Method for reducing wall damage by regional light control based on penetration model

Abstract

The invention belongs to the field of laser processing, and relates to a method for reducing wall damage by regional light control based on a penetration model, which comprises the following steps of: 1) constructing the penetration model; 2) interpreting the penetration model constructed in the step 1) based on a camera visual interpretation technology to obtain an interpretation result; 3) capturing an image of an actual production process in real time, comparing the image of the actual production process with the interpretation result obtained in the step 2), and sending a control signal to a light control system according to a comparison result; and (4) the light control system adjusts the distribution of the light source in time according to the control signal in the step (3), and the actual drilling process is completed. The invention provides the method for reducing the damage to the wall by regional light control based on the penetration model, which can make a judgment in time and can reduce the strict requirement on the real-time performance of the system.

Description

Technical Field

[0001] The invention belongs to the field of laser processing, and relates to a laser processing process monitoring and laser control method, and in particular to a regional light control method based on a penetration model to reduce wall damage. Background Art

[0002] Ultrafast lasers have been widely used in aerospace, electronic circuits, shipbuilding and other fine processing fields in recent years due to their high precision, high efficiency, low thermal impact and wide applicability to difficult-to-process materials. Laser processing is a non-contact processing method. After penetrating the processing surface, it will continue to act on the back of the material. There will inevitably be a cavity structure inside the ultrafast laser processed parts, and the distance between the cavities is generally small. However, during the laser processing process, the energy at the center of the laser beam spot is high, and the heat accumulates quickly, which will result in sufficient laser energy within a certain range of the processing direction. During the processing, if the high-energy laser beam spot passes through the cavity and directly acts on the surface of the cavity wall, it is very likely to cause the removal of the wall material and damage the wall. In severe cases, the entire part will be scrapped. In order to avoid the problem of wall damage, the traditional protection technology is mainly to fill the cavity with material to block the laser from acting on the wall material, but this method is difficult to achieve protection for small cavities, and the removal of the filling material after processing affects the overall processing efficiency of the part. Therefore, in recent years, a method of monitoring the processing process has emerged to achieve wall protection. The moment of penetration is detected, the laser is stopped immediately to prevent damage to the wall. However, this method has extremely high real-time requirements, and the ultrafast laser action time is extremely short, so it is easy to cause wall damage during the process of controlling the laser power reduction.

[0003] Mei Xuesong et al. proposed a combined protection method and system for the back wall of laser processing based on interlayer differences. The method processes on the target material filled with protective materials, collects monitoring signals during the processing, extracts monitoring signal features from the monitoring signals, inputs the obtained monitoring signal features into the interlayer state decision model, and then outputs the processing state. If the state is that the laser is on the protective material, the laser processing parameters are fed back to adjust the target material, and the laser is shut down before the laser penetrates the protective material. This method establishes a feature input interlayer state decision model, which is established using machine learning methods such as support vector machines, random forests, and convolutional neural networks. This method relies on the processing of real-time monitoring signals and complex machine learning models to decide the processing state, which requires the system to have strong data processing capabilities and high-performance computing platforms, increasing the complexity and cost of the system. In addition, model training requires a large amount of sample data and a sophisticated parameter adjustment process to ensure that the model can accurately identify different processing states. Data under different materials and processing conditions may differ. If the model has poor generalization ability, it may lead to misjudgment or inaccurate regulation in practical applications. Summary of the invention

[0004] In order to solve the above technical problems existing in the background technology, the present invention provides a method for alleviating wall damage by controlling regional light based on a penetration model, which can make timely judgments and reduce the strict requirements on the real-time performance of the system.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for reducing wall damage by controlling regional light based on a penetration model, characterized in that the method for reducing wall damage by controlling regional light based on a penetration model comprises the following steps:

[0007] 1) Construct a penetration model;

[0008] 2) Interpreting the penetration model constructed in step 1) based on camera vision interpretation technology to obtain interpretation results;

[0009] 3) capturing images of the actual production process in real time, and comparing the images of the actual production process with the interpretation results obtained in step 2), and sending a control signal to the light control system according to the comparison results;

[0010] 4) The light control system timely adjusts the distribution of the light source according to the control signal of step 3) to complete the actual hole making process.

