Surface roughness monitoring method of embedded physical prediction model

Abstract

The invention provides a surface roughness monitoring method of an embedded physical prediction model. According to the method, in a model training stage, cutting force signals and three-phase current signals of a main shaft servo motor in the machining process are collected through a dynamometer and a current sensor, the signals are preprocessed and then input into a cutting force calculation module, cutting force under the corresponding current signals is output, and the obtained cutting force is used for calculating a surface roughness theoretical value; and the current signal, a current signal and a processing parameter are input into a surface roughness calculation module, and a surface roughness monitoring model is trained. In the machining stage, the roughness value of the machined surface of the workpiece at the moment is calculated in real time by collecting the three-phase current signals of the main shaft servo motor in the machining process, and it is avoided that sensors such as a vibration sensor or an acoustic emission sensor or a dynamometer are installed in a machine tool in the actual machining stage.

Description

Technical Field

[0001] The invention relates to the field of mechanical manufacturing and processing, and in particular to a surface roughness monitoring method with an embedded physical prediction model. Background Art

[0002] Surface roughness has an important influence on the friction coefficient, life, yield limit, corrosion resistance and other properties of parts. In high-end manufacturing fields such as aviation, aerospace, navigation, and automobiles, the service environment of mechanical parts is severe, and the requirements for the surface roughness of these parts are very strict. The traditional surface roughness measurement method usually uses a roughness meter for offline contact measurement. This method easily leaves scratches on the surface of the workpiece and often requires the workpiece to be disassembled, reducing the overall processing efficiency. Secondly, the surface roughness obtained in this way is the surface roughness after processing. If the surface roughness does not meet the process requirements at this time, it is difficult to remedy.

[0003] On the other hand, there are some current explorations of using sensors to monitor machine tool processing signals and then monitor surface roughness. It is often necessary to install vibration sensors, acoustic emission sensors or dynamometers in the machine tool, but these sensors may affect normal processing because they are installed in the processing area. Summary of the invention

[0004] In order to solve the problems existing in the prior art and improve the processing efficiency, the surface roughness monitoring needs of the workpiece during the processing process are met, and the workpiece is prevented from failing due to unqualified surface roughness. The present invention uses the deep learning neural network technology with an embedded physical model to propose a surface roughness monitoring method with an embedded physical prediction model. In the model training stage, the cutting force signal and the three-phase current signal of the spindle servo motor in the processing process are collected by the dynamometer and the current sensor, and the signal is pre-processed and input into the cutting force calculation module, and the cutting force under the corresponding current signal is output. The obtained cutting force is used to calculate the theoretical value of the surface roughness, and it is input into the surface roughness calculation module together with the current signal and the processing parameters to train the surface roughness monitoring model. In the processing stage, by collecting the three-phase current signal of the spindle servo motor during the processing process, the roughness value of the processed surface of the workpiece is calculated in real time, avoiding the installation of sensors such as vibration sensors, acoustic emission sensors or dynamometers in the machine tool during the actual processing stage.

[0005] The technical solution of the present invention is:

[0006] A surface roughness monitoring method with an embedded physical prediction model comprises the following steps:

[0007] Step 1: Collect the three-phase current signal of the spindle servo motor during the CNC machine tool processing of the workpiece surface, perform noise reduction processing on the three-phase current signal according to the main frequency of the three-phase current of the spindle servo motor, and convert the three-phase current signal after noise reduction processing into equivalent DC current;

[0008] Step 2: The equivalent DC current obtained in step 1, as well as the CNC machine tool spindle speed, tool feed speed, cutting width, and cutting depth at the corresponding moment are input into the trained cutting force calculation module to obtain the cutting force;

[0009] Step 3: Based on the cutting force F obtained in step 2, use the formula

[0010]

[0011] The theoretical value of surface roughness R is calculated. a ; where C az is the setting proportional coefficient, r is the cutting edge angle, HB is the Brinell hardness of the workpiece material, σ is the flow stress, E1 is the elastic modulus of the tool material, E2 is the elastic modulus of the workpiece material, v1 is the Poisson's ratio of the tool material, and v2 is the Poisson's ratio of the workpiece material;

[0012] Step 4: Substitute the theoretical value of surface roughness R obtained in step 3 a , and the input of the cutting force calculation module in the corresponding step 2 is used as input and input into the trained surface roughness calculation module to finally obtain the surface roughness value of the machined surface.

