Laser processing real-time defect monitoring device and method based on scattered light detection

Abstract

The invention belongs to the related technical field of laser processing quality detection, and discloses a laser processing real-time defect monitoring device based on scattered light detection, which comprises a scattered light detection module, a signal processing module and a display early warning module, the detection module is used for realizing a scattered light detection function of focusing, filtering and attenuating; the signal processing module carries out various processing and enhancement on the detection signal based on a coordinate attention mechanism so as to obtain a defect judgment result; and the display early warning module is used for displaying data such as defect judgment results and the like, and giving alarms of different levels according to the severity of the defects. The invention further discloses a corresponding monitoring method. According to the invention, real-time and accurate identification of surface layer and subsurface layer defects in the laser processing process can be realized, so that the method is suitable for providing a more efficient and reliable solution for laser processing quality control in the fields of semiconductors, optical elements, aerospace and the like.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of laser processing quality detection, and more particularly relates to a laser processing real-time defect monitoring device and method based on scattered light detection. BACKGROUND

[0002] In the field of precision machining, laser has been widely used due to its characteristics such as high energy focusing, non-contact processing and material universality. However, during the processing, due to factors such as laser power fluctuation, material property difference and processing parameter matching degree, defects such as cracks, pits, micro-holes and subsurface damage layer are easily produced. These defects not only reduce the mechanical and optical properties of the workpiece, but also may cause failure risk in subsequent use.

[0003] In the prior art, the detection methods of laser processing defects mainly include offline detection and online detection. For the offline detection technology, the workpiece needs to be transferred to the detection equipment after processing, which has the problems of detection lag, inability to adjust the processing parameters in time, high detection cost, low efficiency and difficulty in meeting the demand of large-scale batch production. For the online detection technology, it is mainly based on imaging principle, which shoots the processing area image through an industrial camera, and then judges the defects through image recognition algorithm.

[0004] However, further research shows that the existing various online detection technologies still have the following defects or deficiencies: first, the imaging quality is easily disturbed by the plasma flash and debris generated during the processing, resulting in low defect recognition accuracy; second, it is difficult to detect subsurface defects because the subsurface defects cannot be directly presented through surface imaging; third, in order to avoid damage to the camera by strong light, a complex light shielding structure is needed, which increases the equipment volume and cost.

[0005] Correspondingly, there is an urgent need in the art to improve and improve the online detection scheme for laser processing defects in order to better meet the higher comprehensive quality requirements of various precision machining. SUMMARY

[0006] In order to solve one or more of the above defects or needs of the prior art, the present application provides a laser processing real-time defect monitoring device and method based on scattered light detection, wherein the key component modules such as the scattered light detection module and the signal processing module are redesigned in terms of specific component structure and working mechanism, which can collect a wide range of scattered light caused by defects compared with the prior art, realize real-time and accurate identification of surface and subsurface defects in the laser processing process, and provide integrated detection functions of focusing, filtering and attenuation, and have the characteristics of compact structure, easy operation and good adaptability. In addition, by combining a deep learning algorithm, the characteristics of the scattered light can be automatically extracted and learned, further improving the efficiency, accuracy and work efficiency, and thus the device is especially suitable for real-time defect monitoring applications such as femtosecond laser processing.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a laser processing real-time defect monitoring device based on scattered light detection is provided, which comprises a scattered light detection module, a signal processing module and a display warning module, characterized in that: The scattered light detection module is arranged on the side of the laser processing platform, and comprises a focusing lens, a filter, an attenuation sheet and an industrial CCD camera which are integrally designed. The focusing lens is fixed by a lens holder with a locking structure, and is used to converge the surface scattered light and subsurface scattered light during laser processing. The filter and the attenuation sheet are installed in the lens sleeve, and one side of the lens sleeve is threadedly connected with the lens holder. The filter only allows the scattered light to enter the photosensitive target surface of the industrial CCD camera, and the attenuation sheet is used to reduce the intensity of the scattered light to within the light intensity control safety range of the industrial CCD camera. The industrial CCD camera is installed on the other side of the lens sleeve, and is used to collect the scattered light signals and transmit them to the signal processing module in real time. The signal processing module comprises a signal preprocessing unit, a feature extraction unit and a defect judgment unit. The signal preprocessing unit is used to preprocess the real-time received scattered light original signal by denoising and normalization. The feature extraction unit is used to extract the features of the preprocessed scattered light signal, and to enhance some key features of the scattered light by applying a coordinate attention mechanism, and to output a multi-scale feature map. The defect judgment unit judges based on the multi-scale feature map, and obtains the judgment results including defect positioning and existence, defect type classification and defect level evaluation. The display warning module is used to display the defect judgment results and the scattered light signal in real time, and to alarm at different levels according to the severity of the defects.

