Laser processing detection method and equipment

Abstract

The embodiment of the invention discloses a laser processing detection method and equipment. The detection method comprises the following steps: receiving an optical radiation signal of at least one processing point in a laser processing path, wherein the optical radiation signal comprises one or more of an infrared radiation signal, a visible light radiation signal and a processing laser reflection signal; performing photoelectric conversion on the received optical radiation signal through a single-point photoelectric sensor to obtain an electric signal; establishing a corresponding relation representing the change of the electric signal corresponding to the laser processing point in the laser processing process; according to the corresponding relation and a pre-stored normal electric signal corresponding to the processing point of the laser processing standard part, determining processing point data with defects in the initial quality of the processing point of the laser processing part; wherein the normal electric signal is a corresponding changing electric signal value amplitude range in the qualified machining process of the machining point of the laser machining part under one technological parameter; and according to a pre-trained machine learning model, carrying out quality judgment on the machining point data with the defects again so as to determine the quality of the machining points of the laser machining part. According to the embodiment of the invention, an existing multispectral optical detection algorithm and an AI detection algorithm are fused, so that the over-killing number is reduced under the condition of approaching zero missed killing, and the effect of saving the production cost is achieved.

Description

Technical Field

[0001] This application relates to the technical field of laser processing, and particularly relates to a laser processing detection method and device. Background Art

[0002] Laser processing is a high-precision and high-efficiency manufacturing technology, which is widely used in fields such as automobiles, aerospace, electronics, and machinery. Laser processing detection refers to using various detection technologies and devices to monitor and evaluate the laser processing process and results to ensure processing quality and efficiency.

[0003] With the continuous development of AI (Artificial Intelligence) technology, AI technology has gradually been applied in laser processing detection. For example, in defect detection: AI can automatically detect defects in the laser processing process, such as cracks, holes, non-uniformities, etc., through image recognition and machine learning algorithms. The AI system can analyze the image or signal after laser processing, compare it with the normal processing result, and quickly identify abnormal situations. In dimension measurement: AI can be used to accurately measure the dimensions of the workpiece after laser processing. Through laser scanning and computer vision technology, the AI system can obtain the three-dimensional dimension data of the workpiece and compare it with the design drawing to automatically detect dimension deviations. In process monitoring: AI can real-time monitor various parameters in the laser processing process, such as laser power, frequency, pulse width, etc. Through the analysis of these parameters, the AI system can predict the quality of the processing result and timely adjust the processing parameters to ensure the stability and consistency of the processing process. In quality prediction: AI can predict the quality after laser processing through the analysis and modeling of historical data. Through the comprehensive analysis of factors such as processing parameters, material properties, and equipment status, the AI system can predict possible quality problems in advance and thus take preventive measures.

[0004] It can be understood that in the trend of technological development, in the future, AI can improve the accuracy and reliability of detection by learning and analyzing a large amount of data. Realize the automation of the detection process, reduce manual intervention, and improve detection efficiency. Can real-time monitor the laser processing process and timely feedback the results, which helps to discover and solve problems in a timely manner. Can also conduct in-depth analysis of the detection data, predict the processing quality, and provide support for production decision-making.

[0005] The existing problem of the current technology is that when integrating AI general technology into the analysis of laser processing detection data, the analysis result is still disjointed from the actual business requirements, and the AI input and output still cannot accurately reflect the laser processing quality. Summary of the Invention

[0006] An embodiment of the present application provides a method and device for detecting the quality of laser processing. Using the embodiment of the present application is beneficial to reducing the overkill rate of laser processing detection and reducing production costs.

[0007] The present application is implemented by the following technical solutions.

[0008] In a first aspect, an embodiment of the present application provides a method for detecting the quality of laser processing, including:

[0009] Step 100: Receive the optical radiation signals of at least one processing point in the laser processing path. The optical radiation signals include one or more of infrared radiation signals, visible light radiation signals, and processing laser reflection signals; convert the received optical radiation signals into electrical signals through a single-point photoelectric sensor; establish a correspondence relationship between the laser processing points and the changes in the electrical signals during the laser processing process; determine the processing point data with defective initial quality of the laser processing part according to the correspondence relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts; wherein, the normal electrical signal is the range of the value amplitude of the electrical signal corresponding to the qualified processing process of the laser processing part processing point under a certain process parameter.

[0010] Step 200: Perform another quality judgment on the processing point data with defects according to a pre-trained machine learning model to determine the quality of the laser processing part processing points.

[0011] It can be seen that by fusing and inheriting between the existing multi-spectral optical detection algorithm and the AI detection algorithm, the overkill number is reduced under the condition of approaching zero or setting zero missed killings, achieving the effect of saving production costs.

[0012] Combined with the first aspect, in a possible implementation manner, the specific steps for obtaining the pre-trained machine learning model include:

[0013] Step 301: Obtain unqualified processing electrical signals under defective processing process parameters by simulating the change mode of laser processing parameters.

[0014] Step 302: Process the noise of the unqualified processing electrical signals through a filtering algorithm to unify the data scale of the unqualified processing electrical signals.

[0015] Step 303: Extract time-domain and / or frequency-domain features from the unified unqualified processing electrical signal data.

[0016] Step 304: Inject the time-domain and / or frequency-domain features into the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts to obtain simulated abnormal electrical signals.

[0017] Step 305: Perform pre-training through a deep learning network based on the normal electrical signals and simulated abnormal electrical signals corresponding to the laser processing standard part processing points stored in advance; obtain a pre-trained machine learning model.

[0018] It can be seen that by pre-simulating the changes in laser processing parameters to obtain a batch of unqualified laser processing detection electrical signals, and after data alignment, extracting the time domain and / or frequency domain features of unqualified processing, and injecting the time domain and / or frequency domain features into the OK signal, a pre-trained machine learning model can be obtained based on the OK signal and the simulated NG signal.

[0019] Combined with the first aspect, in a possible implementation manner, the specific steps for obtaining the pre-trained machine learning model include:

[0020] Step 311: Determine a segment of abnormal electrical signals for one or more failure causes according to the historical pre-stored unqualified processing electrical signals and / or the on-site pre-collected unqualified processing electrical signals;

[0021] Step 312: Simulate the segment of abnormal electrical signals through one or more functions such as a constant failure function, a cosine failure function, and a variance failure function;

[0022] Step 313: Randomly select a certain segment of the normal electrical signals corresponding to the laser processing standard part processing points stored in advance;

[0023] Step 314: Inject the selected segment of the normal electrical signals in Step 313 into the simulated abnormal electrical signals of the corresponding segment selected in Step 312 to obtain simulated abnormal electrical signals;

[0024] Step 315: Perform pre-training through a deep learning network based on the normal electrical signals and simulated abnormal electrical signals corresponding to the laser processing standard part processing points stored in advance; obtain a pre-trained machine learning model.

