Feedback control system for the laser-induced thermal crack cutting process of brittle materials

Abstract

The present invention relates to the field of laser processing technology, and specifically relates to a feedback control system for the processing of brittle materials by laser-induced thermal cracking cutting, including a numerical control processing module, a visual signal acquisition module, an acoustic emission signal acquisition module, a multi-modal information fusion module, and a real-time feedback control module; wherein: the numerical control processing module: is used to control the focal position, spot diameter, and processing speed of the laser; the visual signal acquisition module: is used to collect image data at the processing site; the acoustic emission signal acquisition module: is used to collect sound fluctuation data during the processing; the multi-modal information fusion module is used to integrate visual and acoustic emission signals to generate comprehensive features. The present invention significantly improves the accuracy and efficiency of the laser cutting process, reduces material waste, and enhances the adaptability to complex processing environments by integrating multi-modal signal processing and dynamic parameter adjustment.

Description

Technical Field

[0001] The present invention relates to the technical field of laser processing, and particularly to a feedback control system for the processing process of laser-induced thermal cracking of brittle materials. Background Art

[0002] In modern manufacturing, laser-induced thermal cracking cutting technology is widely used in the precision machining of brittle materials such as glass or ceramics. Although this technology is known for its high precision and low damage, the complexity of the processing process and the high demand for precise control still pose significant challenges. Traditional laser cutting systems mainly rely on single-modal signal monitoring, such as visual or acoustic emission monitoring. These methods often cannot comprehensively capture the key physical and chemical changes during the processing when used alone, resulting in difficulty in ensuring the processing quality in real time. In addition, when dealing with complex or rapidly changing processing environments, these systems often cannot adjust the operation parameters in time due to limitations in data processing and response speed, thus affecting the quality and yield of the final product.

[0003] Aiming at the problems existing in the prior art, the present invention aims to solve the deficiencies in real-time monitoring and control during laser cutting. The existing single-modal signal processing technology cannot provide sufficient data support for efficient cutting quality control, especially when dealing with complex material responses and variable working environments. In addition, the response mechanism of the existing system is often delayed and cannot adjust cutting parameters such as laser power, spot diameter, and scanning speed according to the process state obtained in real time to adapt to sudden changes during the processing.

[0004] Therefore, a feedback control system for the processing process of laser-induced thermal cracking of brittle materials is proposed. By integrating advanced fusion analysis technologies of visual and acoustic emission signals, it can achieve precise monitoring and immediate adjustment of the processing process. Summary of the Invention

[0005] Based on the above purposes, the present invention provides a feedback control system for the processing process of laser-induced thermal cracking of brittle materials.

[0006] The feedback control system for the processing process of laser-induced thermal cracking of brittle materials includes a numerical control processing module, a visual signal acquisition module, an acoustic emission signal acquisition module, a multi-modal information fusion module, and a real-time feedback control module; wherein:

[0007] The numerical control processing module: is used to receive adjustment instructions from the real-time feedback control module to control the focal position, spot diameter, and processing speed of the laser to adapt to the processing requirements;

[0008] The visual signal acquisition module: is used to acquire image data at the processing site and send the image information to the multi-modal information fusion module to provide visual basic data for signal fusion;

[0009] Acoustic emission signal acquisition module: It is used to collect the sound fluctuation data during the processing, and after converting the sound data into digital signals, it transmits them to the multi-modal information fusion module to provide acoustic information input;

[0010] Multi-modal information fusion module: It adopts an end-to-end fusion model constructed based on a deep convolutional neural network and a multi-head self-attention mechanism. This end-to-end fusion model is used to integrate visual and acoustic emission signals, strengthen the cross-correlation analysis between different modalities through the self-attention mechanism, and generate comprehensive features and output them to the real-time feedback control module;

[0011] Real-time feedback control module: It receives the analysis results of the multi-modal information fusion module, and dynamically adjusts the operating parameters of the numerical control processing module based on the comprehensive features, including laser power, spot diameter, and scanning speed, to optimize the processing process in real time.

[0012] Furthermore, the numerical control processing module includes a laser generator unit, a numerical control moving platform unit, and a control interface unit; among them:

[0013] Laser generator unit: It is used to generate a laser beam and adjust its power output. Specifically, according to the instructions from the real-time feedback control module, it dynamically adjusts the laser power, and the power adjustment is achieved by controlling the power input of the laser generator through a frequency converter;

[0014] Numerical control moving platform unit: It includes a stepping motor and a linear slide rail, and is used to control the position of the laser head or the workpiece to be processed. This numerical control moving platform unit receives the coordinate adjustment signal from the real-time feedback control module, and precisely moves the laser head through a high-precision position control system to adjust the focal position and spot diameter of the laser;

[0015] Control interface unit: It is responsible for parsing the adjustment instructions from the real-time feedback control module. This instruction is based on the output of the multi-modal information fusion module, including processing quality indicators and prediction results, and indicates the specific adjustments of laser power, focal position, spot diameter, and moving speed.