[0008] As a further preferred, for the said scattered light detection module, it is preferably mounted in the vicinity of the workpiece through a three-dimensional adjustable support, and can be freely adjusted in space pose according to the working condition.

[0009] As a further preferred, for the said focusing lens, its focal length and minimum diameter are preferably designed in the following way, thereby ensuring the convergence of most of the scattered light scattered during laser processing: The focal length of the focusing lens is preferably selected according to the emission angle of the processable laser, the size of the photosensitive target surface of the industrial CCD camera and the mounting distance; The minimum diameter D of the focusing lens is preferably designed according to the formula , wherein, represents the mounting distance between the focusing lens and the workpiece, represents the minimum measurement half-angle that needs to be reached by the scattered light.

[0010] As a further preferred, for the said band-pass filter, it is preferably a replaceable band-pass filter, and the center wavelength is consistent with the wavelength of the processing laser, with a bandwidth control within 5nm, and a stray light suppression ratio greater than 10000:1.

[0011] As a further preferred, for the said attenuator, it is preferably an absorption type medium density attenuator, and its attenuation ratio is set to more than 100,000 times.

[0012] As a further preferred, for the said feature extraction unit, in the process of applying the coordinate attention mechanism, it is preferably enhanced for some scattered light key features related to the spatial directivity features of the defect morphology profile, the spatial distribution density difference features of the defect scattered light spot, and the weak signal spatial positioning features of the subsurface defects.

[0013] As a further preferred, for the said feature extraction unit, it preferably uses MobileNetV3-Large as the basic feature extraction method, and then enhances some scattered light key features by applying the coordinate attention mechanism, corresponding to output 3 sizes of feature maps.

[0014] As a further preferred, for the defect judgment unit, it is preferably further divided into three branches, wherein the first branch, i.e. the defect monitoring branch, preferably adopts YOLOv8-Nano lightweight detection head, and realizes the defect positioning and existence judgment of laser processing through CIoU loss function; the second branch, i.e. the defect classification branch, preferably adopts a fully connected layer plus a Softmax activation function to classify and identify multiple defects; the third branch, i.e. the defect level evaluation branch, preferably adopts a regression type fully connected layer to quantize the defect level into multiple levels, and the level evaluation is based on the defect size, feature intensity and morphological integrity.

[0015] As a further preferred, for the signal processing module, it preferably further includes a model iteration unit for storing various types of historical data to construct an incremental learning data set, and preferably adopts an incremental learning algorithm based on knowledge distillation to adjust regularly, thereby ensuring the long-term adaptability of the entire signal processing module.

[0016] As a further preferred, for the display warning module, it preferably includes a display screen unit and a multi-level sound and light alarm unit, wherein the display screen unit is used to display the defect judgment result and the scattered light signal in real time, and the multi-level sound and light alarm unit is used to trigger different levels of alarm according to the defect severity.

[0017] As a further preferred, for the multi-level sound and light alarm unit, it preferably triggers two levels of alarm according to the defect severity, wherein the first level alarm is triggered when the defect is relatively mild, in which case the indicator light flashes and the buzzer alarms intermittently; the second level alarm is triggered when the defect is relatively serious, in which case the indicator light is always on and the buzzer alarms continuously, while sending a signal to the laser processing equipment to trigger the processing pause or automatic parameter adjustment to avoid the defect from expanding.

[0018] According to another aspect of the present application, a corresponding laser processing real-time defect monitoring method is also provided.