[0025] Practice shows that in the laser processing detection application scenario, simulating the segment of abnormal electrical signals through one or more functions such as a constant failure function, a cosine failure function, and a variance failure function can effectively simulate and restore the defect characteristics of laser processing detection.

[0026] Combined with the first aspect, in a possible implementation manner, the specific steps for obtaining the pre-trained machine learning model include:

[0027] Step 321: Determine a segment of abnormal electrical signals for one or more failure causes according to the historical pre-stored unqualified processing electrical signals and / or the on-site pre-collected unqualified processing electrical signals;

[0028] Step 322: Randomly use one or more functions to simulate the abnormal electrical signals of this segment; the functions include constant fault functions, cosine fault functions, and variance fault functions.

[0029] Step 323: Randomly select a certain segment of the normal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts;

[0030] Step 324: Inject a certain segment of the normal electrical signal selected in Step 323 into the simulated abnormal electrical signal of the corresponding segment selected in Step 322 to obtain a simulated abnormal electrical signal;

[0031] Step 325: Perform pre-training through a deep learning network based on the normal electrical signal and the simulated abnormal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts; obtain a pre-trained machine learning model;

[0032] Step 326: Fine-tune the pre-trained machine learning model according to the abnormal electrical signal obtained during online processing to obtain a fine-tuned sub-model;

[0033] Step 327: Repeat Steps 321 to 326 to obtain multiple fine-tuned sub-models, and integrate the multiple fine-tuned sub-models into a unified machine learning model in an ensemble model manner.

[0034] It can be seen that in an ensemble model manner, multiple fine-tuned sub-models are integrated into a unified machine learning model. And by combining the prediction results of multiple models, better performance than a single model can be obtained.

[0035] Combined with the first aspect, in a possible implementation manner, the specific steps for performing a quality judgment on the defective processing point data again according to the pre-trained machine learning model are as follows:

[0036] Step 401: Based on the pre-trained machine learning model, perform data screening on the processing point data with defective initial quality of the laser processed parts obtained in Step 100;

[0037] Step 402: Divide the processing point data with defects in Step 100 into OK re-judgment data and NG re-judgment data, and determine that the OK re-judgment data meets the laser processing quality requirements.

[0038] It can be seen that through the method of this embodiment, by performing a quality judgment on the defective processing point data again through a pre-trained machine learning model, the re-verification of the laser processing NG data can be achieved, and the difficulty of setting the previous feature threshold can be reduced, saving the workload of personnel. Through the setting of the machine learning model, the objectivity and result consistency of the subsequent laser processing quality judgment are ensured, the accuracy of the detection result obtained by the weld seam detection is improved, the practical problem of high overkill rate under the conventional feature threshold setting algorithm is avoided, and the stability of the detection result is ensured.

[0039] In combination with the first aspect, in a possible implementation manner, in step 100, according to the corresponding relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts, determine the processing point data with defective preliminary quality of the laser processing part; it further includes the following steps:

[0040] Receive the external defect data of the processing points of the laser processing part transmitted back externally, and after merging the external defect data with the processing point data with defective preliminary quality, form the processing point data with defects to be analyzed in step 200.

[0041] It can be seen that in the laser on-line detection, the basic data is updated in real time for the process of re-judging the quality through a pre-trained machine learning model. In view of the requirement of many variable factors in the complex optoelectronic scenario of laser processing, the data is updated in time to improve the accuracy of the re-judgment of the quality.

[0042] In combination with the first aspect, in a possible implementation manner, step 100 further includes the following steps:

[0043] Establish a corresponding relationship representing the change of the electrical signal corresponding to the laser processing point during the laser processing process; according to the corresponding relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts, determine the processing point data with normal preliminary quality of the laser processing part;

[0044] Determine both the normal processing point data and the OK re-judgment data in step 402 as the data meeting the laser processing quality requirements, that is, judge that the processing points of the laser processing part meet the laser processing quality requirements.

[0045] It can be seen that by adding the re-judgment data, the overkill rate during the laser processing detection is reduced, the overkill workpieces (workpieces with qualified quality but judged as NG by the detection system) are reduced, and the production cost is reduced.

[0046] In a second aspect, an embodiment of the present application provides a laser processing quality detection device, including:

[0047] Optoelectronic processing module: It is used to receive the optical radiation signals of at least one processing point in the laser processing path. The optical radiation signals include one or more of infrared radiation signals, visible light radiation signals, and processed laser reflection signals; the received optical radiation signals are photoelectrically converted into electrical signals through a single-point optoelectronic sensor; a correspondence relationship between the laser processing point and the change of the electrical signal during the laser processing is established; according to the correspondence relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts, the processing point data with defective initial quality of the laser processing part processing points is determined; wherein, the normal electrical signal is the range of the variation of the electrical signal value corresponding to the qualified processing of the laser processing part processing points under a certain process parameter.

[0048] Artificial intelligence data processing unit, which is used to perform another quality judgment on the defective processing point data according to a pre-trained machine learning model to determine the quality of the laser processing part processing points.

[0049] It can be seen that the laser processing quality detection equipment realizes reducing the overkill number under the condition of zero or near-zero missed killing by fusing the existing multi-spectral optical detection algorithm and the AI detection algorithm, achieving the effect of saving production costs. For each workpiece, first use the traditional algorithm to make a preliminary judgment. When the traditional algorithm judges it as OK, no further judgment is made, and the result of the traditional algorithm is used as the final result; when the traditional algorithm judges it as NG, it enters the AI detection process for rejudgment, and the AI detection result is used as the final result. The two-stage detection process can combine the advantages of the traditional algorithm and AI detection, and finally realize reducing the overkill number under the condition of approaching zero missed killing or setting zero missed killing.

[0050] Combined with the second aspect, in a possible implementation manner, the artificial intelligence data processing unit is further used to execute the acquisition of a pre-trained machine learning model, and its characteristics are:

[0051] The artificial intelligence data processing unit obtains unqualified processing electrical signals under defective processing process parameters by simulating changes in laser processing parameters (such as power jitter changes of processing lasers, defocus amount changes of processing lasers, processing temperature changes, etc.); processes the noise of unqualified processing electrical signals through a filtering algorithm to unify the data scale of unqualified processing electrical signals; extracts time-domain and / or frequency-domain features from the unified unqualified processing electrical signal data; injects the time-domain and / or frequency-domain features into the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts to obtain simulated abnormal electrical signals; performs pre-training on the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts and the simulated abnormal electrical signals through a deep learning network; obtains a pre-trained machine learning model.