[0016] Furthermore, the visual signal acquisition module includes a high-resolution CCD industrial camera unit, an image acquisition card unit, and an image preprocessing unit; among them:

[0017] High-resolution CCD industrial camera unit: It is used to capture high-definition images of the processing site in real time. This high-resolution CCD industrial camera unit is equipped with an imaging sensor, which can capture images with rich details under various lighting conditions, ensuring stable and clear visual data even in a fast-moving or changing processing environment;

[0018] Image acquisition card unit: It is used to receive the original image data from the CCD industrial camera and perform preliminary digital signal processing. This image acquisition card unit also supports high-speed data transmission and large-capacity caching, and can process and transfer image data without delay;

[0019] Image preprocessing unit: It is used to perform preliminary analysis and preprocessing on the acquired images, including denoising, contrast adjustment, and edge enhancement.

[0020] Furthermore, the acoustic emission signal acquisition module includes an acoustic emission sensor unit, a preamplifier unit, and a digital signal converter unit; among them:

[0021] Acoustic emission sensor unit: It is used to capture the acoustic wave fluctuations generated during the processing in real time. This acoustic emission sensor unit is equipped with highly sensitive acoustic sensors, which can accurately capture the acoustic waves generated from tiny material deformations to obvious fracture events, so as to provide real-time data for monitoring the material processing state;

[0022] Preamplifier unit: It is used to enhance the weak acoustic wave signals received from the acoustic emission sensors and improve the accuracy and efficiency of subsequent digital conversion;

[0023] Digital signal converter unit: It converts the analog acoustic wave signals output by the preamplifier into digital signals, and specifically operates through analog-to-digital conversion technology, so that the digital signals can be processed by the multi-modal information fusion module.

[0024] Furthermore, the multi-modal information fusion module includes a deep convolutional neural network unit, an acoustic emission feature extraction unit, a multi-head self-attention mechanism unit, and a comprehensive feature generation unit; among them:

[0025] Deep convolutional neural network unit: It is used to extract visual features from the image data received by the visual signal acquisition module. This deep convolutional neural network includes multiple convolutional layers, pooling layers, and activation layers, and can identify and analyze the predetermined features in the images, including edges, textures, and patterns;

[0026] Acoustic emission feature extraction unit: It extracts acoustic features from the digital signals provided by the acoustic emission signal acquisition module through preset signal processing algorithms, including frequency analysis, energy distribution, and time-domain features;

[0027] Multi-head self-attention mechanism unit: It is used to fuse the visual and acoustic features provided by the deep convolutional neural network unit and the acoustic emission feature extraction unit, and uses the attention model to strengthen the correlation and complementarity between different modal features and optimize the feature fusion process;

[0028] Comprehensive Feature Generation Unit: Based on the multi-head self-attention mechanism, it converts the fused visual and acoustic features into a comprehensive feature vector, which details the quality indicators, abnormal state identifications, and potential prediction results during the processing, providing decision-making support for real-time feedback control.

[0029] Further, the acoustic emission feature extraction unit includes:

[0030] Signal preprocessing: First, preprocess the digital signals received from the acoustic emission signal acquisition module to reduce noise and improve signal quality. The preprocessing steps include the application of a digital filter to remove signal components of non-target frequencies. The preprocessing formula is:

[0031] Y(f) = H(f)X(f), where X(f) is the frequency-domain representation of the original signal, H(f) is the transfer function of the selected filter, and Y(f) is the filtered signal;

[0032] Frequency analysis: Next, use the fast Fourier transform algorithm to perform frequency analysis on the preprocessed signal to identify the main frequency components contained in the signal and determine the characteristic frequencies in the acoustic emission events, which are related to the rupture and stress changes of the material. The calculation formula for frequency analysis is:

[0033] where x(n) is the sample point of the preprocessed time-domain acoustic emission signal; N is the number of sample points for the FFT transform; k is the frequency index; F(k) represents the k-th complex value in the fast Fourier transform result, representing the amplitude and phase of the component with frequency k;

[0034] Energy distribution: Use Parseval's theorem to calculate the total energy of the signal and the energy distribution within a predetermined frequency band to assist in evaluating the energy magnitude of the acoustic emission event, thereby inferring the severity of material deformation or rupture. The calculation formula for energy distribution is: where x(n) is the time-domain sample point of the acoustic emission signal, and E is the total energy of the signal;

[0035] Time-domain feature extraction: Finally, calculate the time-domain features of the signal, including peak time, signal duration, and rise time. For the peak time t peak it is calculated by locating the point with the maximum signal amplitude. The signal duration t duration and the rise time t rise can be determined by analyzing the time interval from the start of the signal to the maximum amplitude.