[0019] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages: 1. By redesigning the specific component structure and working mechanism of some key component modules such as the scattered light detection module and the signal processing module, the present application can collect a wide range of scattered light caused by defects, realize real-time and accurate identification of surface and subsurface defects during laser processing, and effectively solve the technical pain points of strong defect detection lag, difficulty in identifying subsurface damage, and weak anti-interference ability in existing laser precision machining. 2、The application also designs a coordinate attention mechanism to enhance some key features of scattered light, and combines a deep learning model to automatically extract and learn these features, so that the real-time monitoring of surface and subsurface defects can be performed, and the accurate identification and active early warning of various defects can be realized during the whole processing process, and various specific judgment results can be output, thereby providing reliable guidance information for subsequent processing; 3、The application further optimizes the key performance indicators of the focusing lens, band-pass filter and attenuation sheet, so that the focusing-filtering-attenuation integrated detection function of the scattered light detection module can be better performed, and the structure is compact, easy to control and good in adaptability; 4、The application further optimizes the algorithm processing of the feature extraction unit and the defect judgment unit, so that the detection equipment and the processing process can work synchronously, the defects can be identified and alarmed immediately when they occur, the processing parameters can be adjusted in time by the operator, batch defects can be avoided, and the processing qualification rate is significantly improved; 5、The whole detection and monitoring device of the application does not need complex imaging light paths and shielding structures, the influence of interference factors on the detection result is reduced, the core components are all conventional industrial components, easy to purchase and integrate, the device is small in size and flexible in installation, and can be adapted to different types of laser processing equipment, and thus is especially suitable for providing efficient and reliable solutions for laser processing quality control in the fields of semiconductors, optical elements and aerospace. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is the overall structure schematic diagram of the laser processing real-time defect monitoring device based on scattered light detection according to the application; Figure 2 It is a schematic diagram for more specifically showing the scattered light detection module according to one preferred embodiment of the application; Figure 3 It is a work flow diagram for exemplarily showing the signal processing module according to the application; Figure 4 It is a work flow diagram for exemplarily showing the display warning module according to the application; In all the drawings, the same reference signs are used to represent the same elements or structures, wherein: 101-laser processing platform; 102-workpiece; 103-scattered light detection module; 104-three-dimensional adjustable support; 105-signal processing module; 106-display warning module; 201-focusing lens; 202-lens holder; 203-filter; 204-attenuation sheet; 205-lens sleeve; 206-industrial CCD camera. DETAILED DESCRIPTION

[0021] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0022] Figure 1 is the overall structure schematic diagram of the laser processing real-time defect monitoring device based on scattered light detection according to the present application, and the present application will be explained more specifically below with reference to Figure 1 .

[0023] The monitoring device mainly includes functional modules such as a scattered light detection module 103, a signal processing module 105 and a display warning module 106, and some key component modules such as the scattered light detection module 103 and the signal processing module 105 are redesigned in terms of specific component structure and working mechanism, etc. Compared with the prior art, a wide range of scattered light caused by defects can be collected, real-time and accurate identification of surface and subsurface defects in the laser processing process can be realized, and technical pain points such as strong defect detection hysteresis, difficulty in identifying subsurface damage and weak anti-interference ability in the existing laser precision machining are effectively solved.

[0024] As shown in Figure 1 , and referring to Figure 2 , the scattered light detection module 103 is arranged on the side of the laser processing platform 101, and includes a focusing lens 201, a filter 203, an attenuation sheet 204 and an industrial CCD camera 206 designed in an integrated manner. The focusing lens 201 is fixed by a lens holder 202 with a locking structure, which is used to converge the surface scattered light and subsurface scattered light during the laser processing process. The filter 203 and the attenuation sheet 204 are installed in a lens sleeve 205, and one side of the lens sleeve 205 is threadedly connected with the lens holder 202. The filter 203 only allows the scattered light to enter the photosensitive target surface of the industrial CCD camera 206, and the attenuation sheet 204 is used to reduce the intensity of the scattered light to within the light intensity control safety range of the industrial CCD camera 206. The industrial CCD camera 206 is installed on the other side of the lens sleeve 205, which is used to collect the scattered light signal reaching here and transmit it to the signal processing module in real time.

[0025] More specifically, the industrial CCD camera 206 may, for example, be a high-sensitivity, low-noise area array industrial CCD camera, and needs to have a high frame rate and a large dynamic range. The high frame rate ensures synchronization with the laser processing pulses, the high dynamic range can simultaneously capture strong surface scattering light and subsurface scattering light, and the low noise characteristic improves weak signal detection accuracy.

[0026] The focusing lens 201 is disposed on the optical path between the industrial CCD camera 206 and the processing area, and is located in front of the optical path filtering unit. The focal length and size of the focusing lens need to be able to converge a large range of scattering light diverging during laser processing to the target surface of the CMOS photosensitive chip of the industrial CCD camera, solving the problem of insufficient collection range caused by scattering light divergence, for example, increasing the effective collection area by 3-5 times.