[0052] It can be seen that by pre-simulating the changes in laser processing parameters to obtain a batch of unqualified laser processing detection electrical signals, and after data alignment, extracting the time-domain and / or frequency-domain characteristics of unqualified processing, and injecting the time-domain and / or frequency-domain characteristics into the OK signal, a pre-trained machine learning model can be obtained based on the OK signal and the simulated NG signal.

[0053] Combined with the second aspect, in a possible implementation, the artificial intelligence data processing unit is further used to perform the acquisition of a pre-trained machine learning model, and is characterized in that:

[0054] The artificial intelligence data processing unit determines an abnormal electrical signal of one or more fault causes based on the pre-stored historical unqualified processing electrical signals and / or the on-site pre-collected unqualified processing electrical signals; simulates the abnormal electrical signal of this section through one or more functions such as a constant fault function, a cosine fault function, and a variance fault function; randomly selects a certain segment of the normal electrical signal corresponding to the processing point of the pre-stored laser processing standard part, and injects the selected segment of the normal electrical signal into the simulated abnormal electrical signal of the selected corresponding segment to obtain a simulated abnormal electrical signal; performs pre-training through a deep learning network based on the pre-stored normal electrical signal corresponding to the processing point of the laser processing standard part and the simulated abnormal electrical signal; and obtains a pre-trained machine learning model.

[0055] Practice shows that in the application scenario of laser processing detection, simulating the abnormal electrical signal of this section through one or more functions such as a constant fault function, a cosine fault function, and a variance fault function can more effectively simulate and restore the defect characteristics of laser processing detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0057] Figure 1 It is a schematic diagram of the architecture of a laser processing detection system in the prior art;

[0058] Figure 2 It is a schematic flowchart of a laser processing quality detection method provided by an embodiment of the present application;

[0059] Figure 3-1 、 3-2 、3-3 are respectively schematic diagrams of injecting the normal electrical signal corresponding to the processing point of the pre-stored laser processing standard part through a constant fault function, a cosine fault function, and a variance fault function provided by the present application; and

[0060] Figure 4 Schematic structural diagram of a laser processing quality detection device provided by an embodiment of the present application. Specific implementation manners

[0061] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0062] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "the" and "said" used in the embodiments of the present application, the description of the drawings and the claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should be understood that the term "and / or" used herein is only a description of the associated relationship of the associated objects, indicating that three relationships may exist, for example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after. It should be understood that although the terms "first", "second" and similar words may be used in the embodiments of the present application, they do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "a" do not indicate a quantity limitation, but indicate that there is at least one. The "multiple" involved in the embodiments of the present application refers to greater than or equal to two.

[0063] As recorded in prior art patents such as CN202011397629.6, please refer to Figure 1 , Figure 1 which is a schematic diagram of the architecture of a prior art laser processing detection system. As Figure 1 shown, the system includes a laser welding system, a multi-optical sensor module 4, a signal processing module 6, and an industrial control computer 8.

[0064] Among them, the laser welding system includes a laser 1, an optical fiber 2, and a laser welding head 3. The laser 1 is connected to the laser welding head 3 through the optical fiber 2. A multi-optical sensor module 4 is coaxially installed on the laser welding head 3. The multi-optical sensor module 4 is connected to a signal processing module 6 through a first signal line 5. The signal processing module 6 is connected to an industrial control computer 8 through a second signal line 7. The industrial control computer 8 is connected to the laser welding system through a third signal line 12. The laser generated by the laser 1 is transmitted to the laser welding head 3 through the optical fiber 2, and then incident on the upper plate 10 of the metal material to be processed by the laser welding head 3. After the upper plate 10 absorbs the laser, it melts and solidifies to form a target weld 9. The infrared light, visible light, and laser reflection signals radiated by the target weld 9 after welding are transmitted to the multi-optical sensor module 4 through the laser welding head 3. The multi-optical sensor module 4 can receive the laser reflection signal, the infrared light signal, and the visible light signal radiated by the target weld 9 and convert them into corresponding electrical signals. The electrical signal is transmitted to the signal processing module 6 through the first signal line 5. After being processed, it is transmitted to the industrial control computer 8 through the second signal line 7. The industrial control computer 8 processes the received signal to determine whether there are defects in the target weld 9.

[0065] In the prior art, when it is determined that there are defects in the target weld 9, the industrial control computer 8 determines the position information of the defects in the target weld 9 and sends this position information to the laser welding system through the third signal line 12 so that the laser welding system reprocesses the position where the defects exist in the target weld 9.

[0066] When applying AI artificial intelligence processing and analysis technology to laser processing detection, in the past, the industrial control computer 8 directly analyzed the received signals, with a large amount of labeled data and computing resources in the early stage, and then applied the AI technology to the field of laser processing detection.

[0067] For example, the multi-optical sensor module 4 is used to monitor the current states of the laser welding head 3 and / or corresponding components to form detection and measurement data to be analyzed. The relevant measurement data includes the sum of different measurement parameters and / or the time variation process of each measured value of the measurement parameter. The measurement parameter can be: temperature, pressure, scattered light or scattered light intensity, (air) humidity, acceleration, voltage, current, and digital communication signals of optical components, mechanical components, electrical components, and / or electronic components.

[0068] The relevant measurement data may also include the sum of different process parameters and / or the time variation process of each value of the process parameters during laser processing. For example, laser parameters, processing parameters, and / or environmental parameters. The laser parameters may be the current laser power, beam parameter product, pulse peak power, pulse length, pulse frequency, wavelength, and / or fiber diameter. The processing parameters may be the laser processing focus position, air pressure, gas type (such as nitrogen, oxygen, compressed air, etc.), nozzle type, nozzle diameter, imaging ratio of the optical system, current acceleration of the laser processing head, feed speed of the laser processing head, distance between the laser processing head and the workpiece to be processed, actuator data of the laser processing head, and other machine components. The environmental parameters may include environmental air temperature and environmental air humidity.

[0069] When analyzing the above relevant data through AI technology, the measurement data and, if necessary, the process data are processed or analyzed by a pre-trained machine learning model such as a training model, so as to obtain a prediction or estimation of the state during laser processing based on the current state of the laser processing head. Subsequently, the determined state can be evaluated or classified. Based on the determined state, it can be output whether there are defects in the laser processing process, the cause of the defects, or improvement measures, etc. The machine learning model can adopt a neural network, a recurrent neural network, a convolutional neural network, a deep convolutional neural network, a random forest, a support vector machine, fuzzy logic, a decision tree, or a combination thereof (for example, a fuzzy decision tree).

[0070] The applicant found that during laser processing, due to the complex processing scenario, involving a large number of data sources that are widely distributed, it is impossible to obtain accurate AI judgment results. Therefore, the applicant proposes a laser processing detection method, device, and system, which apply AI technology based on the determined restricted range data to avoid being affected by factors such as relevant noise and interference in the complex environment of laser processing, and integrate AI technology with the existing laser processing detection system to achieve more intelligent and accurate laser processing detection.

