Method and system for controlling feed rate in numerical control machining based on data analysis

Abstract

The invention relates to the technical field of speed control, in particular to a feed rate control method and system in numerical control machining based on data analysis, and the method comprises the steps: obtaining a vibration signal and a current signal in a machining process; calculating the cutting impact strength at each moment, wherein the cutting impact strength is positively correlated with the kurtosis of the amplitude of the vibration signal in the window corresponding to the moment and the logarithm of the absolute mean value of the current signal; and the instability index of each moment is calculated. According to the method, the penalty factor and the kernel function parameter can be adjusted in real time according to the instability of the machining process, high fitting precision is kept when the signal is stable, overfitting is prevented by reducing the penalty factor in the stage of severe abrasion and large noise, and therefore accurate prediction of the surface roughness is achieved, closed-loop correction is conducted on the feeding rate, and the precision of the surface roughness is improved. And the processing efficiency and the surface quality are effectively balanced.

Description

Technical Field

[0001] This invention relates to the field of speed control technology. More specifically, this invention relates to a method and system for feed rate control in CNC machining based on data analysis. Background Technology

[0002] In long-duration CNC machining processes such as mold manufacturing, tool wear is an unavoidable physical phenomenon. As machining time progresses, the tool condition goes through three stages: initial wear, stable wear, and severe wear. Tool wear leads to dulling of the cutting edge, which in turn causes increased cutting force, higher cutting temperature, and intensified system vibration, ultimately resulting in a deterioration of the workpiece surface roughness and even causing the part to be scrapped.

[0003] In existing technologies, constant feed rate or experience-based stepped feed rate control is commonly used, which is insufficient to handle complex time-varying wear processes. To address this issue, predictive control methods based on Support Vector Regression (SVR) have been introduced to attempt to establish a mapping relationship between sensor signals and surface roughness. However, most existing SVR methods determine a fixed set of hyperparameters, such as penalty factors and kernel function parameters, through cross-validation in an offline phase. This static parameter setting has significant limitations: fixed hyperparameters cannot simultaneously adapt to different wear stages throughout the tool's lifecycle. During the stable wear phase, the signal is relatively stable, and the model requires high fitting accuracy; however, during the severe wear phase, the signal contains a large amount of nonlinear noise and abrupt changes. A fixed and high penalty factor can lead to overfitting noise in the model, which not only fails to accurately predict the changing trend of surface quality but also causes improper feed rate adjustment due to incorrect predictions. Summary of the Invention This invention provides a feed rate control method and system for CNC machining based on data analysis, aiming to solve the problem in related technologies that fixed hyperparameters cannot simultaneously adapt to different wear stages throughout the tool's entire life cycle.

[0004] In a first aspect, the present invention provides a feed rate control method and system for CNC machining based on data analysis, comprising: acquiring vibration signals and current signals during the machining process; calculating the cutting impact intensity at each moment, wherein the cutting impact intensity is positively correlated with the kurtosis of the vibration signal amplitude within the corresponding window at that moment and the logarithm of the absolute mean of the current signal; calculating the instability index at each moment, wherein the instability index is positively correlated with the standard deviation of the cutting impact intensity within the corresponding window at that moment and the difference between the current cutting impact intensity and the initial cutting impact intensity; inputting the cutting impact intensity at the current moment into an SVR model to obtain a predicted surface roughness value for the next moment; updating the hyperparameters of the SVR model according to the instability index, including a penalty factor parameter inversely proportional to the instability index and a kernel function parameter directly proportional to the instability index; correcting the baseline feed rate based on the comparison result of the predicted surface roughness value for the next moment and a warning threshold to obtain a corrected feed rate; and sending the corrected feed rate to the CNC system controller. It can dynamically update the penalty factor and kernel parameters of the SVR model based on the instability index, effectively solving the overfitting problem caused by noise during the severe wear period, and realizing accurate prediction of surface roughness and adaptive optimization control of feed rate throughout the tool's entire life cycle.