[0036] Further, the multi-head self-attention mechanism unit includes:

[0037] Preparation before feature fusion: First, receive the visual features extracted by the deep convolutional neural network unit and the acoustic features extracted by the acoustic emission feature extraction unit, and integrate these features into two independent feature vectors, namely the visual feature vector S and the acoustic feature vector A, and perform normalization processing to match their scales;

[0038] Attention score calculation: Use the multi-head attention mechanism to calculate the attention scores between different modalities. Each head uses a different set of weights to process the visual and acoustic features separately. The specific attention calculation formula is: where the query matrix Q, the key matrix K, and the value matrix V are respectively obtained by transforming the visual and acoustic features; d k is the dimension of the key vector, used to scale the dot product result to prevent gradient disappearance;

[0039] Feature fusion strategy: Based on the calculated attention scores, fuse the information of different features together. For each pair of visual and acoustic features, this multi-head attention mechanism will dynamically adjust their weights to strengthen the contribution of the features that are more important for the current task. The feature fusion is specifically carried out through the following formula:

[0040] where, α i and β i are the weight coefficients of the visual and acoustic features calculated by the attention mechanism, S i and A i are the specific feature vectors, and n is the number of features;

[0041] Output of the comprehensive feature vector: Through the above steps, the obtained fused feature vector F fused will contain the predetermined visual and acoustic information during the processing, and this information has been optimized to highlight the most relevant features.

[0042] Furthermore, the comprehensive feature generation unit includes:

[0043] Feature integration: After being processed by the multi-head self-attention mechanism, the visual and acoustic features have been weighted and fused accordingly. The comprehensive feature generation unit first receives these fused feature data and uses a fully connected layer to integrate the features to form a unified feature vector: F integrated = W f ·F fused + b f , F fused is the fused feature vector from the multi-head self-attention mechanism, W f and b f are respectively the weight matrix and the bias vector of the fully connected layer; F integrated is the integrated feature vector;

[0044] Feature mapping: Then, the comprehensive feature vector F integrated is converted through the mapping layer into a form that describes specific quality indicators, abnormal state identifiers, and prediction results during the processing. The mapping layer includes a linear layer, an activation function, or a machine learning model. The specific formulas are as follows:

[0045] F quality = W q ·F integrated + b q ;

[0046] F anomaly = W a ·F integrated + b a ;

[0047] F prediction = W p ·F integrated + b p ; where F quality describes the indicators related to processing quality; F anomaly indicates potential abnormal states; F predictiom provides predictions of future processing results; W q , W a , W p and b q , b a , b p are the weight matrices and bias vectors of each mapping layer respectively;

[0048] Output comprehensive feature vector: Finally, F quality , F anomaly and F prediction are combined or processed separately to form a comprehensive feature vector for real-time feedback to the control system to guide the adjustment and optimization of the processing process.

[0049] Furthermore, the real-time feedback control module includes a parameter decision unit, an execution control unit, and a feedback adjustment unit; where:

[0050] Parameter decision unit: Used to receive the comprehensive feature vector from the comprehensive feature generation unit. This vector includes detailed information on processing quality, abnormal states, and prediction results. Based on this detailed information, the parameter decision unit uses a predefined decision algorithm to determine the operating parameters that need to be adjusted. The specific decision formula is:

[0051] ΔP = f power (F quality , F anomaly );

[0052] ΔD = f diameter (F quality , Fprediction );

[0053] ΔS = f speed (F ano,aly , F prediction ), where ΔP, ΔD, and ΔS represent the adjustment amounts of laser power, spot diameter, and scanning speed respectively, and f power , f diameter and f speed are adjustment functions extracted from the eigenvector and calculated;

[0054] Execution control unit: Directly controls the corresponding hardware in the numerical control machining module according to the adjustment amount instructions provided by the parameter decision unit. Specifically, it changes the laser power by adjusting the input voltage of the laser generator unit, or adjusts the numerical control moving platform unit to change the spot diameter and scanning speed. The specific calculation formula for execution control is: P new = P current + ΔP; D new = D current + ΔD; S new = S current + ΔS; where P current , D current and S current are the current laser power, spot diameter, and scanning speed, while P new , D new and S new are the new adjusted parameters;

[0055] Feedback regulation unit: Used to monitor the operation results after adjustment and feed back real-time data to the parameter decision unit for the next round of adjustment decision-making.

[0056] Advantages of the present invention:

[0057] The present invention, by integrating multi-modal signals and visual and acoustic emission signals, provides more comprehensive data analysis and processing capabilities than traditional single-modal signal monitoring methods. Using the multi-head self-attention mechanism to perform advanced fusion and analysis on these multi-modal signals enables the system to more accurately identify key variables and potential problems during the machining process. This enhanced monitoring ability ensures more precise quality control of the machining process and greatly improves the machining quality and material utilization rate.

[0058] The present invention can not only monitor the machining state in real time but also dynamically adjust key operating parameters such as laser power, spot diameter, and scanning speed to adapt to the immediate changes during the machining process. This ability stems from the detailed output of the comprehensive feature generation unit, which reflects the quality indicators, abnormal states, and prediction results of the machining process and provides a decision-making basis for the real-time feedback control module. This dynamic adjustment mechanism greatly enhances the adaptability and flexibility of the system to complex machining environments.

[0059] In the present invention, by precisely controlling and instantaneously adjusting the processing parameters, the material waste caused by processing errors is significantly reduced, and the entire processing flow is optimized. Higher processing accuracy and fewer production defects mean higher material utilization rate and production efficiency, which is particularly important for large-scale production. In addition, the automated characteristics of the system reduce the workload of operators, lower the operation complexity, and further improve the overall operation efficiency and safety of the production line. Brief Description of the Drawings

[0060] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0061] Figure 1 Schematic diagram of the real-time feedback control system according to an embodiment of the present invention;

[0062] Figure 2 Schematic diagram of the real-time feedback control module according to an embodiment of the present invention. Detailed Embodiments

[0063] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the following further details the present invention in combination with specific embodiments.