[0011] The specific implementation of the above step 1) is:

[0012] 1.1) Collect data from previous penetration processes;

[0013] 1.2) A mathematical model between the penetration process and the machining process is established based on the data collected in step 1.1), wherein the mathematical model between the penetration process and the machining process is a penetration model; the penetration model is a penetration process under the joint action of multiple variables; the variables include the number of layers, the inclination angle of the workpiece, the aperture size and specific process parameters.

[0014] The specific method of establishing the mathematical model in the above step 1.2) is:

[0015] 1.2.1) establishing the relationship between the penetration process and the number of processing layers based on the data collected in step 1.1);

[0016] 1.2.2) Establish the relationship between the penetration process and the tilt angle of the workpiece based on step 1.2.1);

[0017] 1.2.3) Establish the relationship between the penetration process and the pore size based on step 1.2.2);

[0018] 1.2.4) Based on step 1.2.3), the relationship between the penetration process and the process parameters is established, and finally a mathematical model between the penetration process and the processing process is obtained.

[0019] The specific implementation of the above step 2) is:

[0020] 2.1) using a camera to capture images during the actual penetration process in real time; the images during the actual penetration process include images at the beginning of the actual penetration process and images at the complete penetration process;

[0021] 2.2) The images of the actual penetration process captured in step 2.1) are processed and analyzed by a deep learning training model to obtain a judgment result, wherein the judgment result is the starting penetration layer of the penetration model and the complete penetration layer of the penetration model.

[0022] In the above step 2.2), the deep learning model is an image recognition model established by combining support vector regression, decision tree regression, long short-term memory neural network or Transformer;

[0023] The image processing method in step 2.2) is edge detection, pattern recognition or dynamic analysis;

[0024] The initial penetration layer in step 2.2) is determined based on tiny cracks, color changes and / or brightness changes on the surface of the material.

[0025] The specific implementation of the above step 3) is:

[0026] 3.1) Capture images of the actual production process in real time;

[0027] 3.2) Compare the image of the actual production process with the interpretation result obtained in step 2) to obtain a comparison result;

[0028] 3.3) If the comparison result is the initial penetration layer of the penetration model, a starting penetration signal is sent to the light control system; if the comparison result is the complete penetration layer of the penetration model, a complete penetration signal is sent to the light control system.

[0029] The light control system comprises a controller and a laser connected to the controller.

[0030] The specific implementation of the above step 4) is:

[0031] 4.1) The controller receives the control signal sent by step 3);

[0032] 4.2) Convert the control signal into an electrical signal suitable for the laser through the driving circuit;

[0033] 4.3) According to the electrical signal, adjust the output power and wavelength of the laser, control the light-on or light-off state of the laser, and adjust the light source distribution according to the penetration state, and finally the actual hole making process.

[0034] The penetration states in the above step 4.3) are an initial penetration state and a complete penetration state.

[0035] The specific implementation method of adjusting the light source distribution according to the penetration state in the above step 4.3) is:

[0036] When the controller receives the initial penetration signal, it is determined to be the initial penetration state, and the controller controls the laser to continue working until the controller receives the complete penetration signal, and then turns off all light sources;

[0037] When the controller receives the full penetration signal and determines the full penetration state, the controller controls the laser to stop working and turns off all light sources.

[0038] The advantages of the present invention are:

[0039] The present invention provides a method for reducing wall damage by regional light control based on a penetration model. Compared with the prior art, the present invention adopts a more direct and efficient strategy. It no longer relies on a deep learning model to continuously identify the state of the entire processing process, but focuses on two key penetration moments: initial penetration and complete penetration. The camera can capture the image changes during the processing process in real time, and use image processing technology to accurately judge the two penetration signals. Once the camera determines the occurrence of initial penetration or complete penetration, it will immediately send a signal to the light control system. Since this signal is sent after the penetration is determined, there is no need to continuously identify and feedback the state like a deep learning model, thereby greatly reducing the system's pressure on real-time performance. In general, the present invention simplifies the feedback mechanism and shifts the focus from continuous monitoring of the entire processing process to accurate judgment of key penetration events, effectively reducing the system's direct dependence on real-time performance, while still being able to perform regional light control according to the processing status to reduce wall damage. DETAILED DESCRIPTION