[0013] A further preferred scheme, in step 1, the process of performing noise reduction processing on the three-phase current signal according to the main frequency of the three-phase current of the spindle servo motor is: setting the cutoff frequency of the low-pass filter to the main frequency or no more than 5 Hz higher than the main frequency, and using the low-pass filter to filter and reduce noise on the three-phase current signal.

[0014] According to a further preferred embodiment, the process of determining the main frequency of the three-phase current of the spindle servo motor is as follows:

[0015] Step a: Establish a linear relationship between the main frequency of the three-phase current of the spindle servo motor and the spindle speed;

[0016] Step b: According to the current actual spindle speed, the linear relationship obtained in step a is used to determine the current main frequency of the three-phase current of the spindle servo motor.

[0017] In a further preferred embodiment, in step a, the process of establishing a linear relationship between the three-phase current main frequency of the spindle servo motor and the spindle speed is:

[0018] Multiple spindle speeds are set. When the spindle rotates at a set speed, the three-phase current signal of the spindle servo motor is synchronously collected, and the collected three-phase current signal of the spindle servo motor is Fourier transformed, and the frequency corresponding to the highest amplitude within the working frequency range of the spindle servo motor is taken as the main frequency; according to multiple spindle speeds and the corresponding main frequencies, the linear relationship between the three-phase current main frequency of the spindle servo motor and the spindle speed is obtained by linear fitting.

[0019] In a further preferred embodiment, the trained cutting force calculation module and the trained surface roughness calculation module are obtained by training through the following process:

[0020] Step A: Establishing a neural network model for calculating cutting force and a neural network model for calculating surface roughness respectively;

[0021] Step B: Establishing a sample data set; the sample data consists of equivalent DC current, spindle speed, feed speed, cutting width, cutting depth, and cutting force and measured surface roughness under corresponding processing parameters;

[0022] The equivalent DC current, spindle speed, feed rate, cutting width, and cutting depth are used as training inputs of the neural network model for cutting force calculation, and the cutting force measured under the corresponding machining parameters is used as the label corresponding to the training input;

[0023] The equivalent DC current, spindle speed, feed speed, cutting width, cutting depth and theoretical values ​​of surface roughness calculated based on the measured cutting force are used as training inputs of the surface roughness calculation neural network model, and the measured surface roughness under the corresponding processing parameters is used as the label corresponding to the training input;

[0024] Step C: Use the established sample data set to train the cutting force calculation neural network model and the surface roughness calculation neural network model to obtain the trained cutting force calculation module and the surface roughness calculation module.

[0025] A further preferred solution is that when collecting sample data, the cutting force is measured by adding a force sensor to the CNC machine tool. When actually monitoring the surface roughness, the cutting force is not measured by adding a force sensor to the CNC machine tool, but is output by a trained cutting force calculation module.

[0026] A further preferred solution is that when collecting sample data, the collected three-phase current signal of the spindle servo motor and the cutting force signal generated during the knife feeding process are intercepted at set time intervals, the intercepted three-phase current signal of the spindle servo motor is subjected to noise reduction processing and converted into an equivalent DC current, and the average value of the intercepted cutting force signal is calculated as a label for training the cutting force calculation neural network model.

[0027] In a further preferred embodiment, the cutting force is a positive pressure exerted by the tool on the surface of the workpiece being processed.