[0064] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those of ordinary skill in the field to which the present invention pertains. The "first", "second", and similar terms used in the present invention do not indicate any order, quantity, or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0065] As Figure 1 - Figure 2 shown, the feedback control system for the laser-induced thermal cracking cutting of brittle materials during the processing process includes a numerical control processing module, a visual signal acquisition module, an acoustic emission signal acquisition module, a multi-modal information fusion module, and a real-time feedback control module; among them:

[0066] Numerical control processing module: It is used to receive adjustment instructions from the real-time feedback control module to control the focus position, spot diameter and processing speed of the laser to adapt to the processing requirements;

[0067] Visual signal acquisition module: It is used to collect image data at the processing site and send the image information to the multi-modal information fusion module to provide visual basic data for signal fusion;

[0068] Acoustic emission signal acquisition module: It is used to collect sound fluctuation data during the processing process. After converting the sound data into digital signals, it transmits them to the multi-modal information fusion module to provide acoustic information input;

[0069] Multi-modal information fusion module: It adopts an end-to-end fusion model constructed based on a deep convolutional neural network and a multi-head self-attention mechanism. This end-to-end fusion model is used to integrate visual and acoustic emission signals, strengthen the cross-correlation analysis between different modalities through the self-attention mechanism, and generate comprehensive features and output them to the real-time feedback control module;

[0070] Real-time feedback control module: It receives the analysis results of the multi-modal information fusion module and dynamically adjusts the operating parameters of the numerical control processing module based on the comprehensive features, including laser power, spot diameter and scanning speed, to optimize the processing process in real time.

[0071] The numerical control processing module includes a laser generator unit, a numerical control moving platform unit and a control interface unit; among them:

[0072] Laser generator unit: It is used to generate a laser beam and adjust its power output. Specifically, according to the instructions from the real-time feedback control module, it dynamically adjusts the laser power. The power adjustment is achieved by controlling the power input of the laser generator through a frequency converter to match the processing conditions analyzed by the multi-modal information fusion module and ensure the best matching of laser energy and efficient cutting;

[0073] Numerical control moving platform unit: It includes a stepping motor and a linear slide rail and is used to control the position of the laser head or the workpiece to be processed. This numerical control moving platform unit receives coordinate adjustment signals from the real-time feedback control module and accurately moves the laser head through a high-precision position control system to adjust the focus position and spot diameter of the laser. This adjustment relies on microstepping drive technology to ensure the high-precision positioning of the laser focus, thereby optimizing the cutting path and quality;

[0074] Control Interface Unit: As the communication core between the numerical control machining module and the real-time feedback control module, it is responsible for parsing the adjustment instructions from the real-time feedback control module. This instruction is based on the output of the multi-modal information fusion module, including machining quality indicators and prediction results, and indicates specific adjustments to laser power, focus position, spot diameter, and moving speed. The control interface unit uses a microprocessor and a predetermined software algorithm to convert the received adjustment instructions into control signals for the motor and laser generator, ensuring a fast and accurate response to the requirements of real-time feedback. The above design allows the numerical control machining module to make real-time adjustments according to the comprehensive feature output, dynamically optimize the laser cutting parameters, thereby maximizing the cutting accuracy and material utilization rate while ensuring the machining efficiency. This not only improves the product quality but also reduces material waste and machining costs, meeting the needs of modern high-efficiency intelligent manufacturing.

[0075] The visual signal acquisition module includes a high-resolution CCD industrial camera unit, an image acquisition card unit, and an image preprocessing unit; among them:

[0076] High-resolution CCD industrial camera unit: It is used to capture high-definition images of the machining site in real time. This high-resolution CCD industrial camera unit is equipped with an imaging sensor, which can capture detailed images under various lighting conditions, ensuring stable and clear visual data even in a fast-moving or changing machining environment.

[0077] Image acquisition card unit: It is used to receive the raw image data from the CCD industrial camera and perform preliminary digital signal processing. This image acquisition card unit also supports high-speed data transmission and large-capacity caching, and can process and transfer image data without delay, thus ensuring the integrity and real-time nature of the data during transmission.

[0078] Image preprocessing unit: It is used to perform preliminary analysis and preprocessing on the acquired images, including denoising, contrast adjustment, and edge enhancement. These preprocessing steps are to improve the image quality and optimize the subsequent feature extraction and analysis process. The visual signal acquisition module sends the preprocessed image data to the multi-modal information fusion module, providing the necessary visual basic data for this module. This process is achieved through a high-speed communication interface, ensuring the timeliness and accuracy of data transmission, so that the multi-modal information fusion module can effectively combine visual information with acoustic emission data for in-depth analysis and feature fusion. In this way, the visual signal acquisition module provides key visual input for the entire system, supporting the system for high-precision machining monitoring and real-time feedback control.