[0071] Based on Figure 1 the existing technical solution, a photoelectric sensor is used to collect the light reflection information of different bands, and then a benchmark is established and a fault is detected using the characteristic threshold determination logic. The process includes:

[0072] Step 1: Collect the light reflection data of 30 - 50 times of manually confirmed qualified welding quality (the welding of each workpiece will generate time series data of three channels, representing the visible light channel, the reflected light channel, and the infrared light channel respectively). Aggregate this batch of data, and then calculate the upper envelope curve, lower envelope curve, and upper and lower envelope mean curve of each channel data to obtain three reference lines for each channel, which are respectively called: upper reference, lower reference, and reference median line.

[0073] Step 2. The reference line established in Step 1 will be used in subsequent fault detection. The specific process is as follows: For the welding of a certain workpiece, the workpiece will also generate time series data of three channels. For each channel, the eigenvalue between the channel and the upper reference, lower reference, and reference midline of the corresponding channel will be calculated. The specific eigenvalues mainly include the following features: the average offset value between the sequence and the reference midline, the area between the sequence and the reference midline, the width, height, area of the sequence exceeding the upper reference (or below the lower reference)... A total of about 50 features will be extracted.

[0074] It can be understood that a threshold range is set for each feature. As long as one of the calculated features of the workpiece exceeds the set feature threshold range, the workpiece will be determined as an NG workpiece, and vice versa for an OK workpiece.

[0075] In practical applications, since the accuracy of the detection system depends on the dimension of feature extraction and the set threshold range of each feature, in the application scenario of laser processing detection, our attention to NG workpieces is significantly higher than that to OK workpieces. Therefore, we hope that all NG workpieces can be detected. The problem brought about by this is that new features need to be continuously extracted according to the signal characteristics of NG workpieces. And in order to ensure that all NG workpieces can be detected, the artificially defined feature threshold range will be relatively narrow, resulting in a certain number of over-killed workpieces (workpieces with qualified quality but determined as NG by the detection system). If the over-kill rate is too high, it will increase the production cost.

[0076] For this reason, please refer to Figure 2 This application embodiment provides a laser processing quality detection method, including the following steps:

[0077] Step 100: Receive the optical radiation signal of at least one processing point in the laser processing path. The optical radiation signal includes one or more of infrared radiation signal, visible light radiation signal, and processing laser reflection signal; perform photoelectric conversion on the received optical radiation signal into an electrical signal through a single-point photoelectric sensor; establish a corresponding relationship representing the change of the electrical signal corresponding to the laser processing point during the laser processing process; determine the processing point data with defects in the preliminary quality of the laser processing part processing point according to the corresponding relationship and the normal electrical signal corresponding to the processing point of the pre-stored laser processing standard part; wherein, the normal electrical signal is the range of the variation amplitude of the electrical signal corresponding to the qualified processing process of the laser processing part processing point under a certain process parameter.

[0078] Step 200: Perform another quality judgment on the processing point data with defects according to the pre-trained machine learning model to determine the quality of the laser processing part processing point.

[0079] In the embodiment of the present application, by fusing the existing multi-spectral optical detection algorithm and the AI detection algorithm, the overkill number is reduced under the condition of approaching zero false negatives or setting zero false negatives, so as to achieve the effect of saving production costs. For each workpiece, first use the traditional algorithm to make a preliminary judgment on it. When the traditional algorithm determines it as OK, no further judgment is made, and the result of the traditional algorithm is used as the final result; when the traditional algorithm determines it as NG, it enters the AI detection process for re-judgment, and the AI detection result is used as the final result. The two-stage detection process can combine the advantages of the traditional algorithm and AI detection, and finally achieve reducing the overkill number under the condition of approaching zero false negatives or setting zero false negatives.

[0080] Optionally, in step 200, the specific steps for obtaining the pre-trained machine learning model are as follows:

[0081] Step 301: By simulating the change of laser processing parameters (such as the power jitter change of the processing laser, the defocus amount change of the processing laser, the processing temperature change, etc.), obtain the unqualified processing electrical signal under the defective processing process parameters;

[0082] Step 302: Process the noise of the unqualified processing electrical signal through a filtering algorithm (such as wavelet transform, Kalman filter) to unify the data scale of the unqualified processing electrical signal;

[0083] Step 303: Extract time-domain and / or frequency-domain features (such as FFT, peak energy, etc.) from the unified unqualified processing electrical signal data;

[0084] Step 304: Inject the time-domain and / or frequency-domain features into the normal electrical signal corresponding to the processing point of the pre-stored laser processing standard part to obtain a simulated abnormal electrical signal;

[0085] Step 305: Perform pre-training through a deep learning network according to the normal electrical signal corresponding to the processing point of the pre-stored laser processing standard part and the simulated abnormal electrical signal; obtain the pre-trained machine learning model.

[0086] In the embodiment of the present application, a batch of unqualified laser processing detection electrical signals are obtained by pre-simulating the change of laser processing parameters. After data alignment, the time-domain and / or frequency-domain features of the unqualified processing are extracted, and the time-domain and / or frequency-domain features are injected into the OK signal. Furthermore, a pre-trained machine learning model can be obtained according to the OK signal and the simulated NG signal.

[0087] Optionally, in step 200, the specific steps for obtaining the pre-trained machine learning model are as follows:

[0088] Step 311: Determine an abnormal electrical signal of one or more failure causes according to the historically pre-stored unqualified processing electrical signal and / or the unqualified processing electrical signal pre-collected on site;

[0089] Step 312: Simulate the abnormal electrical signals of this section through one or more of the constant failure function, cosine failure function, and variance failure function;

[0090] Step 313: Randomly select a certain segment of the normal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts;

[0091] Step 314: Inject a certain segment of the normal electrical signal selected in Step 313 into the simulated abnormal electrical signal of the corresponding segment selected in Step 312 to obtain a simulated abnormal electrical signal;

[0092] Step 315: Perform pre-training through the deep learning network based on the normal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts and the simulated abnormal electrical signal; obtain a pre-trained machine learning model.

[0093] It can be understood that the pre-trained machine learning model in this embodiment is obtained based on injecting faults into the normal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts.

[0094] It can be understood that the constant failure function, cosine failure function, and variance failure function can be widely applied in reliability engineering and fault analysis. The related failure functions (or failure rate functions) are usually used to describe the probability or frequency of a device or system failing over time. Practice has shown that in the application scenario of laser processing detection, simulating the abnormal electrical signals of this section through one or more of the constant failure function, cosine failure function, and variance failure function can more effectively simulate and restore the defect characteristics of laser processing detection.

