application with publication number CN103170877A discloses a method for precise positioning of a diamond turning tool during micro-turning. The method uses high-precision CCD measurement technology to extract the tool image, calculate the coordinate position of the tool tip in the CCD image system, and detect the three-dimensional positioning error of the diamond turning tool. Then, the measured tool tip position is compared with the original tool tip position, and error compensation is performed online in conjunction with the precision positioning platform of the machine tool system to improve the tool positioning accuracy.

[0005] Existing technologies utilize high-precision CCD measurement to extract tool images for tool positioning. This requires sophisticated measurement equipment, is costly, and the accuracy of the extracted images is difficult to guarantee, thus affecting the positioning accuracy of the tool. Therefore, there are areas for improvement. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a single-point diamond turning tool positioning method and system based on a variational autoencoder.

[0007] A single-point diamond turning tool positioning method based on a variational autoencoder, provided by the present invention, includes the following steps:

[0008] Step S1: Extract features from the input force signal and learn the data distribution of the input force signal based on the encoder network;

[0009] Step S2: Perform batch normalization processing on the signals output by the encoder network;

[0010] Step S3: Reconstruct the force signal based on the data distribution of the force signal in the batch-normalized signal using the decoder network;

[0011] Step S4: Calculate the reconstruction error between the reconstructed force signal and the input force signal;

[0012] Step S5: Compare the reconstruction error value with the pre-generated state threshold to determine whether the cutting tool is in the free-running stage or the workpiece contact stage.

[0013] Preferably, for step S1, the encoder network includes an adaptive function.

[0014] Preferably, for step S4, the reconstruction error value between the reconstructed force signal and the input force signal is calculated using the root mean square error.

[0015] Preferably, step S5 includes:

[0016] Step S5.1: Construct the state threshold for the tool's idle travel phase;

[0017] Step S5.2: Determine whether the cutting tool is in the free-running stage or the workpiece contact stage.

[0018] Preferably, step S5.1 includes:

[0019] Step S5.11: Obtain the force signal reconstruction error value during the tool idle travel phase;

[0020] Step S5.12: Calculate the kernel density function for the force signal reconstruction error value during the tool idle travel phase;

[0021] Step S5.13: Calculate the cumulative distribution function from the kernel density function calculation results;

[0022] Step S5.14: Obtain the state threshold belonging to the tool idle travel stage based on the significance level.

[0023] A single-point diamond turning tool positioning system based on a variational autoencoder provided by the present invention includes:

[0024] Modular block M1, based on encoder network, performs feature extraction of input force signal and learns the data distribution of input force signal;

[0025] Modular block M2 performs batch normalization processing on the signals output by the encoder network;

[0026] Modular M3 reconstructs the force signal based on the data distribution of the force signal in the batch-normalized signal using a decoder network.

[0027] Modular block M4 calculates the reconstruction error between the reconstructed force signal and the input force signal;

[0028] The module M5 compares the reconstruction error value with the pre-generated state threshold to determine whether the cutting tool is in the free-running stage or the workpiece contact stage.

[0029] Preferably, module M1, the encoder network includes an adaptive function.

[0030] Preferably, module M4 calculates the reconstruction error between the reconstructed force signal and the input force signal using the root mean square error.

[0031] Preferably, module M5 includes:

[0032] Submodule M5.1: Construct the state threshold for the tool's idle travel phase;

[0033] Submodule M5.2 determines whether the cutting tool is in the idle tooling stage or the workpiece contact stage.

[0034] Preferably, submodule M5.1 includes:

[0035] Sub-mold fast M5.11, obtain the force signal reconstruction error value during the tool idle travel phase;

[0036] The kernel density function is used to calculate the force signal reconstruction error value during the tool idle movement phase of the sub-mold fast M5.12.

[0037] Sub-module M5.13: Calculate the cumulative distribution function based on the kernel density function calculation results;

[0038] Sub-mode fast M5.14: Based on the significance level, obtain the state threshold belonging to the tool idle travel stage.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. This invention learns the signal feature distribution of the tool during the idle travel phase by using a neural network based on a variational autoencoder, and reconstructs the input force signal. The signal error of the reconstructed model is used to determine whether the tool is in contact with the workpiece, so as to achieve precise positioning of a single-point diamond turning tool. This helps to improve the positioning accuracy, and is low in cost and easy to operate.

[0041] 2. This invention solves the problems of feature learning robustness and model generalization by using batch normalization and adaptive activation functions, which helps to improve the robustness of feature learning.

[0042] 3. This invention solves the problem of adaptive threshold determination under non-stationary signals by using kernel density estimation method, which helps to improve the convenience of adaptive threshold determination. Attached Figure Description

[0043] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0044] Figure 1 This is a flowchart illustrating the diamond turning tool positioning method of this invention;

[0045] Figure 2 This invention primarily demonstrates the effect of the reconstruction force signal during the tool's empty-path movement.

[0046] Figure 3 This invention primarily demonstrates the effect of the reconstruction force signal during the contact stage between the tool and the workpiece.