[0005] Furthermore, the formula for calculating cutting impact strength is: In the formula, This represents the cutting impact intensity within the window corresponding to time t. This indicates the first vibration signal in the vibration signal window sequence corresponding to time t. The amplitude of each sampling point This represents the mean of the vibration signal sequence. This indicates the number of sampling points in the vibration signal window sequence; This indicates the current signal window sequence at time t. The amplitude of each sampling point This represents the number of sampling points in the current signal window sequence. Compared to single signal features, this calculation method utilizes both vibration kurtosis to reflect the instantaneous impact and sharpness during the cutting process, and current signals to reflect the magnitude of the macroscopic cutting load. It can more comprehensively and sensitively characterize the degree of energy accumulation caused by cutting edge dulling and viscous-slip phenomena, providing a robust physical characteristic basis for subsequent instability analysis.

[0006] Furthermore, the window corresponding to time t is a window consisting of time t and the previous N times. By limiting the time window to the current time and the previous N times, this technical solution introduces a sliding window mechanism. This allows the system to consider not only instantaneous values ​​when calculating features, but also to capture the signal fluctuation trend over a short period of time.

[0007] Furthermore, the corrected feed rate is obtained, including: In the formula, For a moment Corrected feed rate; The set baseline feed rate; This is the predicted roughness value of the machined surface at the current moment; This is the warning threshold for surface roughness; Allowable roughness tolerance bandwidth; This is the maximum reduction ratio factor; It is a hyperbolic tangent function. Compared with simple linear adjustment or threshold switching control, the tanh function has nonlinear saturation characteristics. When the predicted surface roughness value exceeds the warning threshold, it can smoothly reduce the feed rate according to the magnitude of the deviation. This avoids system oscillation caused by excessive adjustment and effectively limits roughness deterioration, thus achieving flexible speed planning.

[0008] Furthermore, the updated penalty factor parameter is calculated using the following formula: In the formula, The penalty factor parameters are the updated SVR model parameters at time t. Let t be the instability index of the processing within the window corresponding to time t. This represents the gain coefficient corresponding to the penalty factor. When the processing enters a stage of severe wear (increased instability index), the penalty factor is automatically reduced, thereby increasing the model's tolerance to training errors. This makes the model more inclined to learn the main wear trend rather than forcibly fitting nonlinear noise, significantly improving the model's generalization ability under complex working conditions and avoiding erroneous predictions caused by overfitting noise.

[0009] Furthermore, the method for obtaining the initial cutting impact intensity includes: obtaining an initial cutting interval of a set length after the new tool enters the workpiece; calculating multiple cutting impact intensities within the initial cutting interval to obtain an initial cutting impact intensity sequence; and using the average of the initial cutting impact intensity sequence as the initial cutting impact intensity. This method can eliminate systematic deviations caused by fluctuations in the hardness of different machine tools and workpiece materials. Comparing subsequent cutting impact intensities with this initial value allows the instability index to purely reflect the state evolution caused by tool wear, improving the universality and accuracy of the monitoring index under different machining scenarios.

[0010] Furthermore, the method for obtaining the gain coefficient corresponding to the penalty factor includes: calculating the standard deviation of the initial cutting impact intensity sequence as the initial volatility, wherein the gain coefficient corresponding to the penalty factor is negatively correlated with the initial volatility. For machining systems with large initial vibrations, such as machine tools with poor rigidity, the system automatically reduces the adjustment sensitivity to prevent misjudgment; while for stable systems, the sensitivity is increased. This design allows the control algorithm to adapt to different machining environment background noise levels without tedious manual debugging.

[0011] Furthermore, after acquiring the vibration signal and current signal, the method further includes performing wavelet threshold denoising processing on the vibration signal and current signal. Compared to traditional low-pass filtering, which may cause the loss of effective impact signal, wavelet denoising ensures the data purity of the input SVR model, further improving the signal-to-noise ratio and accuracy of surface roughness prediction.

[0012] Furthermore, the sampling frequency range for acquiring the vibration signal and current signal is 10kHz~20kHz.

[0013] In a second aspect, a feed rate control system for CNC machining based on data analysis is also provided, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the feed rate control method for CNC machining based on data analysis described in any of the above embodiments.

[0014] Beneficial effects: It can adjust the penalty factor and kernel function parameters in real time according to the instability of the processing. It maintains high fitting accuracy when the signal is stable and prevents overfitting by reducing the penalty factor in the stage of severe wear and high noise. This enables accurate prediction of surface roughness and closed-loop correction of the feed rate, effectively balancing processing efficiency and surface quality. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the feed rate control flowchart according to an embodiment of the present invention. Detailed Implementation

[0016] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] like Figure 1 As shown, S101: Multi-source signal acquisition and preprocessing.