[0028] Beneficial Effects

[0029] The surface roughness monitoring method with an embedded physical prediction model proposed in the present invention, by establishing a mapping relationship between the cutting force signal and the current signal, enables the acquisition of the cutting force in the processing process to get rid of the expensive and difficult to install dynamometer. The present invention uses the cutting force output by the cutting force calculation module to calculate the theoretical value of the roughness of the processed surface, and inputs it as a characteristic value into the surface roughness calculation neural network, thereby reducing the computational complexity of the neural network, reducing the number of samples required for training the surface roughness calculation neural network, reducing the monitoring cost of the surface roughness, and ensuring the monitoring accuracy.

[0030] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0032] Figure 1 Flowchart of surface roughness monitoring method based on embedded physical model neural network

[0033] Figure 2 Schematic diagram of elastic-plastic deformation in the meshing area between tool and workpiece

[0034] Figure 3 Model of a certain type of box casing cover

[0035] Figure 4 Comparison of original current signal, cutting force signal and filtered current signal

[0036] Figure 5 Cutting force calculation module verification data results

[0037] Figure 6 Surface roughness calculation module verification data results DETAILED DESCRIPTION

[0038] Embodiments of the present invention are described in detail below. The embodiments are exemplary and intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0039] This embodiment takes the end surface processing of a certain type of box casing as an example to achieve real-time monitoring of the surface roughness of the processed surface during the processing of the end surface of the box casing.

[0040] like Figure 1As shown, the method mainly includes two parts, namely, the model training part before real-time monitoring, and the real-time monitoring part of the surface roughness of the machined surface during the actual machining process of the box casing end face after the model training is completed.

[0041] In the model training part, two neural network models are mainly trained, namely the cutting force calculation neural network model and the surface roughness calculation neural network model.

[0042] First, a sample data set is established. The sample data consists of equivalent DC current, spindle speed, feed speed, cutting width, cutting depth, and the cutting force and surface roughness measured under the corresponding processing parameters. The spindle speed, feed speed, cutting width, and cutting depth are directly obtained from the CNC machine tool data; the equivalent DC current is obtained by installing a current sensor on the CNC machine tool, and the three-phase current signal of the spindle servo motor is measured after noise reduction and solution; the cutting force is measured by installing a dynamometer on the CNC machine tool; and the surface roughness is measured on the workpiece processing surface. The specific construction process is:

[0043] During the sample collection phase, a dynamometer and a current sensor are installed on the CNC machine tool. The cutting force signal and the three-phase current signal of the spindle servo motor are collected through the dynamometer and the current sensor. The collected cutting force signal is the positive pressure signal of the tool acting on the machined surface. Taking the milling plane processing of the vertical CNC machine tool as an example, the tool axis direction of the vertical CNC machine tool is consistent with the Z-axis direction of the machine tool coordinate system. When the machined surface is parallel to the XOY plane of the machine tool coordinate system, the positive pressure of the tool acting on the machined surface is the axial cutting force F. z , the positive pressure can be collected by a force gauge.

[0044] The spindle servo motor of a CNC machine tool is generally a three-phase asynchronous AC motor. Its complete current signal consists of three-phase AC currents of u, v and w. The phase difference between the three-phase AC currents is Therefore, the three-phase current signals are collected simultaneously in the electric control box of the machine tool through the current sensor, and the three-phase current signals are subjected to noise reduction processing according to the main frequency of the three-phase current of the spindle servo motor:

[0045] Set the cutoff frequency to be the main frequency or a frequency no higher than 5Hz, and use the Butterworth low-pass filter to filter and reduce noise on the three-phase current signal. The process of determining the main frequency of the three-phase current of the spindle servo motor is as follows:

[0046] Step a: Establish a linear relationship between the main frequency of the three-phase current of the spindle servo motor and the spindle speed;

[0047] Multiple spindle speeds are set. When the spindle rotates at a set speed, the three-phase current signal of the spindle servo motor is synchronously collected, and the collected three-phase current signal of the spindle servo motor is Fourier transformed, and the frequency corresponding to the highest amplitude within the working frequency range of the spindle servo motor is taken as the main frequency; according to multiple spindle speeds and the corresponding main frequencies, the linear relationship between the three-phase current main frequency of the spindle servo motor and the spindle speed is obtained by linear fitting.