[0047] Figure 4 This is a schematic diagram illustrating the overall structure of the diamond turning tool positioning system of this invention;

[0048] Figure 5 This diagram illustrates the output results of the state recognition module, which is the main feature of this invention.

[0049] Figure 6 This invention primarily demonstrates the tool positioning accuracy detected by a white light interferometer. Detailed Implementation

[0050] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0051] like Figure 1 As shown, a single-point diamond turning tool positioning method based on a variational autoencoder provided by the present invention includes the following steps:

[0052] Step S1: Feature extraction of the input force signal and learning of the data distribution of the input force signal based on the encoder network. Specifically, the encoder network includes an adaptive function. The encoder network of this application belongs to the variational autoencoder in existing neural network learning techniques. The variational autoencoder is a directed model that uses learned approximations for inference and can be trained using gradient-based optimization methods. Adding an adaptive function to the encoder network can improve the robustness of feature extraction of the input force signal.

[0053] Step S2: Perform batch normalization on the encoder network output signal to avoid gradient dissipation during the learning of the input force signal data distribution. Batch normalization, also known as batch standardization, is a method that re-centers and re-scales the layer inputs to make the training of artificial neural networks faster and more stable. During training, the distribution of the inputs to each layer changes continuously, requiring the next layer to constantly adapt to the new data distribution, making training very complex and slow. To solve this problem, a smaller learning rate and more stringent parameter initialization are needed. By using batch normalization, the mean and variance of a small batch are used to adjust the intermediate outputs of the neural network during model training, so that the outputs of each layer conform to a Gaussian distribution with the same mean and variance. This makes the data more stable; regardless of how the parameters of the hidden layers change, the mean and variance of the output data of the previous layer are known and fixed, thus solving the problems of slow training and small learning rates caused by constantly changing data distributions.

[0054] Step S3: Reconstruct the force signal based on the data distribution of the force signal in the batch-normalized signal using the decoder network. The decoder network in this application is part of the variational autoencoder in existing neural network learning techniques.

[0055] Step S4: Calculate the reconstruction error between the reconstructed force signal and the input force signal. Specifically, the root mean square error (RMSE) is used to calculate the reconstruction error between the reconstructed force signal and the input force signal. Figure 2 and Figure 3 As shown, there is a significant difference between the reconstruction error during the tool idle travel phase and the reconstruction error during the tool-workpiece contact phase.

[0056] Step S5: Compare the reconstruction error value with the pre-generated state threshold to determine whether the cutting tool is in the free-running stage or the workpiece contact stage.

[0057] Specifically, step S5 includes: step S5.1, constructing the state threshold for the tool idle travel phase.

[0058] More specifically, step S5.1 includes:

[0059] Step S5.11: Obtain the force signal reconstruction error value during the tool idle travel phase.

[0060] Step S5.12: Calculate the kernel density function for the force signal reconstruction error value during the tool idle travel phase.

[0061] Step S5.13: Calculate the cumulative distribution function based on the kernel density function calculation results.

[0062] Step S5.14: Obtain the state threshold belonging to the tool idle travel stage based on the significance level. In this application, the significance level is defined as the value with the highest significance level in the distribution function, which is used as the state threshold. The higher the significance level, the more accurate the obtained state threshold.

[0063] It should be noted that the method for calculating the state threshold during the tool idle travel phase can be the same as the method for obtaining the reconstruction error value in the above steps.

[0064] Step S5.2: Determine whether the cutting tool is in the idle tool movement stage or the workpiece contact stage. Specifically, compare the reconstruction error value with the state threshold: if the reconstruction error value is greater than the state threshold, the corresponding input force signal belongs to the tool-workpiece contact stage, and the cutting tool is in the workpiece contact stage. If the reconstruction error value is less than or equal to the state threshold, the corresponding input force signal belongs to the tool idle tool movement stage, and the cutting tool is in the idle tool movement stage.

[0065] like Figure 4 As shown, this invention also provides a single-point diamond turning tool positioning system based on a variational autoencoder, employing the aforementioned single-point diamond turning tool positioning method based on a variational autoencoder, comprising:

[0066] Module M1 extracts features from the input force signal and learns its data distribution based on the encoder network. The encoder network in module M1 includes an adaptive function. Module M2 performs batch normalization on the signal output from the encoder network. Module M3 reconstructs the force signal based on the data distribution of the force signal in the batch normalized signal using the decoder network.

[0067] Among them, modules M1, M2 and M3 together form a force signal reconstruction module, which is used to reconstruct the input force signal. Furthermore, the input force signal can be used to sense the force on the cutting tool based on the piezoelectric effect.

[0068] Modular block M4 is used to calculate the reconstruction error between the reconstructed force signal and the input force signal. Specifically, modular block M4 uses the root mean square error to calculate the reconstruction error between the reconstructed force signal and the input force signal.

[0069] The module M5 is used to compare the reconstruction error value with the pre-generated state threshold to determine whether the cutting tool is in the free-running stage or the workpiece contact stage.

[0070] Specifically, module M5 includes:

[0071] Submodule M5.1 is used to construct the state thresholds for the tool idle travel phase. Submodule M5.1 is a state threshold module used to adaptively construct the state thresholds for the tool idle travel phase. It constructs the probability density distribution function of the force signal reconstruction error during the tool idle travel phase based on the kernel density estimation method, and then obtains the cumulative distribution function. Based on the significance level, it obtains the state thresholds belonging to the tool idle travel phase.