[0079] The acoustic emission signal acquisition module includes an acoustic emission sensor unit, a preamplifier unit, and a digital signal converter unit; among them:

[0080] Acoustic Emission Sensor Unit: It is used to capture the acoustic wave fluctuations generated during the machining process in real time. This acoustic emission sensor unit is equipped with highly sensitive acoustic sensors, which can accurately capture the acoustic waves generated from tiny material deformations to obvious fracture events, thereby providing real-time data for monitoring the material machining state;

[0081] Pre-amplifier Unit: It is used to enhance the weak acoustic wave signals received from the acoustic emission sensors, improving the accuracy and efficiency of subsequent digital conversion;

[0082] Digital Signal Converter Unit: It converts the analog acoustic wave signals output by the pre-amplifier into digital signals, specifically operating through analog-to-digital conversion (ADC) technology to ensure the faithful expression and high-speed processing of acoustic wave data, enabling the digital signals to be processed by the multi-modal information fusion module. After converting the acoustic wave data into digital signals, the acoustic emission signal acquisition module transfers the digital signals to the multi-modal information fusion module through a high-speed data interface. In the multi-modal information fusion module, these acoustic information will be analyzed and fused with visual information to form a comprehensive understanding of the material behavior during the machining process. This fused data is crucial for accurately evaluating the machining quality, predicting potential defects, and real-time adjusting machining parameters, thereby improving the machining accuracy and efficiency.

[0083] The multi-modal information fusion module includes a deep convolutional neural network unit, an acoustic emission feature extraction unit, a multi-head self-attention mechanism unit, and a comprehensive feature generation unit; among which:

[0084] Deep Convolutional Neural Network Unit: It is used to extract visual features from the image data received by the visual signal acquisition module. This deep convolutional neural network includes multiple convolutional layers, pooling layers, and activation layers, and can identify and analyze the predetermined features in the image, including edges, textures, and patterns, providing highly detailed data analysis for the visual monitoring of the machining process;

[0085] Acoustic Emission Feature Extraction Unit: It extracts acoustic features from the digital signals provided by the acoustic emission signal acquisition module through preset signal processing algorithms, including frequency analysis, energy distribution, and time-domain features. This acoustic data provides important information about the material stress state and possible structural changes;

[0086] Multi-head Self-attention Mechanism Unit: It is used to fuse the visual and acoustic features provided by the deep convolutional neural network unit and the acoustic emission feature extraction unit, using the attention model to strengthen the correlation and complementarity between different modal features, optimizing the feature fusion process, and improving the accuracy and efficiency of data analysis;

[0087] Comprehensive Feature Generation Unit: Based on the multi-head self-attention mechanism, it converts the fused visual and acoustic features into a comprehensive feature vector, which details the quality indicators, abnormal state identifiers, and potential prediction results during the processing, providing decision-making support for real-time feedback control.

[0088] The acoustic emission feature extraction unit includes:

[0089] Signal preprocessing: First, preprocess the digital signals received from the acoustic emission signal acquisition module to reduce noise and improve signal quality. The preprocessing steps include the application of digital filters such as Butterworth, Chebyshev, or elliptic filters to remove signal components of non-target frequencies. The preprocessing formula is:

[0090] Y(f) = H(f)X(f), where X(f) is the frequency-domain representation of the original signal, H(f) is the transfer function of the selected filter, and Y(f) is the filtered signal;

[0091] Frequency analysis: Next, use the Fast Fourier Transform (FFT) algorithm to perform frequency analysis on the preprocessed signal to identify the main frequency components contained in the signal and determine the characteristic frequencies in the acoustic emission events, which are related to the rupture and stress changes of the material. The calculation formula for frequency analysis is:

[0092] where x(n) are the sample points of the preprocessed time-domain acoustic emission signal, which represent the variation of the acoustic wave intensity of the acoustic emission event over time; N is the number of sample points for the FFT transform, which is usually a power of 2 and determines the frequency-domain resolution and calculation efficiency; k is the frequency index, representing the k-th frequency component in the FFT result; F(k) represents the k-th complex value in the Fast Fourier Transform result, representing the amplitude and phase of the component with frequency k;

[0093] Energy distribution: Use Parseval's theorem to calculate the total energy of the signal and the energy distribution within a predetermined frequency band to assist in evaluating the energy magnitude of the acoustic emission event and thereby infer the severity of material deformation or rupture. The calculation formula for energy distribution is: where x(n) are the time-domain sample points of the acoustic emission signal, E is the total energy of the signal, calculated as the sum of the squares of the amplitudes of all sample points, reflecting the total energy of the acoustic emission event and related to the energy release amount of material rupture or deformation;

[0094] Time-domain feature extraction: Finally, calculate the time-domain features of the signal, including the peak time, signal duration, and rise time. These time-domain features are crucial for depicting the morphology and dynamic characteristics of the acoustic emission signal. For the peak time t peak is calculated by locating the point with the maximum signal amplitude, and the signal duration tduration and the rise time t rise can be determined by analyzing the time interval from the start to the maximum amplitude of the signal; through the above steps, the acoustic emission feature extraction unit can accurately extract key acoustic features from complex acoustic emission signals, and these features will then be sent to the multi-modal information fusion module for further processing and analysis for real-time monitoring and control of the machining process.