[0095] Please refer to Figure 3-1 , in the coordinate system, the signals covered by the gray box are used to represent the state of injecting the constant failure function into the normal signal. It can be seen that the constant failure function means that the failure rate (or failure probability) remains unchanged over time. This failure mode is usually applicable to the random failure stage, such as the random failure of a device during normal use.

[0096] The mathematical expression of the failure rate can be: λ(t) = k

[0097] where k is a constant representing the failure rate per unit time.

[0098] Suppose the failure rate of a certain device is 0.001 failures per hour, that is, k = 0.001. This means that in any given hour, the probability of the device failing is 0.1%.

[0099] Please refer to Figure 3-2, in the coordinate system, the signal covered by the gray box is used to represent the state of the cosine fault function injecting a normal signal. It can be seen that the cosine fault function is a periodic fault mode, and the failure rate changes as a cosine function over time. This fault mode may be applicable to devices affected by periodic stress.

[0100] Please refer to Figure 3-3 , in the coordinate system, the signal covered by the gray box is used to represent the variance fault function

[0101] injecting the state of a normal signal. It can be seen that the variance fault function may refer to that the failure rate is related to a certain variance, such as the variance of device performance or environmental factors. This fault mode may be applicable to systems that are greatly affected by random fluctuations.

[0102] Constant fault function: Applicable to the random fault stage, the failure rate remains unchanged.

[0103] Cosine fault function: Applicable to devices affected by periodic stress, the failure rate changes periodically.

[0104] Variance fault function: Applicable to systems affected by random fluctuations, the failure rate is related to the variance.

[0105] These fault functions can be adjusted and optimized according to specific situations in practical applications to better describe the fault behavior of devices or systems.

[0106] Optionally, in step 200, the specific steps to obtain the pre-trained machine learning model are as follows:

[0107] Step 321: Determine a segment of abnormal electrical signal of one or more fault causes based on the historically pre-stored unqualified processing electrical signals and / or the unqualified processing electrical signals pre-collected on-site;

[0108] Step 322: Randomly use one or more functions to simulate this segment of abnormal electrical signal; the functions include constant fault function, cosine fault function, and variance fault function.

[0109] Step 323: Randomly select a certain segment of the normal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts;

[0110] Step 324: Inject the selected certain segment of the normal electrical signal in step 323 into the simulated abnormal electrical signal of the corresponding segment selected in step 322 to obtain a simulated abnormal electrical signal;

[0111] Step 325: Perform pre-training through a deep learning network based on the normal electrical signal corresponding to the processing points of the pre-stored laser processing standard parts and the simulated abnormal electrical signal; obtain the pre-trained machine learning model;

[0112] Step 326: Fine-tune the pre-trained machine learning model according to the abnormal electrical signals obtained during online processing to obtain a fine-tuned sub-model;

[0113] Step 327: Repeat Steps 321 to 326 to obtain multiple fine-tuned sub-models, and integrate the multiple fine-tuned sub-models into a unified machine learning model in an ensemble model manner. It can be understood that by combining the prediction results of multiple models, better performance than a single model can be obtained. These sub-models can be the same algorithm (such as multiple decision trees) or different algorithms (such as decision trees, neural networks, support vector machines, etc.). The prediction results of these sub-models are integrated through a set strategy to finally form a required model.

[0114] Furthermore, in the embodiments of the present application, in different process application environments, by integrating models and combining the advantages of multiple models, these errors are reduced. For example, some models may perform well on certain data but poorly on other data, while the ensemble model can balance these differences through voting or averaging. At the same time, the ensemble model is more robust to data noise and outliers. Even if some sub-models are affected by optoelectronic detection noise, the correct predictions of other sub-models can compensate for these errors. Through the processing of the ensemble model, the business requirements of online detection in laser processing welding environments and complex process scenarios are better met.

[0115] Specifically, in the embodiments of the present application, the pre-trained machine learning model can solve the training problem of extremely unbalanced sample numbers of OK data and NG data on the production line. The steps for constructing the pre-trained machine learning model can be as follows:

[0116] 1. Collect more than 50,000 OK sample data (sample data with good processing quality at the processing point) and more than 100 NG sample data (sample data with defective processing quality at the processing point);

[0117] 2. Observe the failure modes of the NG sample data, such as determining the cause of the failure through the abnormal shape of the signal;

[0118] 3. According to the observation results, use a certain function to approximately simulate the shape of the failure (one or more functions such as constant failure, cosine failure, variance failure, etc.) can be adopted, and each function has a certain degree of randomness, such as the period, amplitude, noise variance, etc. of the cosine failure;

[0119] 4. For each OK sample, randomly select a certain segment of the OK sample signal (the channel, position, and width of the segment are random), inject the segment into the failure simulated by the failure generation function to obtain the simulated NG sample corresponding to the OK sample, and perform failure injection on each OK sample in this way. Finally, 50,000 real OK samples and 50,000 simulated NG samples will be obtained;

[0120] 5. For the data obtained in the above step 4, construct a deep learning network. Optionally, based on the design idea of a densely connected network, perform large-scale pre-training on the above samples to obtain a pre-trained model.

[0121] 6. After obtaining the pre-trained model, fine-tune the model on the real production line NG data to pull back the fault distribution learned by the pre-trained model to the real production line distribution, and finally obtain a fine-tuned sub-model.

[0122] 7. Repeat step 6 to obtain multiple fine-tuned sub-models, and use the ensemble model method to integrate multiple fine-tuned sub-models into a unified machine learning model.

[0123] Furthermore, after obtaining a pre-trained machine learning model through various different acquisition channels, it can be considered that the performance of the evaluation model passes, and this model can accurately judge whether the data meets the standards. Commonly used evaluation metrics include: Accuracy: the proportion of correct judgments by the model. Recall: the proportion of "non-compliant" data that the model can detect. F1-score: the harmonic mean of accuracy and recall.

[0124] It can be understood that in step 200, the specific steps for performing a quality judgment on the defective processing point data again according to the pre-trained machine learning model are as follows:

[0125] Step 401: Based on the pre-trained machine learning model, perform data screening on the processing point data with preliminary quality defects in the laser processed parts obtained in step 100.

[0126] Step 402: Divide the processing point data with defects in step 100 into OK re-judgment data and NG re-judgment data, and determine that the OK re-judgment data meets the laser processing quality requirements.

[0127] It can be seen that through the method of this embodiment, by performing a quality judgment on the defective processing point data again through the pre-trained machine learning model, it is possible to realize the re-verification of the laser processing NG data, and at the same time, it can also reduce the difficulty of setting the previous feature threshold, saving the workload of personnel. Through the setting of the machine learning model, the objectivity and result consistency of the subsequent laser processing quality judgment are ensured, the accuracy of the detection result obtained by the weld seam detection is improved, the practical problem of high overkill rate under the conventional feature threshold setting algorithm is avoided, and the stability of the detection result is ensured.