[0027] Accordingly, according to one preferred embodiment of the present application, for the focusing lens 201, its focal length and minimum diameter are preferably designed as follows, thereby ensuring that most of the scattering light diverging during laser processing is converged: the focal length of the focusing lens is preferably selected according to the emission angle of the processable laser, the size of the photosensitive target surface of the industrial CCD camera, and the mounting distance; in addition, the minimum diameter D of the focusing lens is preferably designed according to the formula , wherein, represents the mounting distance between the focusing lens and the workpiece, represents the minimum measurement half-angle that needs to be reached by the scattering light.

[0028] In addition, since the focusing lens 201 is fixed in the present application by the lens holder 202 with a locking structure, which is connected to the lens sleeve 205 of the subsequent optical path filtering unit by a threaded connection, in this way the coaxiality of the optical path can be ensured to be ≤0.1mm.

[0029] More specifically, in front of the CMOS photosensitive chip of the industrial CCD camera 206, a detachable optical filter 203 and an attenuation sheet 204 are fixed in sequence along the direction of propagation of the optical path, both of which are integrated with the camera by a threaded lens sleeve 205, facilitating disassembly and replacement.

[0030] According to one preferred embodiment of the present application, the optical filter 203 is a replaceable bandpass filter with a center wavelength consistent with that of the laser and a bandwidth controlled within 5nm, and a stray light suppression ratio greater than 10000:1, for filtering stray light during processing and allowing only scattering light produced by the interaction of the laser and the workpiece to enter the CMOS photosensitive chip.

[0031] According to another preferred embodiment of the present application, the attenuation sheet 204 is an absorption type medium density attenuation sheet, and the attenuation ratio is 100,000 times, which can stably reduce the intensity of scattered light generated by high-power laser, control the light intensity entering the CMOS within a safe range, and effectively avoid damage to the industrial CCD camera caused by strong scattered light.

[0032] In addition, the above-mentioned scattered light detection module is fixed on the side of the laser processing platform 101 by the three-dimensional adjustable support 104, and is preferably arranged within 20 cm from the processing area, so that most of the scattered light can be ensured to be within the CCD acquisition range.

[0033] Referring back to Figure 1 , the signal processing module 105 includes a signal preprocessing unit, a feature extraction unit, and a defect judgment unit, etc., wherein the signal preprocessing unit is used for pre-processing the real-time received scattered light original signal, such as noise reduction and normalization; the feature extraction unit is used for feature extraction on the pre-processed scattered light signal, and some key features of the scattered light are enhanced by applying a coordinate attention mechanism, and a multi-scale feature map is correspondingly output; the defect judgment unit judges based on the multi-scale feature map, and obtains a judgment result including defect positioning and existence, defect type classification, and defect level evaluation. In addition, the display warning module 106 is used for real-time display of the defect judgment result and the scattered light signal, and different levels of alarms are generated according to the defect severity.

[0034] According to another preferred embodiment of the present application, the signal processing module 105 preferably further includes a model iteration unit, which is used for storing various types of historical data to construct an incremental learning data set, and preferably uses an incremental learning algorithm based on knowledge distillation to adjust periodically, thereby ensuring the long-term adaptability of the entire signal processing module.

[0035] More specifically, according to another preferred embodiment of the present application, for the feature extraction unit, it preferably enhances some key features of the scattered light in the process of applying the coordinate attention mechanism, such as the spatial directionality feature of the defect morphology profile, the spatial distribution density difference feature of the defect scattered light spot, and the weak signal spatial positioning feature of the subsurface defect.

[0036] According to another preferred embodiment of the present application, for the feature extraction unit, it preferably uses MobileNetV3-Large as the basic feature extraction method, and then enhances some key features of the scattered light by applying the coordinate attention mechanism, and correspondingly outputs a feature map of 3 sizes.

[0037] According to another preferred embodiment of the present application, for the defect judgment unit, it is preferably further divided into three branches, wherein the first branch, i.e. the defect monitoring branch, preferably adopts a YOLOv8-Nano lightweight detection head and realizes the defect positioning and existence judgment of laser processing through a CIoU loss function; the second branch, i.e. the defect classification branch, preferably adopts a fully connected layer plus a Softmax activation function to classify and identify multiple defects; and the third branch, i.e. the defect level evaluation branch, preferably adopts a regression type fully connected layer to quantize the defect level into multiple levels, and the level evaluation is based on the defect size, feature intensity and morphological integrity.