[0095] The multi-head self-attention mechanism unit includes:

[0096] Preparation before feature fusion: First, receive the visual features (such as edges, textures, and patterns) extracted by the deep convolutional neural network unit and the acoustic features (such as frequency analysis, energy distribution, and time-domain features) extracted by the acoustic emission feature extraction unit, and integrate these features into two independent feature vectors, namely the visual feature vector S and the acoustic feature vector A, and perform normalization processing to match their scales;

[0097] Calculation of attention scores: Use the multi-head attention mechanism to calculate the attention scores between different modalities. Each head uses a different set of weights to process visual and acoustic features separately. The specific attention calculation formula is: where the query matrix Q, the key matrix K, and the value matrix V are respectively obtained by converting visual and acoustic features; d k is the dimension of the key vector, used to scale the dot product result to prevent gradient disappearance;

[0098] Feature fusion strategy: Based on the calculated attention scores, fuse the information of different features together. For each pair of visual and acoustic features, this multi-head attention mechanism will dynamically adjust their weights to strengthen the contribution of features that are more important for the current task. The specific feature fusion is carried out through the following formula:

[0099] where, α i and β i are the weight coefficients of visual and acoustic features calculated by the attention mechanism, S i and A i are specific feature vectors, and n is the number of features;

[0100] Output the comprehensive feature vector: Through the above steps, the obtained fused feature vector F fused will contain the predetermined visual and acoustic information in the machining process, and this information has been optimized to highlight the most relevant features. This comprehensive feature vector will be transmitted to the real-time feedback control module for further processing and decision support; through the implementation of this detailed multi-head self-attention mechanism, the effective fusion between visual and acoustic data is ensured, enhancing the control accuracy and response speed of the system to the machining process, thereby improving the overall machining quality and efficiency.

[0101] The comprehensive feature generation unit includes:

[0102] Feature integration: After being processed by the multi-head self-attention mechanism, the visual and acoustic features have been weighted and fused accordingly. The comprehensive feature generation unit first receives these fused feature data and uses a fully connected layer to integrate the features to form a unified feature vector: F integrated = W f ·F fused + b f , where F fused is the fused feature vector from the multi-head self-attention mechanism, W f and b f are the weight matrix and bias vector of the fully connected layer respectively; F integrated is the integrated feature vector;

[0103] Feature mapping: Then the comprehensive feature vector F integrated is converted through a mapping layer into a form that describes specific quality indicators, abnormal state identifiers, and prediction results during the processing. This mapping layer includes a linear layer, an activation function, or a machine learning model. The specific formulas include:

[0104] F quality = W q ·F integrated + b q ;

[0105] F anomaly = W a ·F integrated + b a ;

[0106] F prediction = W p ·F integrated + b p ; where F quality describes the indicators related to processing quality; F anomaly indicates potential abnormal states; F prediction provides predictions of future processing results; W q , W a , W p and b q , b a , b p are the weight matrices and bias vectors of each mapping layer respectively;

[0107] Output comprehensive feature vector: Finally, F quality , F anomaly and F predictionBe merged or processed separately to form a comprehensive feature vector, which is used to provide real-time feedback to the control system to guide the adjustment and optimization of the machining process. Through this method, the comprehensive feature generation unit not only integrates visual and acoustic information, but also converts this information into practical metrics directly applicable to production process control, enhancing the intelligence and automation levels of the production process.

[0108] The real-time feedback control module includes a parameter decision unit, an execution control unit, and a feedback adjustment unit; among them:

[0109] Parameter decision unit: It is used to receive the comprehensive feature vector from the comprehensive feature generation unit. This vector includes detailed information on machining quality, abnormal status, and prediction results. Based on this detailed information, the parameter decision unit uses a predefined decision algorithm to determine the operating parameters that need to be adjusted. The decision algorithm can be a rule-based system or a more complex machine learning model. The specific decision formula is:

[0110] ΔP = f power (F quality , F anomaly );

[0111] ΔD = f diameter (F quality , F prediction );

[0112] ΔS = f speed (F anomaly , F prediction ),where ΔP, ΔD, and ΔS represent the adjustment amounts of laser power, spot diameter, and scanning speed respectively, and f power , f diameter , and f speed are adjustment functions extracted from the feature vector and calculated;

[0113] Execution control unit: Directly control the corresponding hardware in the numerical control machining module according to the adjustment amount instructions provided by the parameter decision unit. Specifically, change the laser power by adjusting the input voltage of the laser generator unit, or adjust the numerical control moving platform unit to change the spot diameter and scanning speed. The specific calculation formula for execution control is: P new = P current + ΔP; D new = D current + ΔD; S new = S current + ΔS; where P current , D current , and S current are the current laser power, spot diameter, and scanning speed, while P new , D new , and S newIt is the new parameter after adjustment;

[0114] Feedback adjustment unit: used to monitor the operation result after adjustment and feed the real-time data back to the parameter decision-making unit for the next round of adjustment decision-making. This closed-loop control mechanism ensures that the system can continuously optimize the processing process, respond immediately to any sudden quality problems or predicted abnormal states. Through this method, the real-time feedback control module can dynamically adjust the operation parameters of the numerical control processing module according to the comprehensive feature vector obtained from multi-modal information, ensuring the optimization and high efficiency of the laser processing process, while reducing processing defects and improving product quality.