[0018] In one embodiment, a CNC machining signal acquisition environment is first established. Specifically, a vibration acceleration sensor is installed on the spindle side of the CNC machine tool to collect the mechanical vibration of the spindle during machining; simultaneously, the spindle load current signal is synchronously acquired through the PLC port of the CNC system. To ensure signal synchronization and effectiveness, the sampling frequency is set to [value missing]. For example, in a preferred embodiment, Can be set to To capture high-frequency impacts, the time window length is determined as follows: Vibration signal sequence and current signal sequence within the time window.

[0019] Considering the complex environment of CNC machining sites, the original signals often contain high-frequency electromagnetic interference. Therefore, wavelet thresholding denoising is required for the acquired original signals. Specifically, a suitable wavelet basis function (such as the Daubechies wavelet system) is selected to decompose the signal into multiple scales, and soft or hard thresholding is applied to the high-frequency detail coefficients to reconstruct the signal and filter out noise, thereby obtaining a clean machining status signal.

[0020] S102: Extract time-frequency domain characteristic indicators that reflect the cutting state.

[0021] In one embodiment, cutting impact intensity is extracted from the denoised signal, taking into account the physical fact that tool wear alters the signal energy distribution. Considering that increased tool wear leads to disordered vibrations during the cutting process, causing deviations in the signal amplitude distribution, and that blunting of the cutting edge increases cutting resistance and consequently raises the current, this embodiment constructs the cutting impact intensity.

[0022] The formula for calculating cutting impact strength is as follows: In the formula, This represents the cutting impact intensity within the window corresponding to time t. This indicates the first vibration signal in the vibration signal window sequence corresponding to time t. The amplitude of each sampling point This represents the mean of the vibration signal sequence. This indicates the number of sampling points in the vibration signal window sequence; This indicates the current signal window sequence at time t. The amplitude of each sampling point This represents the number of sampling points in the current signal window sequence. The window corresponding to time t refers to the window consisting of time t and the preceding N times; in this embodiment, N is 10.

[0023] As shown in the formula above, the first part calculates the kurtosis of the vibration signal, reflecting the sharpness of the impact component in the signal; the second part is the logarithmic term of the absolute mean of the current signal, reflecting the macroscopic magnitude of the cutting load. When the tool wears, the cutting edge becomes dull, leading to increased cutting resistance (increased current term), and the cutting process is accompanied by unstable viscous-slip phenomena (kurtosis term fluctuations). The combined effect of these two factors leads to… The numerical value increases significantly, thus characterizing the degree of concentration of cutting impact energy in the current physical scenario.

[0024] S103: Construct an instability index for the entire lifecycle of the processing.

[0025] In one embodiment, based on the cutting impact strength obtained from the above steps, its evolution over time is further analyzed to construct a machining process instability index, which is used to quantify the characteristics of the current wear stage. Since the characteristics at a single moment may be affected by random factors such as hard spots in the workpiece material, it is necessary to consider historical trends for judgment.

[0026] The formula for calculating the instability index is as follows: In the formula, Let t be the instability index of the processing within the window corresponding to time t. The window corresponding to time t; For a moment The cutting impact strength; From the start of processing to the current moment The average value of all cutting impact strengths; The initial cutting impact strength when the new tool enters the cutting area; The preset sensitivity coefficient has a preferred value range. .

[0027] It should be noted that the initial cutting impact strength when a new tool enters the market... The method for obtaining the value is as follows: The initial moment is determined by identifying the moment when the tool has contacted the workpiece and started cutting; the length starting from this initial moment is then used to obtain the value. The time series is set, and a window is set to slide on the time series, with a window length of [length missing]. Step size is Each time the slide is made, the initial cutting impact intensity within the sliding window is calculated, thus obtaining the initial cutting impact intensity sequence. Then, the average value of all initial cutting impact intensities in the initial cutting impact intensity sequence is used as the initial cutting impact intensity when the new tool enters.