[0040] The present invention provides a method for reducing wall damage by controlling light in a region based on a penetration model, comprising the following steps:

[0041] 1) The penetration model is derived from the qualitative description of the penetration process based on the actual penetration situation; the penetration model is to collect data of the penetration process and establish a mathematical model between the penetration and processing process based on the collected data; the penetration model describes the penetration process under the joint action of multiple variables, including the number of layers, the tilt angle of the workpiece, the aperture size, and specific process parameters. Through this model, it is possible to systematically describe how the penetration trend affects the specific location of the penetration area and the size of the penetration area.

[0042] Among them, when establishing the penetration model, the specific method of the present invention is:

[0043] In the process of studying the penetration process and its regional changes, the key link of data collection should be paid attention to first. In order to ensure the accuracy and reliability of the data, a systematic iterative method is adopted, based on the relationship between the penetration situation and the number of layers, and the data collection work is gradually carried out. Specifically, the equipment parameters, process parameters and workpiece angles are fixed, and the holes with the same parameters are repeatedly processed many times, and the number of layers is gradually increased each time until complete penetration is achieved. After each processing is completed, the distribution of the penetration area is accurately measured and the data is recorded. The specific establishment method of the mathematical model adopted by the present invention is: first, based on the data collected in step 1.1), the relationship between the penetration process and the number of processing layers is established; secondly, on the basis of the above, the relationship between the penetration process and the inclination angle of the workpiece is established; then, the relationship between the penetration process and the aperture is established; the relationship between the penetration process and the process parameters is established, and finally the mathematical model between the penetration process and the processing process is obtained. In order to eliminate the influence of random factors that may exist in the processing process on the data, it is necessary to implement multiple rounds of testing to improve the representativeness and statistical significance of the data by increasing the sample size. Subsequently, the collected data is deeply processed by using statistical analysis methods to calculate the probability distribution of the penetration area of ​​each layer, thereby revealing the intrinsic connection between the penetration process and the number of layers.

[0044] After collecting all the data, we first need to group the data according to the number of processing layers to ensure that each group of data corresponds to the penetration of the same number of layers, then calculate the mean and variance of the penetration area of ​​each layer, obtain the basic distribution of the penetration area, and finally perform probability distribution fitting. For example, the penetration area distribution can be described by normal distribution. Assuming that the size of the penetration area of ​​the i-th layer is X i Normal distribution where μ i is the mean, is the variance. For each layer i, its μ needs to be estimated i and variance

[0045] The mean is calculated as follows:

[0046]

[0047] in:

[0048] n i is the sample size of the i-th layer;

[0049] x ij is the penetration area size of the jth sample in the i-th layer.

[0050] The variance calculation is as follows:

[0051]

[0052] For a normal distribution, the probability density function is:

[0053]

[0054] The probability distribution of the penetration area of ​​each layer is calculated and used to describe the membership domain and confidence of the penetration area in subsequent modeling, while revealing the intrinsic connection between the penetration process and the number of layers.

[0055] Since the penetration process is not only related to the number of processing layers, but also to the workpiece angle, it is necessary to further test the relationship between penetration and workpiece angle on the basis of understanding the relationship between penetration and number of layers, so as to make the established mathematical model more universal. According to the actual needs of blade processing, the inclination angle of the workpiece is adjusted, and the changes in the penetration area of ​​each layer at different inclination angles are recorded in detail. Through comparative analysis, the influence of the workpiece angle on the size of the penetration area is obtained, thereby ensuring that the mathematical model established subsequently can adapt to different workpiece angles. Next, according to the same logic, the relationship between penetration and aperture and process parameters is tested in turn. In each test, other variables need to be strictly controlled to ensure that the influence of the target factors on the penetration process can be accurately captured. Through these tests, the sizes of penetration areas of different layers at different angles are obtained, and these penetration area values ​​are used to directly construct a functional relationship associated with the processing process. When choosing a suitable function form, it is necessary to decide whether the value of the penetration area changes linearly or nonlinearly according to the change characteristics of the value of the penetration area with the processing process. Finally, a mathematical model that can fully describe the relationship between the value of the penetration area size and the processing process at different processing stages is constructed. The model not only takes into account multiple influencing factors such as the number of layers, workpiece angle, hole diameter and process parameters, but also reveals the interaction mechanism among them.

