Cutter wear prediction method of physically guided Gaussian process based on free energy theory

Abstract

The invention discloses a cutter wear prediction method of a Gaussian process based on physical guidance of a free energy theory. The cutter wear prediction method comprises the following steps: 1, establishing a cutting signal-wear-wear rate data set based on collected cutting force signals; 2, establishing an equation among the tool wear, the wear rate and the signal characteristics by using a free energy theory in statistical mechanics; 3, performing data-driven proxy modeling on two position quantities of free energy and temperature of the system through a physically guided Gaussian process; 4, performing derivation on a physically guided Gaussian process to enable the physically guided Gaussian process to be equal to the wear rate, and establishing a second constraint relation; 5, performing Bayesian optimization solution on the model to obtain optimal tool wear; and 6, directly predicting the wear value by using the model. The method has the advantages of being easy to implement, good in robustness and high in reliability, the accuracy and interpretability of tool wear prediction can be improved, and therefore the milling efficiency and the quality of produced parts are improved.

Description

Technical Field

[0001] The invention belongs to the field of tool state monitoring, and in particular is a tool wear prediction method based on a physically guided Gaussian process of free energy theory. Background Art

[0002] Tool wear is an inevitable natural phenomenon in milling, which affects the quality and processing efficiency of parts. Monitoring the tool wear state to accurately estimate the tool wear value and change the tool in time is of great significance to improve the processing quality and efficiency. The most commonly used tool wear state monitoring method is the indirect method, that is, to infer the tool wear value by real-time acquisition of sensor signals. With the introduction of the concept of "intelligent manufacturing", the indirect method combined with artificial intelligence technology is more intelligent and automated. The mapping from processing signals to tool wear values ​​can be directly established through machine learning methods without any knowledge of the underlying knowledge of the milling process. However, most of the existing methods are purely data-driven, overly relying on the machine learning method itself and ignoring the connection between the internal milling processing signals and the connection between the processing signals and tool wear phenomena. On the one hand, this tool wear estimation method based entirely on the "black box" model is difficult to guarantee accuracy in different scenarios and working conditions, and its generalization and robustness are limited. On the other hand, since it does not involve any underlying processing process mechanism, the reliability of such methods is weakened, which limits their application in actual production and processing tasks. Summary of the invention

[0003] The present invention aims to address the deficiencies in the above-mentioned prior art and proposes a tool wear prediction method based on a physically guided Gaussian process of free energy theory, in order to achieve stable and accurate estimation of tool wear values ​​during milling processing, thereby overcoming the problems of poor generalization and robustness and low reliability of traditional methods in predicting tool wear values ​​under different scenarios and working conditions.

[0004] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0005] The tool wear prediction method of the physical guided Gaussian process based on free energy theory of the present invention is characterized in that it comprises the following steps:

[0006] Step 1: Perform milling on the tool and collect the i-th cutting signal sequence of the tool in the X, Y, and Z directions at a fixed frequency , the i-th segment spindle vibration acceleration signal in the X, Y, and Z directions And the i-th segment acoustic emission signal of the tool , while collecting the i-th section machine tool current signal ; The actual tool wear value of the i-th segment of the tool is recorded as , and is calculated using the difference of adjacent wear values Tool wear rate ; Thus forming the milling processing data set ;in, represents the milling processing data of the i-th segment of the tool, and , represents the wear label of the i-th segment of the tool, and , T represents transpose, express The time series; N represents the total number of time series;

[0007] Step 2: Extract the signal characteristics of milling processing based on the theory of statistical mechanics;

[0008] Step 3: Use the Gaussian process surrogate model to establish the tool wear equation:

[0009] Step 3.1: Construct a mean function that conforms to the physical process of tool wear;

[0010] Step 3.2: Construct a kernel function that conforms to the physical process of tool wear;

[0011] Step 3.3: Construct tool wear equation based on signal features, mean function and kernel function;

[0012] Step 3.4: Based on D, the mechanical energy of the tool wear equation is optimized using the Bayesian optimization algorithm to obtain the tool wear equation under the optimal hyperparameters;

[0013] Step 4: Collect the milling processing data of the current period in real time and input it into the tool wear equation under the optimal hyperparameters after processing according to the process of step 2, and use the fourth-order Runge-Kutta method to integrate the equation to obtain the predicted tool wear value in the current period.