[0048] Step b: According to the current actual spindle speed, the linear relationship obtained in step a is used to determine the current main frequency of the three-phase current of the spindle servo motor.

[0049] The collected three-phase current signal and the cutting force signal collected simultaneously are segmented. Since the sampling rate set in this embodiment is 10000 Hz, these data are segmented into groups of 800 data points, and each group of data becomes a sample. The three-phase current signal in the sample is denoised and converted into an equivalent DC current, and the average value of the cutting force signal in the sample is calculated as the sample label for training the neural network model for cutting force calculation. In addition, the surface roughness of the workpiece is measured, and the measured surface roughness is used as the sample label of the surface roughness calculation module.

[0050] Secondly, the established sample data set is used to train the cutting force calculation neural network model and the surface roughness calculation neural network model.

[0051] When training the neural network model for cutting force calculation, the equivalent DC current, spindle speed, feed speed, cutting width, and cutting depth are used as training inputs of the neural network model for cutting force calculation, and the cutting force measured under the corresponding machining parameters is used as the label corresponding to the training input.

[0052] In order to reduce the computational complexity of the surface roughness calculation neural network model and the number of samples required for training the surface roughness calculation neural network model, a surface roughness physical model is embedded in the surface roughness calculation neural network model, that is, the theoretical value of the surface roughness value calculated based on the cutting force is added as the model input value. Therefore, when training the surface roughness calculation neural network model, the equivalent DC current, spindle speed, feed speed, cutting width, cutting depth and the theoretical value of the surface roughness value calculated based on the measured cutting force are used as the training input of the surface roughness calculation neural network model, and the surface roughness measured under the corresponding processing parameters is used as the label corresponding to the training input.

[0053] The theoretical value of the surface roughness calculated based on the cutting force is one of the innovative points of the present invention, and its theoretical analysis process is as follows:

[0054] Establish a theoretical model of surface roughness: The profile height h of the machined surface is determined by the plastic deformation height h1 and the elastic recovery height h2, such as Figure 2 They can be expressed as:

[0055] h=h1-h2

[0056] Among them, h1 can be calculated by the following formula:

[0057]

[0058] Among them, σ is the flow stress; HB is the Brinell hardness of the workpiece material; ψ is the strain; r is the tip angle of the cutting edge.

[0059] The flow stress can be calculated by the following formula:

[0060]

[0061] Among them, A, B, C, m, n are the constitutive parameters of the workpiece material; T is the workpiece temperature; T r is room temperature; T m is the melting temperature of the workpiece; ε0 is the reference plastic strain rate; ε is the equivalent plastic strain rate, and ε1 is the plastic strain rate, which can be calculated using the following formula:

[0062]

[0063] Among them, α is the tool rake angle; φ is the shear angle; V is the cutting speed; K is the proportion of the main cutting area, h p are the cutting zone thicknesses; they can be calculated using the following formula:

[0064] K=0.5+(cos(2φ-α)) / (2cosα)

[0065]

[0066] Where H is the undeformed chip thickness; it can be expressed as:

[0067]

[0068] Where c is the feed per tooth (mm / rev-tooth); R is the tool radius; a e Radial cutting width; is the maximum immersion angle.

[0069] The elastic recovery height h2 can be calculated using Hertz elastic contact theory:

[0070]

[0071] Among them, v1, v2 are the Poisson's ratios of the tool material and the workpiece material; E1, E2 are the elastic moduli of the tool material and the workpiece material; F is the normal pressure of the tool on the workpiece; ρ1 is the curvature of the tool contact position; ρ2 is the curvature of the machined surface.