[0056] In material penetration technology, especially when it involves multi-layer materials or complex structures, the penetration process often has great uncertainty and randomness. This uncertainty is mainly reflected in the non-fixed number of starting penetration layers, that is, the penetration may start from any layer of the material, depending on many factors, such as the physical properties of the material, the power of the laser, and the operating conditions.

[0057] The penetration model is to describe the penetration process by obtaining the size of the hole penetration area after each layer is processed from the initial penetration to the complete penetration (which can be obtained by two-dimensional detection or similar detection methods) to establish a model. However, in fact, the parameters that affect the penetration model are not only the number of layers required for penetration, but also the aperture, angle, and process parameters. Therefore, it is necessary to obtain the distribution of the penetration area under the number of layers, aperture, angle and process parameters, and establish a penetration model based on the distribution of the penetration area and the processing progress.

[0058] 2) Introduce camera visual interpretation, use the camera to judge the initial penetration layer and the complete penetration layer, and provide a penetration signal for light control in the open area;

[0059] In order to cope with the uncertainty in the hole-making process and improve the accuracy and control precision of the penetration model, camera vision interpretation technology was introduced to visually interpret the starting penetration layer and the complete penetration layer to provide signals for opening the area light control.

[0060] In step 2, visual interpretation uses a camera to capture images of the penetration process in real time, processes and analyzes the images through a deep learning training model to achieve automatic interpretation of the processing process, and finally outputs the interpretation results to the light control system. The deep learning model is combined with support vector regression, decision tree regression, long short-term memory neural network or Transformer to establish an image recognition model. This technology uses a camera to capture images of the processing process in real time, and processes and analyzes the visualized images through an image processing algorithm. The image processing algorithm is the core of this technology, using image analysis techniques such as edge detection, pattern recognition, and dynamic analysis to identify and locate key events in the penetration process. Specifically, the image algorithm will first detect and identify the image features of the starting penetration layer, such as tiny cracks on the surface of the material, color or brightness changes, etc. These features mark the beginning of the penetration process and are sent to the regional light control system as the starting penetration signal. After receiving the starting penetration signal, the regional light control system will automatically adjust the distribution and switching state of the light source according to the preset penetration model. This process is dynamic and aims to optimize the light source distribution according to the real-time changes in the penetration process. As the penetration process progresses, the camera continues to capture images and continuously monitors the penetration status through image processing algorithms. Among them, the image processing algorithm uses deep learning. First, a large amount of image data is collected and annotated, and then a suitable convolutional neural network model is constructed to set the network structure and parameters. Next, the model is trained using the preprocessed image data, the prediction results are calculated through forward propagation, and the model parameters are optimized using back propagation to reduce the prediction error. After training, the model performance is evaluated on an independent test set and adjusted and optimized as needed. Finally, the trained model is used to recognize new images. When it is judged to be fully penetrated, a full penetration signal is issued. This signal is then sent to the regional light control system, triggering it to turn off all light sources, marking the end of the hole making process.

[0061] The penetration signal in step 2 is a software signal and is sent to the control system.

[0062] 3) After receiving the penetration signal, the light control system is triggered to adjust the distribution of the light source according to the penetration state;

[0063] The light control system in step 3 consists of a controller and a laser; the system can receive and analyze instructions from the outside or inside. Based on the analysis results, the controller generates control signals, which are then converted into electrical signals suitable for the laser through the drive circuit. These electrical signals act directly on the laser to adjust key parameters such as output power and wavelength, thereby controlling the laser's light-on or light-off state and adjusting the light source distribution according to the penetration state.

[0064] 4) After obtaining the complete penetration signal, turn off all lights and the hole making is completed.