[0014] The tool wear prediction method of the physical guided Gaussian process based on free energy theory described in the present invention is also characterized in that the step 3 includes the following steps:

[0015] Step 2.1: After normalization, the dimensionless milling data of the i-th segment is obtained: , and use the variational mode decomposition method to Decomposed into n intrinsic mode function components, we get The eigenmode function component matrix of the cutting signal with rows and k columns is ;

[0016] Step 2.2: The importance weight vector ,in, The number of rows and The number of rows is equal, The number of columns is 1;

[0017] Step 2.3: Calculate using formula (1) Enthalpy , thus obtaining the enthalpy vector :

[0018] (1)

[0019] In formula (1), t represents The index of the cutting signal sequence of each row at a discrete time; express The cutting signal sequence of the tth column in ;

[0020] Step 2.4: Calculate using formula (2) The energy probability corresponding to the jth row in :

[0021] (2)

[0022] In formula (2), Indicates the set constant. express The cutting signal of the jth row and the tth column in ; express The j-th importance weight value in ;

[0023] Step 2.5: Calculate using formula (3) Entropy :

[0024] (3).

[0025] Furthermore, the step 3.1 includes the following steps:

[0026] Step 3.1.1: Use Fourier series to Fitting is performed to obtain the fitted wear rate curve , thereby setting the wear rate mean function ;

[0027] Step 3.1.2: Use Fourier series to Perform fitting to obtain the fitted wear value curve , thereby setting the wear value mean function .

[0028] Further, the step 3.2 includes the following steps:

[0029] Step 3.2.1: Use equation (4) to construct a kernel function that conforms to the tool wear rate variation process :

[0030] (4)

[0031] In formula (4), Indicates the tool Segment milling processing data The corresponding tool wear rate, express The corresponding timing, , , , , There are 5 hyperparameters respectively;

[0032] Step 3.2.2: Use equation (7) to construct a kernel function that conforms to the physical process of tool wear :

[0033] (7)

[0034] In formula (7), express The corresponding free energy, , , , There are 4 hyperparameters respectively.

[0035] Further, the step 3.3 includes the following steps:

[0036] Step 3.3.1: Use equation (9) to construct the tool wear equation:

[0037] (9)

[0038] In formula (9), Represents a wear value mean function , kernel function Gaussian process; Represents the wear rate mean function , kernel function Gaussian process;

[0039] Step 3.3.2: Use equation (10) to establish the constraint equation of tool wear equation:

[0040] (10)

[0041] In formula (10), Represents the derivative operator.

[0042] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the tool wear prediction method, and the processor is configured to execute the program stored in the memory.

[0043] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the tool wear prediction method when the computer program is executed by a processor.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention has the advantages of easy extraction of data features and clear physical meaning of features. It uses free energy theory to extract tool feature enthalpy and entropy. Compared with the traditional machine learning module automatic feature extraction method, the features extracted by the method of the present invention are more relevant to the tool wear process and more robust.

[0046] 2. The present invention treats the cutting signal and the actual tool wear value as a dynamic system with free energy G and temperature T, and constructs the tool wear equation accordingly. Compared with the traditional pure data-driven tool wear prediction method, this processing method has a more underlying logic, making the tool wear value estimation result more reliable and more interpretable, and can solve the problem that the traditional pure data-driven tool wear prediction method cannot explain.

[0047] 3. The present invention embeds the known prior knowledge of the milling process into the data-driven method, providing a mean function and a kernel function for the Gaussian process that conform to the physical process of tool wear. Compared with the traditional purely data-driven tool wear value prediction method, the learning efficiency is higher and the training and prediction process is more stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the overall framework of the present invention;

[0049] Figure 2 It is a schematic diagram of the installation of the sensor equipment in the implementation process of the present invention;

[0050] Figure 3 A schematic diagram of the milling process system constructed for the present invention;

[0051] Figure 4 Schematic diagram of comparison between the predicted value and the actual value of tool wear in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described in detail below in conjunction with the accompanying drawings. The specific embodiments described herein are only for explanation and are not intended to limit the present invention.

[0053] In this embodiment, a tool wear prediction method based on a physical guided Gaussian process of free energy theory is implemented by combining free energy theory with a data driven method to achieve a tool wear value prediction with high robustness and reliability. Figure 1 The overall framework diagram of the present invention is shown in the following steps:

[0054] Step 1: The sensor device of this embodiment is installed as follows: Figure 2 As shown, the tool is milled and the i-th cutting signal sequence of the tool in the X, Y, and Z directions is collected at a fixed frequency. , the i-th segment spindle vibration acceleration signal in the X, Y, and Z directions And the i-th segment acoustic emission signal of the tool , while collecting the i-th section machine tool current signal After each cutting section is completed, an optical microscope is used to obtain an image of the tool cutting edge and calculate the wear value. The actual tool wear value of the i-th section of the tool is recorded as , and is calculated using the difference of adjacent wear values Tool wear rate ; Thus forming the milling processing data set ;in, represents the milling processing data of the i-th segment of the tool, and , represents the wear label of the i-th segment of the tool, and , T represents transpose, express The timing sequence; N represents the total number of timing sequences; In this embodiment, the machine tool used is a high-speed CNC milling machine DX-650, the data acquisition card uses Dewesoft SIRIUSi-HS-8xACC, the sampling frequency is 50kHz, and the time length of each cutting signal collected is 5 seconds. The sensors used are Kistler 9129AA dynamometer, Kistler 8763B three-way vibration sensor, and Senderg SS430 acoustic emission sensor. The tool used is a Sanco four-tooth carbide end mill with a helix angle of 48° and a diameter of 6mm. The workpiece material to be milled is aluminum alloy 7030.

