Surface quality real-time monitoring and classifying system

Abstract

The invention relates to a surface quality real-time monitoring and classifying system which comprises a sensor used for collecting displacement signals or acceleration signals in the machining process; the signal processing module comprises a Kalman filter, and the Kalman filter is used for estimating cutting force from the displacement signal or the acceleration signal; the frequency domain transformation module is used for carrying out fast Fourier transformation on the estimated cutting force; and the machine learning module is trained and can classify the surface quality based on the data after the fast Fourier transform. The surface quality real-time monitoring and classifying system provided by the invention is low in cost and high in precision, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of machining, and in particular to a system and method for real-time monitoring and classification of surface quality. Background Art

[0002] During the cutting process of a workpiece by a machine tool, machining vibrations will reduce the surface quality of the workpiece. In particular, chatter is a self-excited vibration phenomenon that occurs during the chip formation process in machining. Even if the tool-side parameters are adjusted, the vibration is still inevitable. Especially for flexible workpieces, as the machining progresses, the vibration frequency of the flexible workpiece will change dynamically. Machining vibrations cause dimensional deviations in the final product by damaging the surface quality, increasing the repair cost of the workpiece, delaying production, wasting raw materials, and affecting the overall production efficiency. Therefore, good surface quality is crucial for high-precision industrial components.

[0003] In the related art, a cutting force dynamometer is used to measure the force signal during the cutting process between the tool and the workpiece, so as to monitor the chatter during the machining process and then judge the surface quality of the machined workpiece. However, the cutting force dynamometer and the supporting amplifier are expensive and not easy to install in complex or space-constrained structures. In addition, for complex or space-constrained occasions, it is difficult to reasonably integrate the force sensor into the monitoring system, which is not convenient for application. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems raised in the above background art, and provide a system and method for real-time monitoring and classification of surface quality.

[0005] To achieve the above purpose, the technical solution of the present invention is as follows:

[0006] A system for real-time monitoring and classification of surface quality, comprising:

[0007] A sensor for collecting displacement signals or acceleration signals during the machining process;

[0008] A signal processing module including a Kalman filter, and the Kalman filter is used to estimate the cutting force from the displacement signal or acceleration signal,

[0009] A frequency domain transformation module for performing a fast Fourier transform on the estimated cutting force;

[0010] A machine learning module, which is trained to be able to classify the surface quality based on the data after the fast Fourier transform.

[0011] In one embodiment, during the initial calibration process, the Kalman filter gain is optimized by minimizing the error between the estimated cutting force and the actual measured cutting force;

[0012] Among them, the system noise parameter and the measurement noise parameter are optimized through the Kalman filter framework algorithm.

[0013] In one embodiment, the sensor is set as a displacement sensor or an acceleration sensor;

[0014] Among them, when the sensor is set as an acceleration sensor, after the acceleration signal collected by the acceleration sensor is converted into a displacement signal, the displacement signal is input into the Kalman filter.

[0015] In one embodiment, the machine learning module includes a multi-scale convolutional feature extractor, a parallel dual-domain attention mechanism composed of channel domain and frequency domain convolutions, and a Transformer encoder.

[0016] In one embodiment, it further includes:

[0017] A data acquisition unit for receiving the signal of the sensor;

[0018] A software interface module connected to the machine learning module for real-time output of the surface quality classification result;

[0019] Among them, the machine learning module is a pre-trained model embedded in the system and can be locally executed during the processing.