[0128] To improve the fault discovery, diagnosis and analysis of processing defects by photoelectric sensors during laser online processing, and to achieve early prediction and early intervention of equipment failures to ensure continuous production. In step 100: According to the corresponding relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts, determine the processing point data of the processing points with preliminary quality defects in the laser processing parts; The method further includes the following steps:

[0129] Receive the external defect data of the processing points of the laser processing parts transmitted back, and after merging the external defect data with the processing point data of the processing points with preliminary quality defects, form the processing point data with defects to be analyzed in step 200. It can be understood that in this embodiment, the external defect data of the processing points of the laser processing parts transmitted back can be data input manually, detected and fed back, or remotely edited. Furthermore, in the laser online detection, the basic data can be updated in real time for the quality re-judgment process through a pre-trained machine learning model. In view of the requirement of many variant factors in the complex photoelectric scenario of laser processing, the data is updated in time to improve the accuracy of quality re-judgment.

[0130] It can be understood that step 100 further includes the following steps:

[0131] Establish a corresponding relationship representing the change of the electrical signal corresponding to the laser processing point during the laser processing process; According to the corresponding relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts, determine the processing point data of the processing points with normal preliminary quality in the laser processing parts;

[0132] Determine both the normal processing point data and the OK re-judgment data in step 402 as the data meeting the laser processing quality requirements, that is, determine that the processing points of the laser processing parts meet the laser processing quality requirements.

[0133] Furthermore, by adding re-judgment data, the overkill rate during laser processing detection is reduced, the overkilled workpieces (workpieces with qualified quality but determined as NG by the detection system) are reduced, and the production cost is lowered.

[0134] See Figure 4 , Figure 4 which is a schematic structural diagram of a laser processing quality detection device provided by an embodiment of the present application. As Figure 4 shown, the laser processing quality detection device 500 includes:

[0135] Optical and electrical sensor 501: It is used to receive the optical radiation signals of at least one processing point in the laser processing path. The optical radiation signals include one or more of infrared radiation signals, visible light radiation signals, and processing laser reflection signals; the received optical radiation signals are photoelectrically converted into electrical signals through a single-point optical and electrical sensor; a corresponding relationship is established to characterize the change of the electrical signals corresponding to the laser processing points during the laser processing process; according to the corresponding relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts, the processing point data with defects in the initial quality of the laser processing part processing points is determined; wherein, the normal electrical signals are the value range of the corresponding variable electrical signals during the qualified processing process of the laser processing part processing points under a certain process parameter.

[0136] Artificial intelligence data processing unit 502: It is used to perform another quality judgment on the processing point data with defects according to the pre-trained machine learning model to determine the quality of the laser processing part processing points.

[0137] Furthermore, the laser processing quality detection device 500 realizes reducing the overkill number under the condition of zero or near-zero missed killings by fusing the existing multi-spectral optical detection algorithm and the AI detection algorithm, achieving the effect of saving production costs. For each workpiece, first use the traditional algorithm to make a preliminary judgment on it. When the traditional algorithm judges it as OK, no further judgment is made, and the result of the traditional algorithm is used as the final result; when the traditional algorithm judges it as NG, it enters the AI detection process for re-judgment, and the AI detection result is used as the final result. The two-stage detection process can combine the advantages of the traditional algorithm and the AI detection, and finally realize reducing the overkill number under the condition of zero or near-zero missed killings.

[0138] Optionally, the artificial intelligence data processing unit 502 is also used to perform the steps of obtaining the pre-trained machine learning model in the foregoing embodiments, such as:

[0139] By simulating the changes in laser processing parameters (such as the power jitter change of the processing laser, the defocus amount change of the processing laser, the processing temperature change, etc.), the unqualified processing electrical signals under the defective processing process parameters are obtained; the noise of the unqualified processing electrical signals is processed through a filtering algorithm (such as wavelet transform, Kalman filter) to unify the data scale of the unqualified processing electrical signals; the time domain and / or frequency domain features (such as FFT, peak energy, etc.) are extracted from the unified unqualified processing electrical signal data; the time domain and / or frequency domain features are correspondingly injected into the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts to obtain simulated abnormal electrical signals; the pre-stored normal electrical signals corresponding to the processing points of the laser processing standard parts and the simulated abnormal electrical signals are pre-trained through a deep learning network; the pre-trained machine learning model is obtained.

[0140] In the embodiments of the present application, by pre-simulating the changes in laser processing parameters, a batch of unqualified laser processing detection electrical signals are obtained. After data alignment, the time-domain and / or frequency-domain characteristics of the unqualified processing are extracted, and the time-domain and / or frequency-domain characteristics are injected into the OK signal. Furthermore, a pre-trained machine learning model can be obtained according to the OK signal and the simulated NG signal.

[0141] Optionally, the specific steps for the artificial intelligence data processing unit 502 to execute the pre-trained machine learning model in the foregoing embodiments may also be:

[0142] According to the historical pre-stored unqualified processing electrical signals and / or the on-site pre-collected unqualified processing electrical signals, an abnormal electrical signal of one or more fault causes is determined; the abnormal electrical signal of this section is simulated by one or more functions such as a constant fault function, a cosine fault function, and a variance fault function; a certain segment of the normal electrical signal corresponding to the pre-stored laser processing standard part processing point is randomly selected, and the selected segment of the normal electrical signal is injected into the simulated abnormal electrical signal of the selected corresponding segment to obtain a simulated abnormal electrical signal; pre-training is performed through a deep learning network according to the normal electrical signal corresponding to the pre-stored laser processing standard part processing point and the simulated abnormal electrical signal; a pre-trained machine learning model is obtained.

[0143] It can be understood that the pre-trained machine learning model in this embodiment is obtained based on the normal electrical signal corresponding to the fault-injected pre-stored laser processing standard part processing point.

[0144] It can be understood that the constant fault function, the cosine fault function, and the variance fault function can be widely used in reliability engineering and fault analysis. Related fault functions (or failure functions) are usually used to describe the probability or frequency of a device or system failing over time.