[0055] Step 2: Extract the signal characteristics of milling processing based on the theory of statistical mechanics:

[0056] like Figure 3As shown in the figure, based on the free energy theory, the milling tool and workpiece are constructed as a system, and the free energy of the system is equal to the enthalpy of the system minus the product of the entropy of the system and the temperature. Among them, the tool wear value is defined as the actual state of the system and is related to the free energy, and the tool wear rate is related to the system temperature; the sensor signal is defined as the external representation of the actual state of the system, and the entropy and enthalpy of the milling signal are extracted as features.

[0057] Step 2.1: After normalization, the dimensionless milling data of the i-th segment is obtained: , and use the variational mode decomposition method to Decomposed into n intrinsic mode function components, we get The eigenmode function component matrix of the cutting signal with rows and k columns is In this embodiment, the matrix is decomposed into 60 components.

[0058] Step 2.2: The importance weight vector ,in, The number of rows and The number of rows is equal, The number of columns is 1;.

[0059] Step 2.3: Calculate using formula (1) Enthalpy , thus obtaining the enthalpy vector :

[0060] (1)

[0061] In formula (1), t represents The index of the cutting signal sequence of each row at a discrete time; express The cutting signal sequence of the tth column in ;

[0062] Step 2.4: Calculate using formula (2) The energy probability corresponding to the jth row in :

[0063] (2)

[0064] In formula (2), Indicates the set constant. express The cutting signal of the jth row and the tth column in ; express The jth importance weight value in ; in this embodiment, the setting , the importance weight vector is set to a periodic exponential decay form: .

[0065] Step 2.5: Calculate using formula (3) Entropy :

[0066] (3)

[0067] Step 3: Use the Gaussian process surrogate model to establish the tool wear equation:

[0068] Step 3.1: Construct a mean function that conforms to the physical process of tool wear:

[0069] Step 3.1.1: Use Fourier series to Fitting is performed to obtain the fitted wear rate curve , thereby setting the wear rate mean function ;

[0070] Step 3.1.2: Use Fourier series to Perform fitting to obtain the fitted wear value curve , thereby setting the wear value mean function ;

[0071] Step 3.2: Construct a kernel function that conforms to the physical process of tool wear:

[0072] Step 3.2.1: Use equation (4) to construct a kernel function that conforms to the tool wear rate variation process :

[0073] (4)

[0074] In formula (4), Indicates the tool Segment milling processing data The corresponding tool wear rate, express The corresponding timing, , , , , There are 5 hyperparameters respectively;

[0075] Step 3.2.2: Use equation (7) to construct a kernel function that conforms to the physical process of tool wear :

[0076] (7)

[0077] In formula (7), express The corresponding free energy, , , , There are 4 hyperparameters respectively.

[0078] Step 3.3: Based on the fact that the free energy of the system is equal to the product of enthalpy minus entropy and temperature, the tool wear equation is constructed using equation (9):

[0079] (9)

[0080] In formula (9), Represents a wear value mean function , kernel function Gaussian process; Represents the wear rate mean function , kernel function Gaussian process;

[0081] Step 3.4: Use equation (10) to establish the constraint equation of tool wear equation:

[0082] (10)

[0083] In formula (10), represents a derivative operator; in this embodiment, the derivative is specifically to use automatic gradient to obtain the model output mean The gradient of , this calculation can also be replaced by theoretical derivation, through The mixed covariance function between and its first-order partial derivative is obtained.

[0084] Step 3.5: Based on D, use the Bayesian optimization algorithm to update the hyperparameters in equation (9) under equation (10), so as to obtain the tool wear equation under the optimal hyperparameters; this embodiment uses the Bayesian Optimization object provided by Python to create an instance and solve it.

[0085] Step 4: Collect the milling processing data of the current period in real time and input it into the tool wear equation under the optimal hyperparameters after processing according to the process of step 2. Use the 4th-order Runge-Kutta method to integrate the equation. The tool wear value in the i+1th milling period is ,in, , , , , f represents the function represented by tool wear equation (9). h represents the time period, which is 5 seconds in this embodiment. Thus, the predicted tool wear value in the current time period is obtained. The visualization result of the comparison between the wear prediction value and the actual value in this embodiment is as follows Figure 4 As shown;

[0086] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0087] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

[0088] In summary, the prediction results of the cases of the present invention prove that the method combining free energy theory and Gaussian process proxy model can accurately and stably predict tool wear values. This method can improve the accuracy of traditional tool wear prediction methods, further ensure the safe progress of the machining process, and help improve the quality and machining efficiency of parts; the data used in the present invention is easy to obtain, easy to process, and easy to extract features, which alleviates the need for traditional complex manual extraction of time, frequency, and time-frequency domain features, and is more physically meaningful than the features automatically extracted by machine learning, and enhances the reliability of the features. The present invention combines the free energy theory that conforms to the physical process of tool wear with the Gaussian process proxy model, enhances the interpretability of the tool wear prediction method, and provides machine tool users with a more reliable way to predict tool wear values.