[0072] Combining the above equations, we can get:

[0073]

[0074] The profile height h can then be expressed as:

[0075]

[0076] According to the surface roughness value R z From the definition of , we can know that the profile height calculated here is the surface roughness value R z , but the arithmetic mean height R is commonly used in engineering a To express the surface roughness value, set R z With R a The proportional coefficient between them is C az , then the arithmetic mean height R a It can be expressed as:

[0077]

[0078] For the cutting force collected or obtained through the cutting force calculation model, the cutting force is substituted into the above formula to obtain the theoretical surface roughness of the machined surface.

[0079] Through the above process, the training of the neural network model for cutting force calculation and the neural network model for surface roughness calculation was completed, and the cutting force calculation module and the surface roughness calculation module were obtained. When the surface roughness of the machined surface is monitored in real time during the actual processing of the end face of the box casing, there is no need to install a dynamometer on the CNC machine tool. It is only necessary to install a current sensor in the electronic control box of the machine tool to collect the three-phase current signal of the spindle servo motor. After processing, the equivalent DC current is obtained, and the spindle speed, feed speed, cutting width and cutting depth are combined as the input of the cutting force calculation module to obtain the average positive pressure of the tool on the workpiece as the cutting force. The cutting force is then used to calculate the theoretical value of the roughness of the machined surface, and it is input into the surface roughness calculation module together with the spindle speed, feed speed, cutting width, cutting depth and equivalent DC current to finally obtain the surface roughness value of the machined surface.

[0080] The surface roughness monitoring system used in this embodiment is: JDMR600 five-axis CNC machining center, FLUKEi200s clamp current sensor, Kistlter 9170A131 rotary dynamometer, Kistler 5238B dynamometer signal amplifier, DEWETron SIRIUSi digital acquisition card. Figure 3 shown. Figure 4 Shown is the comparison between the cutting force signal and the spindle servo motor current signal. Figure 5 Shown are the training and verification results of the neural network model for cutting force calculation, with an accuracy of 96.6%. Figure 6 The figure shows the training and verification results of the surface roughness calculation neural network model, with an accuracy of up to 98.23%.

[0081] Finally, a practical processing example of a certain type of box casing is given. The protruding end face of the box casing and the end face where the hole feature is located are assembly surfaces, and the surface roughness requirement of the former is R a 1.0μm, which is called surface A, and the surface roughness requirement of the latter is R a 0.5μm, which is called surface B. When processing these two end faces, only the current sensor is installed in the machine tool electrical control box to collect the current signal of the spindle servo motor during the processing, and input it into the cutting force calculation module to obtain the real-time cutting force, and then input the cutting force into the surface roughness calculation module to obtain the real-time surface roughness. Finally, the average surface roughness of the end face and the measured surface roughness are monitored as shown in Table 1. The relative errors of the two surface roughness monitoring values ​​are 6.26% and 3.12% respectively. The relative error is small, which can meet the monitoring requirements of surface roughness during the processing.

[0082] Table 1

[0083]

[0084] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and purpose of the present invention.

Claims

1. A surface roughness monitoring method with an embedded physical prediction model, characterized in that: The following steps are involved: Step 1: Collect the three-phase current signal of the spindle servo motor during the CNC machine tool processing of the workpiece surface, perform noise reduction processing on the three-phase current signal according to the main frequency of the three-phase current of the spindle servo motor, and convert the three-phase current signal after noise reduction processing into equivalent DC current; Step 2: The equivalent DC current obtained in step 1, as well as the CNC machine tool spindle speed, tool feed speed, cutting width, and cutting depth at the corresponding moment are input into the trained cutting force calculation module to obtain the cutting force; Step 3: Based on the cutting force F obtained in step 2, use the formula The theoretical value of surface roughness R is calculated. a ; where C az is the setting proportional coefficient, r is the cutting edge angle, HB is the Brinell hardness of the workpiece material, σ is the flow stress, E1 is the elastic modulus of the tool material, E2 is the elastic modulus of the workpiece material, v1 is the Poisson's ratio of the tool material, and v2 is the Poisson's ratio of the workpiece material; Step 4: Substitute the theoretical value of surface roughness R obtained in step 3 a , and the input of the cutting force calculation module in the corresponding step 2 is used as input and input into the trained surface roughness calculation module to finally obtain the surface roughness value of the machined surface.