[0028] The first term of the formula calculates the standard deviation of the cutting impact intensity fluctuation within a short time window, reflecting the instantaneous dynamic instability of the cutting process. The second term is an exponential growth term, describing the cumulative deviation of the current cutting impact intensity from the initial cutting impact intensity. As machining time increases, if the tool enters a period of rapid wear, not only the cutting impact intensity... It will be much greater than the initial cutting impact strength. Furthermore, due to the deterioration of the cutting interface, the fluctuation of cutting impact strength will also intensify. Therefore, when the independent variable... When its volatility increases, the dependent variable The nonlinear increase indicates that the machining scenario is evolving from stable cutting to a non-stationary stage of severe wear.

[0029] S104: SVR hyperparameter adaptive mapping based on instability index.

[0030] In one embodiment, to address the problem that the fixed hyperparameters of the existing support vector regression (SVR) model cannot adapt to changes in the wear stage, a dynamic mapping mechanism is constructed using the processing instability index obtained in the above steps to calculate the penalty factor and kernel function parameters of the SVR model in real time.

[0031] The constructed hyperparameter adaptive tuning model is as follows: ; In the formula, The penalty factor parameters are the updated SVR model parameters at time t. The kernel function parameters of the SVR model after update at time t. Let t be the instability index of the processing within the window corresponding to time t. and These are the benchmark penalty factor parameter and the benchmark kernel function parameter, respectively. In this embodiment, the benchmark penalty factor parameter is set to 100, and the benchmark kernel function parameter is set to 0.1. This is the gain coefficient corresponding to the penalty factor. This is the gain coefficient corresponding to the kernel function, used to control the adjustment range.

[0032] For penalty factors : The value represents the model's tolerance for training error. The larger the value, the lower the tolerance. When When the wear increases (i.e., tool wear becomes severe), the nonlinear noise and random interference in the signal increase significantly. If a high level is maintained at this point... The value, the model will force a fit to the noise, leading to overfitting. Through the above inverse proportional relationship, when instability As it rises, Automatic reduction increases the model's tolerance to errors, thereby extracting the main trend of wear rather than fitting instantaneous noise and enhancing the model's generalization ability.

[0033] For kernel parameters : The value controls the range of influence of the radial basis function (RBF). The larger the value, the smaller the scope of influence; the more complex the model. When As the surface roughness increases, the mapping relationship between surface roughness and signal characteristics becomes more complex and highly nonlinear. Using the aforementioned proportional power function logic, as instability increases, Increased size allows the model to capture more localized and complex feature changes, preventing underfitting during periods of severe wear.

[0034] Following on from the above, in the formula and The method for obtaining it is as follows: calculate the standard deviation of the initial cutting impact strength sequence, and use it as the initial volatility. This value characterizes the inherent noise level of the current cutting system under healthy conditions. The calculation formula is as follows: ; In the formula, and All are tolerance multiples. In this embodiment, The preferred value is , The preferred value is ; This is the initial volatility. This setting gives the system adaptive capability: if the initial processing is relatively stable ( If it is smaller, then the calculated and Larger vibrations mean the system is more sensitive to wear-induced changes in characteristics; if the initial machining process involves significant vibrations ( (larger), then and It automatically reduces vibrations, preventing the system from misinterpreting normal cutting vibrations as wear characteristics, thus achieving parameter self-tuning without manual intervention.

[0035] S105: Online prediction of surface roughness and closed-loop control of feed rate.

[0036] In one embodiment, the SVR regression model is reconstructed using updated penalty factor parameters and kernel function parameters. Specifically, the current cutting impact intensity is input as an input vector into the SVR model, and the predicted surface roughness value at the current moment is output.

[0037] Based on the prediction results, a feed rate correction strategy is constructed, and the feed rate command for the next time step is calculated. The calculation formula is as follows: In the formula, For a moment Corrected feed rate; The set baseline feed rate; This is the predicted roughness value of the machined surface at the current moment; This is the warning threshold for surface roughness; The allowable roughness tolerance bandwidth is 0.2; For example, the maximum deceleration ratio factor. The value is ; It is the hyperbolic tangent function.

[0038] From the above formula, it can be seen that when the predicted surface roughness... Below the warning threshold hour, Function output The feed rate remains at the baseline feed rate. To ensure processing efficiency. Once If the threshold is exceeded, the difference increases. The function output value is in the range The increase in internal nonlinearity causes the feed rate to decrease proportionally from the baseline feed rate. Reducing the feed rate can decrease the amount of material removed per tooth and the cutting force, thereby suppressing further deterioration of surface quality under tool wear conditions.