[0020] The effects of the present invention:

[0021] The present invention uses sensors with relatively low costs, which are arranged at positions far from the processing area, and constructs a Kalman filter for estimating the dynamic relationship between the processing point and the sensor position. Through the Kalman filter, expensive cutting force dynamometers do not need to be used, and the cutting force can be accurately estimated; then, the estimated cutting force signal after fast Fourier transform is input into the machine learning module. The machine learning module, based on the deep learning architecture, realizes the real-time monitoring and classification of the surface quality caused by chatter; the present invention integrates all links from data acquisition to surface quality classification into one, avoiding the usual isolated processing of each stage in the prior art, establishing a continuous and unified workflow. Using the system and method of the present invention can reduce workpiece surface defects, avoid waste of raw materials, minimize the repair cost of workpiece defects, and improve production and processing efficiency by avoiding subsequent repair processes. Description of the Drawings

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0023] Figure 1 Schematic diagram of the overall application solution of the system of the present invention in a numerically controlled machine tool;

[0024] Figure 2 Schematic diagram of the overall process of the present invention;

[0025] Figure 3 Flowchart of the acceleration-displacement conversion algorithm of the present invention;

[0026] Figure 4 Flowchart of the algorithm of the Kalman filter constructed by the present invention;

[0027] Figure 5 Multi-scale kernel dual-domain attention model with a Transformer encoder of the present invention.

[0028] Reference numerals: 11, spindle; 12, tool holder; 13, fixture; 14, feeding table; 15, machining area; 20, sensor; 30, data acquisition card; 40, computer; 41, computer window; 200, workpiece. Detailed implementation manners

[0029] The following further describes the present invention with reference to the drawings, but does not limit the present invention in any way. Any transformation or replacement based on the teachings of the present invention falls within the protection scope of the present invention.

[0030] The surface quality real-time monitoring and classification system provided by the present invention, based on Kalman filtering and machine learning, can be integrated into existing numerically controlled machine tools or used as an independent module.

[0031] If the sensor is installed close to the machining area, the sensor near the machining area is vulnerable to high-temperature chips, mechanical stress, vibration, high-speed tools and temperature, and is prone to damage. To solve this problem, in the prior art, non-contact optical sensing components are mostly used. However, due to the occlusion of the rotating tool and the chip flow, the non-contact optical sensing component cannot directly monitor the machining position of the tool-workpiece contact surface.

[0032] Such as Figure 1As shown, the arrangement of the sensor 20 in the numerical control machine tool is presented. There is a workpiece fixture 13 on the feeding table 14, the workpiece 200 is located on the fixture 13, the sensor 20 is arranged at the bottom of the workpiece 100, the tool holder 12 holds the tool against the workpiece 200, and the tool holder 12 is connected to the spindle 11. Installing the sensor 20 on the workpiece 200 away from the machining area 15 can avoid damage and reduce noise interference, and can effectively reduce the machine tool downtime caused by the maintenance and replacement of the sensor 20. Thus, through the indirect sensing method, the present invention flexibly arranges the sensor 20 outside the machining area 15, significantly improving the durability and simplifying the system integration process.

[0033] In this embodiment, the sensor 20 is arranged at the bottom of the flexible workpiece 200. Of course, according to needs, the sensor 20 can also be arranged on the tool side.

[0034] Specifically, the sensor 20 can be set as a displacement sensor or an acceleration sensor, with a lower cost. The acceleration sensor or displacement sensor is easier to integrate and apply. Among them, the acceleration sensor is usually easier to install and can adapt to the integration requirements of various configurations.

[0035] The system of the present invention further includes a data acquisition unit for receiving the signals of the sensor. The data acquisition unit can be set as a data acquisition card 30. The data collected by the sensor 20 is input and stored in the data acquisition card 30, and the calculator 40 processes the data of the data acquisition card 30 and can display it through the computer window 41.

[0036] The present invention installs the sensor 20 at a safe distance. Although it avoids damage and interference, this will increase the distance between the sensor 20 and the machining area 15, and the directly obtained data accuracy is low. Therefore, below, the present invention will finely calibrate the signals collected by the sensor 20 through a Kalman filter to maintain the measurement accuracy.

[0037] The real-time surface quality monitoring and classification system of the present invention further includes a signal processing module. The signal processing module includes a Kalman filter, and the Kalman filter is used to estimate the cutting force from the displacement signal or acceleration signal.