Claims

1. A tool wear prediction method based on a physically guided Gaussian process of free energy theory, characterized in that: The following steps are involved: Step 1: Perform milling on the tool and collect the i-th cutting signal sequence of the tool in the X, Y, and Z directions at a fixed frequency , the i-th segment spindle vibration acceleration signal in the X, Y, and Z directions And the i-th segment acoustic emission signal of the tool , while collecting the i-th section machine tool current signal ; The actual tool wear value of the i-th segment of the tool is recorded as , and is calculated using the difference of adjacent wear values Tool wear rate ; Thus forming the milling processing data set ;in, represents the milling processing data of the i-th segment of the tool, and , represents the wear label of the i-th segment of the tool, and , T represents transpose, express The time series; N represents the total number of time series; Step 2: Extract the signal characteristics of milling processing based on the theory of statistical mechanics; Step 3: Use the Gaussian process surrogate model to establish the tool wear equation: Step 3.1: Construct a mean function that conforms to the physical process of tool wear; Step 3.2: Construct a kernel function that conforms to the physical process of tool wear; Step 3.3: Construct tool wear equation based on signal features, mean function and kernel function; Step 3.4: Based on D, the mechanical energy of the tool wear equation is optimized using the Bayesian optimization algorithm to obtain the tool wear equation under the optimal hyperparameters; Step 4: Collect the milling processing data of the current period in real time and input it into the tool wear equation under the optimal hyperparameters after processing according to the process of step 2, and use the fourth-order Runge-Kutta method to integrate the equation to obtain the predicted tool wear value in the current period.

2. The tool wear prediction method based on the physical guided Gaussian process of free energy theory according to claim 1 is characterized in that: The step 3 comprises the following steps: Step 2.1: After normalization, the dimensionless milling data of the i-th segment is obtained: , and use the variational mode decomposition method to Decomposed into n intrinsic mode function components, we get The eigenmode function component matrix of the cutting signal with rows and k columns is ; Step 2.2: The importance weight vector ,in, The number of rows and The number of rows is equal, The number of columns is 1; Step 2.3: Calculate using formula (1) Enthalpy , thus obtaining the enthalpy vector : (1) In formula (1), t represents The index of the cutting signal sequence of each row at a discrete time; express The cutting signal sequence of the tth column in ; Step 2.4: Use formula (2) to calculate The energy probability corresponding to the jth row in : (2) In formula (2), Indicates the set constant. express The cutting signal of the j-th row and the t-th column in ; express The j-th importance weight value in ; Step 2.5: Calculate using formula (3) Entropy : (3)。 3. The tool wear prediction method based on the physical guided Gaussian process of free energy theory according to claim 2 is characterized in that: The step 3.1 comprises the following steps: Step 3.1.1: Use Fourier series to Fitting is performed to obtain the fitted wear rate curve , thereby setting the wear rate mean function ; Step 3.1.2: Use Fourier series to Perform fitting to obtain the fitted wear value curve , thereby setting the wear value mean function .

4. The tool wear prediction method based on the physical guided Gaussian process of free energy theory according to claim 3 is characterized in that: The step 3.2 comprises the following steps: Step 3.2.1: Use equation (4) to construct a kernel function that conforms to the tool wear rate variation process : (4) In formula (4), Indicates the tool Segment milling processing data The corresponding tool wear rate, express The corresponding timing, , , , , There are 5 hyperparameters respectively; Step 3.2.2: Use equation (7) to construct a kernel function that conforms to the physical process of tool wear : (7) In formula (7), express The corresponding free energy, , , , There are 4 hyperparameters respectively.

5. The tool wear prediction method based on the physical guided Gaussian process of free energy theory according to claim 4 is characterized in that: The step 3.3 comprises the following steps: Step 3.3.1: Use equation (9) to construct the tool wear equation: (9) In formula (9), Represents a wear value mean function , kernel function Gaussian process; Represents the wear rate mean function , kernel function Gaussian process; Step 3.3.2: Use equation (10) to establish the constraint equation of tool wear equation: (10) In formula (10), Represents the derivative operator.

6. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the tool wear prediction method according to any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the tool wear prediction method according to any one of claims 1 to 5 are executed.

Images