[0145] Optionally, the specific steps for the artificial intelligence data processing unit 502 to execute to obtain the pre-trained machine learning model in the foregoing embodiments may also be:

[0146] Determine a segment of abnormal electrical signal for one or more fault causes based on the pre-stored historical unqualified processing electrical signals and / or the unqualified processing electrical signals pre-collected on-site; randomly use one or more functions to simulate the abnormal electrical signal of this segment; the functions include a constant fault function, a cosine fault function, and a variance fault function. Randomly select a certain segment of the normal electrical signal corresponding to the pre-stored laser processing standard part processing point, inject the selected certain segment of the normal electrical signal into the simulated abnormal electrical signal of the selected corresponding segment to obtain a simulated abnormal electrical signal; perform pre-training on the pre-stored normal electrical signal corresponding to the laser processing standard part processing point and the simulated abnormal electrical signal through a deep learning network; obtain a pre-trained machine learning model; fine-tune the pre-trained machine learning model according to the abnormal electrical signal obtained during on-line processing to obtain a fine-tuned sub-model;

[0147] Repeat the steps of the fine-tuned sub-model, and integrate multiple fine-tuned sub-models into a unified machine learning model in an integrated model manner.

[0148] It can be understood that in a laser processing quality detection device provided by an embodiment of the present application, the photoelectric sensor 501 and the artificial intelligence data processing unit 502 are used to execute each step of the laser processing quality detection in the foregoing embodiment, and the present application will not elaborate here.

[0149] An embodiment of the present application further provides an artificial intelligence data processing chip, the chip includes a processor and a data interface, and the processor reads instructions stored on a memory through the data interface to implement the laser processing quality detection method described above.

[0150] Optionally, as an implementation manner, the chip may further include a memory, and instructions are stored in the memory, and the processor is used to execute the instructions stored on the memory. When the instructions are executed, the processor is used to execute the laser processing defect detection parameter determination method described above.

[0151] An embodiment of the present application further provides a computer-readable storage medium, and instructions are stored in the computer-readable storage medium. When it runs on a computer or a processor, it causes the computer or the processor to execute one or more steps in any of the above methods.

[0152] An embodiment of the present application further provides a computer program product containing instructions. When the computer program product runs on a computer or a processor, it causes the computer or the processor to execute one or more steps in any of the above methods.

[0153] Those skilled in the art will appreciate that the functions described in connection with the various illustrative logical blocks, modules, and algorithm steps disclosed herein can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions described in the various illustrative logical blocks, modules, and steps can be stored or transmitted as one or more instructions or code on a computer-readable medium and executed by a hardware-based processing unit. The computer-readable medium can include a computer-readable storage medium corresponding to a tangible medium, such as a data storage medium, or a communication medium including any medium that facilitates transfer of a computer program from one place to another (e.g., based on a communication protocol). In this way, the computer-readable medium generally can correspond to (1) a non-transitory tangible computer-readable storage medium, or (2) a communication medium, such as a signal or carrier wave. The data storage medium can be any available medium that can be accessed by one or more computers or one or more processors to retrieve instructions, code, and / or data structures for implementing the techniques described in this application. A computer program product can include a computer-readable medium.

[0154] By way of example, and not limitation, such computer-readable storage media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory, or any other medium that can be used to store the desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if instructions are transmitted using coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of the medium. However, it should be understood that the computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but rather are directed to non-transitory tangible storage media. As used herein, disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), and Blu-ray disc, where disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0155] Instructions may be executed by one or more processors, such as one or more digital signal processors (DSPs), general microprocessors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Thus, the term “processor” as used herein may refer to any of the foregoing structures or any other structure suitable for implementing the techniques described herein. Additionally, in some aspects, the functions described for the various illustrative logical blocks, modules, and steps described herein may be provided within dedicated hardware and / or software modules configured for encoding and decoding, or incorporated in a combined codec. Moreover, the techniques may be fully implemented in one or more circuits or logic elements.

[0156] The techniques of this application may be implemented in a variety of devices or apparatuses, including wireless handsets, integrated circuits (ICs) or a set of ICs (e.g., a chipset). Various components, modules, or units are described in this application to emphasize functional aspects of the devices for performing the disclosed techniques, but need not be implemented by distinct hardware units. In fact, as described above, the various units may be combined in an encoding hardware unit with suitable software and / or firmware, or provided by interoperating hardware units, including one or more processors as described above.

[0157] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, the specific working processes of the systems, apparatuses, and units described above may refer to the corresponding step processes in the foregoing method embodiments, and will not be elaborated herein.

[0158] It should be understood that in the description of this application, unless otherwise specified, " / " means that the objects associated before and after are in an "or" relationship. For example, A / B can represent A or B; where A and B can be singular or plural. Also, in the description of this application, unless otherwise specified, "a plurality of" means two or more than two. "At least one (item)" or similar expressions thereof refer to any combination of these items, including any combination of single item(s) or plural item(s). For example, at least one (item) of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple. Additionally, for the convenience of clearly describing the technical solutions of the embodiments of this application, in the embodiments of this application, terms such as "first" and "second" are used to distinguish identical or similar items with basically the same functions and roles. Those skilled in the art can understand that terms such as "first" and "second" do not limit the quantity and execution order, and "first", "second", etc. do not necessarily mean different. At the same time, in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner for easy understanding.

[0159] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the division of the unit is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The couplings shown or discussed, either direct couplings or communication connections, can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0160] The unit described as a separate component may or may not be physically separated, and the component shown as a unit may or may not be a physical unit, that is, it can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0161] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a read-only memory (ROM), a random access memory (RAM), a magnetic medium, such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium, such as a digital versatile disc (DVD), or a semiconductor medium, such as a solid state disk (SSD), etc.

[0162] As described above, it is only the specific implementation manner of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the embodiments of the present application should be covered by the protection scope of the embodiments of the present application. Therefore, the protection scope of the embodiments of the present application should be subject to the protection scope of the claims.

[0163] The device embodiments described above are merely illustrative. The units and modules described as separate components may or may not be physically separated. Additionally, some or all of the units and modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative work.

[0164] The above description is only the specific implementation manner of the present application. It should be noted that for those of ordinary skill in the technical field, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for detecting the quality of laser processing, characterized in that, The method includes: Step 100: Receive the optical radiation signals of at least one processing point in the laser processing path. The optical radiation signals include one or more of infrared radiation signals, visible light radiation signals, and processing laser reflection signals; photoelectrically convert the received optical radiation signals into electrical signals through a single-point photoelectric sensor; establish a correspondence relationship between the laser processing points and the changes in the electrical signals during the laser processing process; determine the processing point data with defective preliminary quality of the laser processing part according to the correspondence relationship and the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts. Wherein, the normal electrical signal is the range of the value amplitude of the corresponding variable electrical signal during the qualified processing of the processing points of the laser processing part under a certain process parameter; Step 200: Perform a quality judgment on the processing point data with defects again according to the pre-trained machine learning model to determine the quality of the processing points of the laser processing part.