[0115] The present invention aims to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. Feedback control system for laser induced thermal cracking cutting of brittle materials, characterized in that: It includes CNC machining module, visual signal acquisition module, acoustic emission signal acquisition module, multimodal information fusion module and real-time feedback control module; among which: CNC machining module: used to receive adjustment instructions from the real-time feedback control module to control the focal position, spot diameter and machining speed of the laser to meet the machining requirements; Visual signal acquisition module: used to collect image data at the processing site and send the image information to the multimodal information fusion module to provide visual basic data for signal fusion; Acoustic emission signal acquisition module: used to collect sound fluctuation data during the processing, convert the sound data into digital signals, and transmit them to the multimodal information fusion module to provide acoustic information input; Multimodal information fusion module: It adopts an end-to-end fusion model based on deep convolutional neural network and multi-head self-attention mechanism. This end-to-end fusion model is used to integrate visual and acoustic emission signals, strengthen the cross-correlation analysis between different modalities through the self-attention mechanism, and generate comprehensive features to output to the real-time feedback control module; The multimodal information fusion module includes a deep convolutional neural network unit, an acoustic emission feature extraction unit, a multi-head self-attention mechanism unit and a comprehensive feature generation unit; wherein: A deep convolutional neural network unit: used to extract visual features from the image data received by the visual signal acquisition module. The deep convolutional neural network includes multiple layers of convolutional layers, pooling layers, and activation layers, and can recognize and analyze predetermined features in the image, including edges, textures, and patterns; Acoustic emission feature extraction unit: extracts acoustic features from the digital signal provided by the acoustic emission signal acquisition module through a preset signal processing algorithm, including frequency analysis, energy distribution and time domain features; Multi-head self-attention mechanism unit: used to fuse the visual and acoustic features provided by the deep convolutional neural network unit and the acoustic emission feature extraction unit, using the attention model to strengthen the correlation and complementarity between different modal features and optimize the feature fusion process; Comprehensive feature generation unit: Based on the multi-head self-attention mechanism, the fused visual and acoustic features are converted into a comprehensive feature vector, which describes in detail the quality indicators, abnormal state identification and potential prediction results of the processing process, providing decision support for real-time feedback control; Real-time feedback control module: receives the analysis results of the multimodal information fusion module and dynamically adjusts the operating parameters of the CNC machining module, including laser power, spot diameter and scanning speed, based on the comprehensive characteristics to optimize the machining process in real time.

2. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 1, characterized in that: The CNC machining module includes a laser generator unit, a CNC mobile platform unit and a control interface unit; wherein: Laser generator unit: used to generate laser beam and adjust its power output. Specifically, according to the instructions from the real-time feedback control module, the laser power is dynamically adjusted. The power adjustment is achieved by controlling the power input of the laser generator through the frequency converter; CNC mobile platform unit: including stepper motor and linear slide rail, used to control the position of laser head or workpiece. The CNC mobile platform unit receives the coordinate adjustment signal from the real-time feedback control module, accurately moves the laser head through the high-precision position control system, and adjusts the focal position and spot diameter of the laser. Control interface unit: responsible for parsing the adjustment instructions from the real-time feedback control module, which are based on the output of the multimodal information fusion module, including processing quality indicators and prediction results, and indicate the specific adjustment of laser power, focal position, spot diameter and moving speed.

3. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 1, characterized in that: The visual signal acquisition module includes a high-resolution CCD industrial camera unit, an image acquisition card unit and an image preprocessing unit; wherein: High-resolution CCD industrial camera unit: used to capture high-definition images of the processing site in real time. The high-resolution CCD industrial camera unit is equipped with an imaging sensor that can capture detailed images under various lighting conditions, ensuring stable and clear visual data in fast-moving or changing processing environments; Image acquisition card unit: used to receive raw image data from CCD industrial cameras and perform preliminary digital signal processing. The image acquisition card unit also supports high-speed data transmission and large-capacity cache, and can process and transfer image data without delay; Image preprocessing unit: used to perform preliminary analysis and preprocessing on the acquired images, including denoising, contrast adjustment and edge enhancement.

4. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 1, characterized in that: The acoustic emission signal acquisition module includes an acoustic emission sensor unit, a preamplifier unit and a digital signal converter unit; wherein: Acoustic emission sensor unit: used to capture the acoustic wave fluctuations generated during the processing process in real time. The acoustic emission sensor unit is equipped with a highly sensitive acoustic sensor that can accurately capture the sound waves generated from tiny material deformation to obvious fracture events, thereby providing real-time data for monitoring the material processing status; Preamplifier unit: used to enhance the weak sound wave signal received from the acoustic emission sensor and improve the accuracy and efficiency of subsequent digital conversion; Digital signal converter unit: converts the analog sound wave signal output by the preamplifier into a digital signal, specifically through analog-to-digital conversion technology, so that the digital signal can be processed by the multimodal information fusion module.

5. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 1, characterized in that: The acoustic emission feature extraction unit comprises: Signal preprocessing: First, the digital signal received from the acoustic emission signal acquisition module is preprocessed to reduce noise and improve signal quality. The preprocessing step includes the application of a digital filter to remove signal components of non-target frequencies. The preprocessing formula is: ,in, is the frequency domain representation of the original signal, is the transfer function of the selected filter, is the filtered signal; Frequency analysis: Next, the preprocessed signal is subjected to frequency analysis using the fast Fourier transform algorithm to identify the main frequency components contained in the signal to determine the characteristic frequency in the acoustic emission event, which is related to the rupture and stress change of the material. The calculation formula for frequency analysis is: in, is the sample point of the preprocessed time domain acoustic emission signal; for The number of sample points for the transformation; is the frequency index; Represents the first complex value, representing the frequency The amplitude and phase of the components of Energy distribution: Parseval's theorem is used to calculate the total energy of the signal and the energy distribution within a predetermined frequency band to assist in evaluating the energy of the acoustic emission event, thereby inferring the severity of material deformation or rupture. The energy distribution calculation formula is: ,in, is the time domain sample point of the acoustic emission signal, is the total energy of the signal; Time domain feature extraction: Finally, calculate the time domain features of the signal, including peak time, signal duration and rise time. By locating the point where the signal amplitude is the largest, the signal duration is calculated. and rise time It can be determined by analyzing the time interval from the beginning of the signal to the time when it reaches the maximum amplitude.

6. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 5, characterized in that: The multi-head self-attention mechanism unit includes: Preparation before feature fusion: First, receive the visual features extracted by the deep convolutional neural network unit and the acoustic features extracted by the acoustic emission feature extraction unit, and integrate the features into two independent feature vectors, namely the visual feature vector Harmonic eigenvector , and normalized to match their scales; Attention score calculation: A multi-head attention mechanism is used to calculate the attention scores between different modalities. Each head uses a different set of weights to process visual and acoustic features respectively. The specific attention calculation formula is: , where the query matrix , key matrix Sum Matrix , converted from visual and acoustic features respectively; is the dimension of the key vector, used to scale the dot product result to prevent gradient vanishing; Feature fusion strategy: Based on the calculated attention scores, the information of different features is fused together. For each pair of visual and acoustic features, the multi-head attention mechanism dynamically adjusts their weights to strengthen the contribution of features that are more important to the current task. The feature fusion is performed using the following formula: in, and is the weight coefficient of visual and acoustic features calculated by the attention mechanism, and is a specific feature vector, is the number of features; Output comprehensive feature vector: Through the above steps, the fused feature vector obtained It will contain predefined visual and acoustic information about the process, which has been optimized to highlight the most relevant features.

7. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 6, characterized in that: The comprehensive feature generating unit comprises: Feature integration: After the multi-head self-attention mechanism, the visual and acoustic features have been weighted and fused accordingly. The comprehensive feature generation unit first receives these fused feature data and uses a fully connected layer to integrate the features to form a unified feature vector: , is the fused feature vector from the multi-head self-attention mechanism, and They are the weight matrix and bias vector of the fully connected layer respectively; is the integrated eigenvector; Feature Mapping: Then synthesize the feature vector It is converted into a form that describes specific quality indicators, abnormal state identification, and prediction results in the processing process through a mapping layer. The mapping layer includes a linear layer, an activation function, or a machine learning model. The specific formulas include: ； ； ;in, Describe the indicators related to processing quality; Indicates a potential abnormal condition; Provide predictions of future processing results; and are the weight matrix and bias vector of each mapping layer respectively; Output comprehensive feature vector: Final , and They are combined or processed separately to form a comprehensive feature vector, which contains all the necessary monitoring and control information and is used to provide real-time feedback to the control system to guide the adjustment and optimization of the machining process.

8. The feedback control system for laser induced thermal cracking cutting of brittle materials according to claim 7, characterized in that: The real-time feedback control module includes a parameter decision unit, an execution control unit and a feedback adjustment unit; wherein: Parameter decision unit: used to receive the comprehensive feature vector from the comprehensive feature generation unit, which includes detailed information on processing quality, abnormal status and predicted results. Based on the detailed information, the parameter decision unit uses a predefined decision algorithm to determine the operating parameters that need to be adjusted. The specific decision formula is: ； ； ,in, and Represent the adjustment amount of laser power, spot diameter and scanning speed respectively, and is the adjustment function calculated by extracting information from the feature vector; Execution control unit: directly controls the corresponding hardware in the CNC machining module according to the adjustment amount instruction provided by the parameter decision unit, specifically by adjusting the input voltage of the laser generator unit to change the laser power, or adjusting the CNC mobile platform unit to change the spot diameter and scanning speed. The specific calculation formula for execution control is: ; ; ;in, and is the current laser power, spot diameter and scanning speed, and is the new parameter after adjustment; Feedback adjustment unit: used to monitor the adjusted operation results and feed back real-time data to the parameter decision unit for the next round of adjustment decisions.

Images