[0038] In the prior art, due to the frequent changes in the machining conditions and the differences in the dynamic characteristics of the workpieces, it is difficult to accurately predict the machining results. In a dynamic machining and manufacturing environment, frequent configuration changes require a fast and automated calibration algorithm to achieve real-time process adjustment. For this reason, below, the present invention constructs a Kalman filter, uses the data of the sensor arranged at the bottom of the workpiece, converts the acceleration signal or displacement signal into an equivalent force signal, and estimates the cutting force acting on the machining area, thereby realizing indirect force measurement. When the collected signal is an acceleration signal, the acceleration signal is converted into a displacement signal, as Figure 3 shown.

[0039] The overview of the Kalman filter equations is as follows:

[0040] When the sensor is installed on the workpiece, the workpiece structural flexibility is calculated according to Equation (1) using the impact modal test:

[0041] The Laplace equation of the dynamic compliance is defined as:

[0042]

[0043] where ω, ξ, ζ, α represent the natural frequency (Hz), damping ratio, and flexibility coefficient (unit: mm / N·s 2 ); d(s) represents the displacement; F a (s) represents the applied external force.

[0044] The state - space model of the system is shown as follows:

[0045]

[0046] where x represents the state vector, u represents the input vector or the actual cutting force applied to the tool (F a ); z represents the measurement vector or the displacement converted from the readings of the acceleration sensor / displacement sensor.

[0047] When reconstructing the matrix using the similarity transformation, the equation is expressed as follows:

[0048]

[0049] where A n = T·A s ·T -1 B n = T·B s C n = C s ·T -1

[0050] The definition of the similarity transformation matrix is as follows:

[0051] T = diag(2 4 , 2 18 , 2 31 , 2 43 , 2 54 , 2 65 )

[0052] Considering the system noise w and measurement noise v existing in the actual process, the equation is modified as follows:

[0053]

[0054] Considering the Kalman filter gain K, the equation is derived as follows:

[0055]

[0056] where K represents the Kalman filter gain matrix, represents the estimated value of the actual cutting force F a of.

[0057] Based on the transfer function of the Kalman filter, it is constructed using the modified state space model. The derivation of the transfer function of the continuous Kalman filter is as follows:

[0058]

[0059] Considering the discreteness of the sensor data, the discrete model is defined as follows:

[0060]

[0061] P is the covariance matrix of the state estimation error of,

[0062]

[0063] The minimum covariance matrix P can be derived from the time-varying Riccati Equation,

[0064]

[0065] where Q and R are the system covariance matrix and the measurement covariance matrix, respectively.

[0066] Then, the final optimal Kalman filter gain matrix is calculated as follows:

[0067]

[0068] Substitute the obtained K into Equation (6) for calculating the transfer function between the displacement d and the estimated cutting force between.

[0069] Therefore, by simply multiplying the measured displacement d by the transfer function, the estimated cutting force

[0070] such as Figure 4 shown, the system noise w and the measurement noise v are calibrated through the Kalman filter algorithm, and K is optimized through this algorithm. In this application, the data of the displacement sensor, or the displacement data converted from the acceleration data, can be input into the relevant equations to estimate the actual cutting force.

[0071] Tests have shown that the Kalman filter developed in the present invention is highly effective in expanding the linear response range of sensors, thus significantly improving the overall measurement accuracy. Comparing the actual force measured by the force sensor with the force estimated by the present invention, within the impact action time, the force estimated by the present invention is highly consistent with the actual force measured by the force sensor, thus verifying the accuracy of the Kalman filter. At the same time, the cost of the sensor of the present invention is at least ten times lower than that of the commercial cutting force dynamometer system, thus making the present system have significant economy.

[0072] The real-time surface quality monitoring and classification system of the present invention further includes a frequency domain transformation module. After obtaining the estimated cutting force, the frequency domain transformation module is then used to perform a fast Fourier transform (FFT) on the estimated cutting force, and the data after the fast Fourier transform is input into the developed machine learning module for processing. Based on the FFT features, it is used to infer the chatter frequency characteristics and associate them with the surface state categories.