[0072] More specifically, submodule M5.1 includes:

[0073] The sub-module M5.11 is used to obtain the force signal reconstruction error value during the tool's idle travel phase.

[0074] The sub-module M5.12 is used to calculate the kernel density function for the force signal reconstruction error value during the tool's idle travel phase.

[0075] The sub-module M5.13 is used to calculate the cumulative distribution function from the results of the kernel density function calculation.

[0076] The sub-module M5.14 is used to obtain the state threshold belonging to the tool idle travel phase based on the significance level.

[0077] Submodule M5.2 is used to determine whether the cutting tool is in the idle tool movement stage or the workpiece contact stage. Submodule M5.2 is a state recognition module used to determine whether the force signal belongs to the tool idle tool movement stage or the tool-workpiece contact stage. After the input force signal passes through the force signal reconstruction module to obtain the reconstruction error value, if this value is greater than the state threshold value obtained by the state threshold module of submodule M5.1, then the input force signal is determined to belong to the tool-workpiece contact stage; otherwise, the input force signal does not belong to the tool-workpiece contact stage.

[0078] It should be further explained that the sub-module M5.1 state threshold module and the sub-module M5.2 state recognition module in the technical solution of this application together constitute the tool state recognition module, which can independently identify and judge the state of the tool.

[0079] like Figure 5 and Figure 6 As shown, the positioning accuracy of the cutting tool can be detected by a white light interferometer, and represented by the depth of the cutting tool into the workpiece. By combining the output of the state recognition module with the tool positioning accuracy detected by the white light interferometer, it is verified that the technical solution of this application has sufficient accuracy in identifying and judging the tool state.

[0080] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0081] In the description of this application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0082] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for positioning a single-point diamond turning tool based on a variational autoencoder, characterized in that, Includes the following steps: Step S1: Extract features from the input force signal and learn the data distribution of the input force signal based on the encoder network; Step S2: Perform batch normalization processing on the signals output by the encoder network; Step S3: Reconstruct the force signal based on the data distribution of the force signal in the batch-normalized signal using the decoder network; Step S4: Calculate the reconstruction error between the reconstructed force signal and the input force signal; Step S5: Compare the reconstruction error value with the pre-generated state threshold to determine whether the cutting tool is in the free-running stage or the workpiece contact stage. Step S5 includes: Step S5.1: Construct the state threshold for the tool's idle travel phase; Step S5.2: Determine whether the cutting tool is in the free-running stage or the workpiece contact stage; Step S5.1 includes: Step S5.11: Obtain the force signal reconstruction error value during the tool idle travel phase; Step S5.12: Calculate the kernel density function for the force signal reconstruction error value during the tool idle travel phase; Step S5.13: Calculate the cumulative distribution function from the kernel density function calculation results; Step S5.14: Obtain the state threshold belonging to the tool idle travel stage based on the significance level.

2. The single-point diamond turning tool positioning method based on variational autoencoder as described in claim 1, characterized in that, For step S1, the encoder network includes an adaptive function.

3. The single-point diamond turning tool positioning method based on variational autoencoder as described in claim 1, characterized in that, For step S4, the reconstruction error between the reconstructed force signal and the input force signal is calculated using the root mean square error.

4. A single-point diamond turning tool positioning system based on a variational autoencoder, characterized in that, The single-point diamond turning tool positioning method based on variational autoencoder as described in any one of claims 1-3 includes: Modular block M1, based on encoder network, performs feature extraction of input force signal and learns the data distribution of input force signal; Modular block M2 performs batch normalization processing on the signals output by the encoder network; Modular M3 reconstructs the force signal based on the data distribution of the force signal in the batch-normalized signal using a decoder network. Modular block M4 calculates the reconstruction error between the reconstructed force signal and the input force signal; M5 module compares the reconstruction error value with the pre-generated state threshold to determine whether the cutting tool is in the empty tool movement stage or the workpiece contact stage. Module M5 includes: Submodule M5.1: Construct the state threshold for the tool's idle travel phase; Submodule M5.2 determines whether the cutting tool is in the idle tool movement stage or the workpiece contact stage; Submodule M5.1 includes: Sub-mold fast M5.11, obtain the force signal reconstruction error value during the tool idle travel phase; The kernel density function is used to calculate the force signal reconstruction error value during the tool idle movement phase of the sub-mold fast M5.

12. Sub-module M5.13: Calculate the cumulative distribution function based on the kernel density function calculation results; Sub-mode fast M5.14: Based on the significance level, obtain the state threshold belonging to the tool idle travel stage.

5. The single-point diamond turning tool positioning system based on a variational autoencoder as described in claim 4, characterized in that, Module M1, the encoder network includes an adaptive function.

6. The single-point diamond turning tool positioning system based on a variational autoencoder as described in claim 4, characterized in that, Module M4 calculates the reconstruction error between the reconstructed force signal and the input force signal using the root mean square error.

Images