2. The surface roughness monitoring method with an embedded physical prediction model according to claim 1, characterized in that: In step 1, the process of performing noise reduction processing on the three-phase current signal according to the main frequency of the three-phase current of the spindle servo motor is as follows: setting the cutoff frequency of the low-pass filter to the main frequency or no higher than the main frequency by no more than 5 Hz, and using the low-pass filter to filter and reduce noise on the three-phase current signal.

3. The surface roughness monitoring method according to claim 2, characterized in that: The process of determining the main frequency of the three-phase current of the spindle servo motor is: Step a: Establish a linear relationship between the main frequency of the three-phase current of the spindle servo motor and the spindle speed; Step b: According to the current actual spindle speed, the linear relationship obtained in step a is used to determine the current main frequency of the three-phase current of the spindle servo motor.

4. The surface roughness monitoring method according to claim 3, characterized in that: In step a, the process of establishing the linear relationship between the three-phase current main frequency of the spindle servo motor and the spindle speed is: Multiple spindle speeds are set. When the spindle rotates at a set speed, the three-phase current signal of the spindle servo motor is synchronously collected, and the collected three-phase current signal of the spindle servo motor is Fourier transformed, and the frequency corresponding to the highest amplitude within the working frequency range of the spindle servo motor is taken as the main frequency; according to multiple spindle speeds and the corresponding main frequencies, the linear relationship between the three-phase current main frequency of the spindle servo motor and the spindle speed is obtained by linear fitting.

5. The surface roughness monitoring method according to claim 1, characterized in that: The trained cutting force calculation module and the trained surface roughness calculation module are obtained by training through the following process: Step A: Establishing a neural network model for calculating cutting force and a neural network model for calculating surface roughness respectively; Step B: Establishing a sample data set; the sample data consists of equivalent DC current, spindle speed, feed speed, cutting width, cutting depth, and cutting force and measured surface roughness under corresponding processing parameters; The equivalent DC current, spindle speed, feed rate, cutting width, and cutting depth are used as training inputs of the neural network model for cutting force calculation, and the cutting force measured under the corresponding machining parameters is used as the label corresponding to the training input; The equivalent DC current, spindle speed, feed speed, cutting width, cutting depth and theoretical values ​​of surface roughness calculated based on the measured cutting force are used as training inputs of the surface roughness calculation neural network model, and the measured surface roughness under the corresponding processing parameters is used as the label corresponding to the training input; Step C: Use the established sample data set to train the cutting force calculation neural network model and the surface roughness calculation neural network model to obtain the trained cutting force calculation module and the surface roughness calculation module.

6. The surface roughness monitoring method with an embedded physical prediction model according to claim 1, characterized in that: When collecting sample data, the cutting force is measured by adding a force sensor to the CNC machine tool. When actually monitoring the surface roughness, the cutting force is not measured by adding a force sensor to the CNC machine tool, but is output by a trained cutting force calculation module.

7. The surface roughness monitoring method with an embedded physical prediction model according to claim 1, characterized in that: When collecting sample data, the three-phase current signal of the spindle servo motor and the cutting force signal generated during the cutting feed process are intercepted at the set time interval, the intercepted three-phase current signal of the spindle servo motor is converted into an equivalent DC current after noise reduction processing, and the average value of the intercepted cutting force signal is calculated as the label for training the cutting force calculation neural network model.

8. The surface roughness monitoring method with an embedded physical prediction model according to claim 1, characterized in that: The cutting force is the positive pressure exerted by the tool on the surface of the workpiece being processed.

Images