[0065] The penetration state in step 4 is the initial penetration state and the full penetration state; the light source distribution during the penetration process is adjusted according to the penetration model, the light is turned off in the penetrated area, and the light is turned on only in the unpenetrated area until the full penetration signal is received, and the hole making is completed. The full penetration signal is when the visual algorithm judges that it is full penetration, and the signal is sent to the light control system, turning off all light sources, and the hole making is completed.

Claims

1. A method for regional light control to reduce wall damage based on a penetration model, characterized in that: The method for regional light control to reduce wall damage based on the penetration model comprises the following steps: 1) Construct a penetration model; 2) Interpreting the penetration model constructed in step 1) based on camera vision interpretation technology to obtain interpretation results; 3) capturing images of the actual production process in real time, and comparing the images of the actual production process with the interpretation results obtained in step 2), and sending a control signal to the light control system according to the comparison results; 4) The light control system timely adjusts the distribution of the light source according to the control signal of step 3) to complete the actual hole making process.

2. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 1, characterized in that: The specific implementation method of step 1) is: 1.1) Collect data from previous penetration processes; 1.2) A mathematical model between the penetration process and the machining process is established based on the data collected in step 1.1), wherein the mathematical model between the penetration process and the machining process is a penetration model; the penetration model is a penetration process under the joint action of multiple variables; the variables include the number of layers, the inclination angle of the workpiece, the aperture size and specific process parameters.

3. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 2, characterized in that: The specific method of establishing the mathematical model in step 1.2) is: 1.2.1) establishing the relationship between the penetration process and the number of processing layers based on the data collected in step 1.1); 1.2.2) Establish the relationship between the penetration process and the tilt angle of the workpiece based on step 1.2.1); 1.2.3) Establish the relationship between the penetration process and the pore size based on step 1.2.2); 1.2.4) Based on step 1.2.3), the relationship between the penetration process and the process parameters is established, and finally a mathematical model between the penetration process and the processing process is obtained.

4. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 3 is characterized in that: The specific implementation method of step 2) is: 2.1) using a camera to capture images during the actual penetration process in real time; the images during the actual penetration process include images at the beginning of the actual penetration process and images at the complete penetration process; 2.2) The images of the actual penetration process captured in step 2.1) are processed and analyzed by a deep learning training model to obtain a judgment result, wherein the judgment result is the starting penetration layer of the penetration model and the complete penetration layer of the penetration model.

5. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 4, characterized in that: In the step 2.2), the deep learning model is established by combining support vector regression, decision tree regression, long short-term memory neural network or Transformer to obtain an image recognition model; The image processing method in step 2.2) is edge detection, pattern recognition or dynamic analysis; The initial penetration layer in step 2.2) is determined based on tiny cracks, color changes and / or brightness changes on the surface of the material.

6. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 5, characterized in that: The specific implementation method of step 3) is: 3.1) Capture images of the actual production process in real time; 3.2) Compare the image of the actual production process with the interpretation result obtained in step 2) to obtain a comparison result; 3.3) If the comparison result is the initial penetration layer of the penetration model, a starting penetration signal is sent to the light control system; if the comparison result is the complete penetration layer of the penetration model, a complete penetration signal is sent to the light control system.

7. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 6, characterized in that: The light control system comprises a controller and a laser connected to the controller.

8. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 7, characterized in that: The specific implementation method of step 4) is: 4.1) The controller receives the control signal sent by step 3); 4.2) Convert the control signal into an electrical signal suitable for the laser through the driving circuit; 4.3) According to the electrical signal, adjust the output power and wavelength of the laser, control the light-on or light-off state of the laser, and adjust the light source distribution according to the penetration state, and finally the actual hole making process.

9. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 8, characterized in that: The penetration states in step 4.3) are an initial penetration state and a complete penetration state.

10. The method for reducing wall damage by controlling light in a region based on a penetration model according to claim 9, characterized in that: The specific implementation method of adjusting the light source distribution according to the penetration state in step 4.3) is: When the controller receives the initial penetration signal, it is determined to be the initial penetration state, and the controller controls the laser to continue working until the controller receives the complete penetration signal, and then turns off all light sources; When the controller receives the full penetration signal and determines the full penetration state, the controller controls the laser to stop working and turns off all light sources.

Images