[0073] The machine learning module includes a multi-scale convolutional feature extractor, a parallel dual-domain attention mechanism composed of channel domain and frequency domain convolutions, and a Transformer encoder (MKDAT). The machine learning module is a pre-trained model embedded in the system and can be executed locally during the machining process.

[0074] As Figure 2 and Figure 5 shown, the machine learning module realizes the real-time classification of surface quality through a deep learning-based monitoring framework. This deep learning architecture combines a multi-scale convolutional kernel module and a parallel dual-fusion attention mechanism with a Transformer encoder, so as to facilitate the extraction of spatial and context features. In this way, it can support efficient surface quality classification and accurate and stable cutting force estimation under different operating conditions, thus promoting the adaptability and intelligence of the machining process.

[0075] The system of the present invention further includes a software interface module. The software interface module is connected to the machine learning module and is used to output the surface quality classification results in real time.

[0076] In the present invention, as Figure 5As shown, a multi-scale kernel dual-domain attention model with a Transformer encoder is developed for deep machine learning. Among them, MKDAT is used for surface quality classification, L is the data length, K1, K2, K3, and K4 are kernels, c is the number of channels, h1 is the number of Transformer heads, c*f1 is the feed-forward dimension, and l1 is the number of layers. The Transformer encoder is used to achieve accurate classification. After testing, when using the developed parallel dual-domain attention mechanism for surface classification, the prediction accuracy of this system is better than that of the serial attention structure, and the average accuracy exceeds 95%, and can reach 100% in flexible workpieces with flip bifurcation flutter. Compared with the serial attention module mechanism, the machine learning architecture developed in the present invention shows better performance.

[0077] The present invention also has an interface developed for monitoring surface quality stability and classification, which enables users to combine the algorithm prediction results with the signals input by the sensor to realize the visualization of the machining process state, analyze the vibration mode in real time, and evaluate and classify the surface quality.

[0078] The present invention integrates artificial intelligence with a perception monitoring module, adopts a modular and reconfigurable design, and has the flexible configuration ability in terms of sensor type, quantity, output format, and communication protocol (such as TCP / IP).

[0079] The system provided by the present invention is applicable to the processes of robotic machining, flexible workpiece handling, and vibration-based surface monitoring.

[0080] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A real-time surface quality monitoring and classification system, characterized in that Comprising: A sensor for collecting displacement signals or acceleration signals during the machining process; A signal processing module including a Kalman filter, which is used to estimate the cutting force from the displacement signal or acceleration signal; A frequency domain transformation module for performing a fast Fourier transform on the estimated cutting force; A machine learning module, which is trained to classify the surface quality based on the data after the fast Fourier transform.

2. The surface quality real-time monitoring and classification system according to claim 1, characterized in that, During the initial calibration process, the Kalman filter gain is optimized by minimizing the error between the estimated cutting force and the actual measured cutting force; Among them, the system noise parameter and the measurement noise parameter are optimized by the Kalman filter framework algorithm.

3. The surface quality real-time monitoring and classification system according to claim 1, wherein The sensor is set as a displacement sensor or an acceleration sensor; Among them, when the sensor is set as an acceleration sensor, the acceleration signal collected by the acceleration sensor is converted into a displacement signal, and then the displacement signal is input into the Kalman filter.

4. The surface quality real-time monitoring and classification system according to claim 1, characterized in that The machine learning module includes a multi-scale convolutional feature extractor, a parallel dual-domain attention mechanism composed of channel domain and frequency domain convolutions, and a Transformer encoder.

5. The surface quality real-time monitoring and classification system according to claim 1, characterized in that, Also comprising: A data acquisition unit for receiving the signal of the sensor; A software interface module connected to the machine learning module for real-time output of the surface quality classification result; Among them, the machine learning module is a pre-trained model embedded in the system and can be executed locally during the machining process.

Images