[0039] Finally, the calculated feed rate for the next moment is sent in real time to the CNC controller of the CNC machine tool via a communication interface, such as the FOCAS library, OPCUA protocol, or Profibus / EtherCAT fieldbus.

[0040] The present invention also provides a feed rate control system for CNC machining based on data analysis. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the feed rate control method for CNC machining based on data analysis according to the first aspect of the present invention.

[0041] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0042] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0043] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for feed rate control in data analysis based numerical control machining, characterized by, The method comprises: acquiring vibration signals and current signals in a machining process; calculating a cutting impact intensity at each time point, the cutting impact intensity being positively correlated with kurtosis of vibration signal amplitudes in a window corresponding to the time point and a logarithm of an absolute mean value of the current signal; calculating an instability index at each time point, the instability index being positively correlated with a standard deviation of the cutting impact intensity in the window corresponding to the time point and a difference between a current cutting impact intensity and an initial cutting impact intensity; inputting the current cutting impact intensity into an SVR model to obtain a surface roughness prediction value at a next time point; super parameters of the SVR model are updated according to the instability index, including a penalty factor parameter being inversely proportional to the instability index and a kernel function parameter being proportional to the instability index; correcting a reference feed rate based on a comparison result of the surface roughness prediction value at the next time point and a warning threshold to obtain a corrected feed rate; and sending the corrected feed rate to a numerical control system controller.

2. The feed rate control method in data analysis based NC machining according to claim 1, characterized by, The calculation formula of the cutting impact intensity is: ； In the formula, represents the cutting impact intensity in the window corresponding to time t, represents the amplitude of the sampling point No. in the vibration signal window sequence corresponding to time t, represents the mean value of the vibration signal sequence, represents the number of sampling points of the vibration signal window sequence; represents the amplitude of the sampling point No. in the current signal window sequence corresponding to time t, represents the number of sampling points of the current signal window sequence.

3. The feed rate control method in data analysis-based NC machining according to claim 1 or 2, characterized by, The window corresponding to the time point t is a window composed of the time point t and N previous time points.

4. The feed rate control method in data analysis based NC machining according to claim 1, characterized by, The corrected feed rate is obtained by: ； wherein, is the modified feed rate at the time t; is the modified feed rate at the time t; is the set reference feed rate; is the roughness prediction value of the machined surface at the current time t; is the warning threshold of the surface roughness; is the allowable roughness tolerance bandwidth; is the maximum deceleration scale factor; is the hyperbolic tangent function.

5. The method for feed rate control in data analysis based NC machining according to claim 1, wherein, The updated penalty factor parameter has a calculation formula: ； In the formula, is the updated penalty factor parameter of the SVR model at time t, is the process instability index corresponding to the window at time t, is the gain coefficient corresponding to the penalty factor.

6. The data analysis based feed rate control method in NC machining according to claim 5, wherein, The initial cutting impact intensity is obtained by: acquiring an initial cutting interval of a set length after a new tool cuts into a workpiece; calculating a plurality of cutting impact intensities in the initial cutting interval to obtain an initial cutting impact intensity sequence; and taking a mean value of the initial cutting impact intensity sequence as the initial cutting impact intensity.

7. The feed rate control method in data analysis based NC machining according to claim 6, characterized in that, The method for obtaining a gain coefficient corresponding to the penalty factor comprises: calculating a standard deviation of the initial cutting impact intensity sequence as an initial volatility rate, the gain coefficient corresponding to the penalty factor being negatively correlated with the initial volatility rate.

8. The data analysis based feed rate control method in NC machining according to claim 1, characterized by, After the vibration signals and the current signals are acquired, the method further comprises: performing wavelet threshold denoising processing on the vibration signals and the current signals.

9. The feed rate control method in data analysis based NC machining according to claim 1, characterized in that, A sampling frequency range for collecting the vibration signals and the current signals is 10 kHz-20 kHz.

10. A feed rate control system in data analysis based NC machining comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the feed rate control method in numerical control machining based on data analysis according to any one of claims 1-9.

Images