[0038] In addition, for the display warning module 106, it preferably includes a display screen unit and a multi-level sound-light alarm unit, wherein the display screen unit is used to display the defect judgment result and the scattered light signal in real time, and the multi-level sound-light alarm unit is used to trigger different levels of alarm according to the defect severity.

[0039] More specifically, for the multi-level sound-light alarm unit, it preferably triggers two levels of alarm according to the defect severity, wherein the first level alarm is triggered when the defect is relatively slight, in which case the indicator light flashes and the buzzer alarms intermittently; and the second level alarm is triggered when the defect is relatively serious, in which case the indicator light is always on and the buzzer alarms continuously, while sending a signal to the laser processing equipment to trigger the processing pause or automatic parameter adjustment to avoid the defect expansion.

[0040] Figure 3 is a workflow diagram for demonstrating the working process of the signal processing module according to the present application. As shown in Figure 3 the signal preprocessing unit first performs noise reduction, normalization and other operations on the real-time received scattered light original signal. The feature extraction unit extracts features from the preprocessed scattered light signal, for example, MobileNetV3-Large can be used as the basic feature extraction network, and a coordinate attention module is inserted to enhance the sensitivity of the model to the key features of the scattered light, finally outputting three scale feature maps.

[0041] The defect judgment unit sets three branches of defect detection, type classification and level evaluation based on the multi-scale feature maps output by the feature extraction unit. The defect detection branch preferably adopts a YOLOv8-Nano lightweight detection head, and the boundary box regression can be optimized through a CIoU loss function to realize the defect positioning and existence judgment. The type classification branch adopts a fully connected layer plus a Softmax activation function to classify and identify multiple typical defects. The level evaluation branch adopts a regression type fully connected layer to quantize the defect severity level into multiple levels, and the evaluation is based on the defect size, feature intensity and morphological integrity. Finally, the identified results are transmitted to the display warning module.

[0042] In addition, the model iteration unit can be used to ensure long-term adaptability of the model. The unit automatically stores the "suspected defect samples" in the detection process and the samples corrected by the operator's labeling, constructs an incremental learning data set, and uses an incremental learning algorithm based on knowledge distillation to fine-tune the model parameters with new samples every month, without the need to retrain the entire model, and the fine-tuning time is ≤2 hours.

[0043] Figure 4 is a workflow diagram for demonstrating the operation of the display warning module according to the present application. As shown in Figure 4 The display warning module serves as the core of human-computer interaction and can realize real-time presentation of detection results and rapid response to abnormal conditions. When a slight defect occurs, the indicator light flashes and the buzzer alarms intermittently. When there is a serious defect, the indicator light is always on, the buzzer continuously alarms, and a signal is sent to the laser processing equipment to trigger processing pause or automatic parameter adjustment to avoid defect expansion. The defect occurs simultaneously with the display screen unit displaying the image of the scattered light and the defect judgment result, including defect type, occurrence time, position coordinates, severity, and other information.

[0044] The general process of the laser processing real-time defect monitoring method according to the present application will be described below.

[0045] Before processing starts, replace the appropriate band-pass filter and attenuator according to the workpiece 102 and the laser wavelength and power, adjust the scattered light detection module 103 to the appropriate position through the three-dimensional adjustable support 104, and may need to load a pre-trained special model from the model library according to the processing material type.

[0046] Next, the laser processing platform 101 processes the workpiece 102, and when a defect occurs at the processing site, it will cause scattered light to diverge in all directions. The scattered light detection module 103 transmits the detected original signal to the signal processing module 105, and the signal processing module 105 processes the signal in a series of processes and transmits the results to the display warning module 106 in real time.

[0047] In summary, compared with the prior art, the laser processing real-time defect monitoring scheme according to the present application can collect a wide range of scattered light caused by defects, realize real-time and accurate identification of surface and subsurface defects during laser processing, and effectively solve the technical pain points of strong defect detection lag, difficulty in identifying subsurface damage, and weak anti-interference ability in existing laser precision processing. The present application can provide an efficient and reliable solution for laser processing quality control in the fields of semiconductors, optical elements, aerospace, etc., and has good practical value and application prospect.