2. The method according to claim 1, characterized in that The specific steps for obtaining the pre-trained machine learning model include: Step 301: Obtain the unqualified processing electrical signals under the defective processing process parameters by simulating the change mode of the laser processing parameters; Step 302: Process the noise of the unqualified processing electrical signals through a filtering algorithm to unify the data scale of the unqualified processing electrical signals; Step 303: Extract time domain and / or frequency domain features from the unified unqualified processing electrical signal data; Step 304: Inject the time domain and / or frequency domain features into the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts respectively to obtain simulated abnormal electrical signals; Step 305: Perform pre-training through a deep learning network according to the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts and the simulated abnormal electrical signals; obtain the pre-trained machine learning model.

3. The method according to claim 1, wherein The specific steps for obtaining the pre-trained machine learning model include: Step 311: Determine a section of abnormal electrical signals for one or more failure reasons according to the historically pre-stored unqualified processing electrical signals and / or the unqualified processing electrical signals pre-collected on-site; Step 312: Simulate the section of abnormal electrical signals through one or more functions of constant failure function, cosine failure function, variance failure function; Step 313: Randomly select a certain segment of the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts; Step 314: Inject the selected segment of the normal electrical signal in Step 313 into the simulated abnormal electrical signal of the corresponding segment selected in Step 312 to obtain a simulated abnormal electrical signal; Step 315: Perform pre-training through a deep learning network according to the normal electrical signals corresponding to the processing points of the pre-stored laser processing standard parts and the simulated abnormal electrical signals; obtain the pre-trained machine learning model.

4. The method according to claim 1, wherein The specific steps for obtaining the pre-trained machine learning model include: Step 321: Determine a section of abnormal electrical signals for one or more failure reasons according to the historically pre-stored unqualified processing electrical signals and / or the unqualified processing electrical signals pre-collected on-site; Step 322: Randomly use one or more functions to simulate the section of abnormal electrical signals; the functions include constant failure function, cosine failure function, variance failure function. Step 323: Randomly select a certain segment of the normal electrical signal corresponding to the pre-stored laser processing standard part processing point; Step 324: Inject a certain segment of the normal electrical signal selected in Step 323 into the simulated abnormal electrical signal of the corresponding segment selected in Step 322 to obtain a simulated abnormal electrical signal; Step 325: Perform pre-training through a deep learning network based on the pre-stored normal electrical signal corresponding to the laser processing standard part processing point and the simulated abnormal electrical signal; obtain a pre-trained machine learning model; Step 326: Fine-tune the pre-trained machine learning model according to the abnormal electrical signal obtained during on-line processing to obtain a fine-tuned sub-model; Step 327: Repeat Steps 321 to 326 to obtain multiple fine-tuned sub-models, and integrate the multiple fine-tuned sub-models into a unified machine learning model in an ensemble model manner.

5. The method according to claim 2 or 3, characterized in that, The specific steps for re-judging the quality of the defective processing point data according to the pre-trained machine learning model are as follows: Step 401: Based on the pre-trained machine learning model, perform data screening on the processing point data with preliminary quality defects in the laser processing part obtained in Step 100; Step 402: Divide the processing point data with defects in Step 100 into OK re-judging data and NG re-judging data, and determine that the OK re-judging data meets the laser processing quality requirements.

6. The method according to claim 1, wherein In Step 100, according to the corresponding relationship and the pre-stored normal electrical signal corresponding to the laser processing standard part processing point, determine the processing point data with preliminary quality defects in the laser processing part; it further includes the following steps: Receive the external defect data of the laser processing part processing point transmitted back, and after merging the external defect data with the processing point data with preliminary quality defects, form the defective processing point data to be analyzed in Step 200.

7. The method according to claim 5, characterized in that Step 100 further includes the following steps: Establish a corresponding relationship characterizing the change of the electrical signal corresponding to the laser processing point during the laser processing process; according to the corresponding relationship and the pre-stored normal electrical signal corresponding to the laser processing standard part processing point, determine the processing point data with normal preliminary quality in the laser processing part; Determine both the normal processing point data and the OK re-judging data in Step 402 as data that meet the laser processing quality requirements, that is, determine that the laser processing part processing point meets the laser processing quality requirements.

8. A laser processing quality detection device, characterized in that, It includes: Optoelectronic processing module: used to receive the optical radiation signals of at least one processing point in the laser processing path, and the optical radiation signals include: one or more of infrared radiation signals, visible light radiation signals, and processing laser reflection signals; perform optoelectronic conversion of the received optical radiation signals into electrical signals through a single-point optoelectronic sensor; establish a corresponding relationship characterizing the change of the electrical signal corresponding to the laser processing point during the laser processing process; according to the corresponding relationship and the pre-stored normal electrical signal corresponding to the laser processing standard part processing point, determine the processing point data with preliminary quality defects in the laser processing part; wherein, the normal electrical signal is the range of the variation of the electrical signal value corresponding to the qualified processing process of the laser processing part processing point under a certain process parameter. An artificial intelligence data processing unit is used to perform another quality judgment on the defective processing point data according to a pre-trained machine learning model to determine the quality of the processing points of the laser processed parts.

9. The laser processing quality detection device according to claim 8, wherein the artificial intelligence data processing unit is further used to execute the acquisition of the pre-trained machine learning model, and is characterized in that: The artificial intelligence data processing unit obtains unqualified processing electrical signals under defective processing process parameters by simulating changes in laser processing parameters (such as power jitter changes of processing lasers, defocus amount changes of processing lasers, processing temperature changes, etc.); Process the noise of the unqualified processing electrical signals through a filtering algorithm to unify the data scale of the unqualified processing electrical signals; extract time domain and / or frequency domain features from the unified unqualified processing electrical signal data; inject the time domain and / or frequency domain features into the corresponding normal electrical signals of the pre-stored processing points of the laser processing standard parts to obtain simulated abnormal electrical signals; perform pre-training through a deep learning network according to the normal electrical signals and the simulated abnormal electrical signals corresponding to the pre-stored processing points of the laser processing standard parts; Obtain a pre-trained machine learning model.

10. The laser processing quality detection device according to claim 8, wherein the artificial intelligence data processing unit is further used to execute the acquisition of the pre-trained machine learning model, and is characterized in that: The artificial intelligence data processing unit determines a section of abnormal electrical signals of one or more failure causes according to the historically pre-stored unqualified processing electrical signals and / or the unqualified processing electrical signals pre-collected on site; simulates the section of abnormal electrical signals through one or more functions such as a constant failure function, a cosine failure function, and a variance failure function; randomly selects a certain segment of the normal electrical signal corresponding to the pre-stored processing point of the laser processing standard part, and injects the selected segment of the normal electrical signal into the simulated abnormal electrical signal of the selected corresponding segment to obtain a simulated abnormal electrical signal; perform pre-training through a deep learning network according to the normal electrical signals and the simulated abnormal electrical signals corresponding to the pre-stored processing points of the laser processing standard parts; Obtain a pre-trained machine learning model.

Images