[0048] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A real-time defect monitoring device for laser processing based on scattered light detection, the device comprising a scattered light detection module (103), a signal processing module (105), and a display and early warning module (106), characterized in that: The scattered light detection module (103) is arranged on the upper side of the laser processing platform (101) and includes an integrated focusing lens (201), a filter (203), an attenuator (204), and an industrial CCD camera (206). The focusing lens (201) is fixed by a lens holder (202) with a locking structure to converge the surface scattered light and subsurface scattered light emitted during laser processing. The filter (203) and the attenuator (204) are installed together in the lens sleeve (205) and the lens... One side of the sleeve (205) is threadedly connected to the lens holder (202), wherein the filter (203) only allows the scattered light to enter the photosensitive target surface of the industrial CCD camera (206), and the attenuator (204) is used to reduce the intensity of the scattered light so that it reaches the light intensity control safety range of the industrial CCD camera (206); the industrial CCD camera (206) is installed on the other side of the lens sleeve (205) for collecting signals from the scattered light arriving there and transmitting them to the signal processing module in real time; The signal processing module (105) includes a signal preprocessing unit, a feature extraction unit, and a defect judgment unit. The signal preprocessing unit is used to perform noise reduction and normalization on the real-time received original scattered light signal. The feature extraction unit is used to extract features from the preprocessed scattered light signal and enhance some key features of the scattered light by applying a coordinate attention mechanism, and output a multi-scale feature map accordingly. The defect judgment unit makes judgments based on the multi-scale feature map and obtains judgment results including defect location and existence, defect type classification, and defect level assessment. The display and early warning module (106) is used to display the defect judgment results and the scattered light signal in real time, and to generate different levels of alarms according to the severity of the defect.

2. The real-time defect monitoring device for laser processing as described in claim 1, characterized in that, The scattered light detection module (103) is installed near the workpiece (102) via a three-dimensional adjustable bracket (104) and its spatial pose can be freely adjusted according to the working conditions.

3. The real-time defect monitoring device for laser processing as described in claim 2, characterized in that, The focal length and minimum diameter of the focusing lens (201) are designed in such a way that most of the scattered light emitted during laser processing is focused: The focal length of the focal lens (201) is selected according to the emission angle of the processable laser, the photosensitive target size of the industrial CCD camera (206), and the installation distance; The minimum diameter D of the focusing lens is preferably according to... The formula is used for design, where, This indicates the installation distance between the focusing lens and the workpiece. This indicates the minimum half-angle that the scattered light needs to be measured.

4. The real-time defect monitoring device for laser processing as described in claim 3, characterized in that, The bandpass filter (203) is a replaceable bandpass filter, and its center wavelength is consistent with the wavelength of the processing laser, its bandwidth is controlled within 5nm, and its stray light suppression ratio is greater than 10000:

1.

5. The real-time defect monitoring device for laser processing as described in claim 4, characterized in that, The attenuator (204) is an absorption-type medium-density attenuator, and its attenuation ratio is set to more than 100,000 times.

6. The real-time defect monitoring device for laser processing as described in any one of claims 1-5, characterized in that, For the feature extraction unit, during the application of the coordinate attention mechanism, it enhances key scattered light features related to the spatial orientation features of the defect morphology contour, the spatial distribution density difference features of the defect scattered light spot, and the weak signal spatial positioning features of the subsurface defect.

7. The real-time defect monitoring device for laser processing as described in claim 6, characterized in that, For the feature extraction unit, it uses MobileNetV3-Large as the basic feature extraction method, and then enhances some key features of scattered light by applying the coordinate attention mechanism, and outputs feature maps of three sizes accordingly.

8. The real-time defect monitoring device for laser processing as described in claim 7, characterized in that, The defect judgment unit is further divided into three branches. The first branch, the defect monitoring branch, preferably uses the YOLOv8-Nano lightweight detection head and uses the CIoU loss function to locate and determine the existence of defects in laser processing. The second branch, the defect classification branch, preferably uses a fully connected layer with a Softmax activation function to classify and identify various defects. The third branch, the defect level evaluation branch, preferably uses a regression-type fully connected layer to quantify the defect level into multiple levels, and the level evaluation criteria include defect size, feature intensity, and morphological integrity.

9. The real-time defect monitoring device for laser processing as described in any one of claims 1-8, characterized in that, The signal processing module (105) further includes a model iteration unit for storing various types of historical data to construct an incremental learning dataset, and preferably employs a knowledge distillation-based incremental learning algorithm for periodic adjustment, thereby ensuring the long-term adaptability of the entire signal processing module.

10. A method for real-time defect monitoring in laser processing, characterized in that, This method is implemented based on the apparatus as claimed in any one of claims 1-9.

Images