An on-line detection method for laser welding penetration based on the recognition of mesoscopic signal arrays

Abstract

The present invention is an on-line detection method for laser welding penetration based on the recognition of a mesoscopic signal array. The present invention relates to on-line detection and intelligent control of welding. The present invention selects the thermally excited mesoscopic signals in the penetration characteristic region within the keyhole as the detection object to obtain mesoscopic detection signals; preprocesses the mesoscopic detection signals to obtain data in a type recognizable by a computer; divides them into a training set, a test set and a validation set after calibration according to the preprocessed data and their corresponding actual penetration states; establishes a neural network model, trains the penetration state recognition model with the data in the training set until the result converges, and then adjusts the model hyperparameters through the validation set to obtain an optimal penetration recognition model; tests the reliability of the model through the test set, and performs on-line detection of laser welding penetration according to the optimal model.

Description

Technical Field

[0001] The present invention relates to the technical field of on-line detection and intelligent control of welding, and is an on-line detection method for laser welding penetration based on the recognition of a mesoscopic signal array. Background Art

[0002] Laser and laser hybrid energy field welding is an indispensable key manufacturing technology in the field of intelligent manufacturing. However, due to fluctuations in the absorption of laser energy by the plume and ejecta during the laser welding process, as well as the non-equilibrium of the metal material vaporization process, etc., the penetration of the welding will be unstable, resulting in stress concentration at the non-penetrated part of the butt weld, and significantly reducing the performance of the welded joint. Therefore, reliable on-line detection of penetration and quality closed-loop control are important issues in the field of laser welding intelligent manufacturing. However, the characteristic area most relevant to the penetration state is the position where the laser beam penetrates the base material at the bottom of the keyhole. The area of this characteristic area is usually less than 1 / 40 of the keyhole opening area, and the diameter is only 0.05 - 0.8 mm, belonging to the mesoscopic scale category. And the detection position of this mesoscopic signal has a great influence on the detection result. For example, a detection deviation of 100 μm is sufficient to significantly reduce the test sensitivity. Moreover, during the laser welding process, the laser keyhole is always in a fluctuating state, and the position of the mesoscopic penetration characteristic area will also swing accordingly. In addition, the laser penetration position does not always exist, but is in a rapid alternating mode of opening / closing. When the bottom opening of the keyhole opens, a low-amplitude signal is generated due to the lack of thermally excited signals at the notch, and when it closes, a high-amplitude signal is generated due to the excitation of laser energy. Therefore, the penetration signal is also a transient signal with bipolar characteristics. So the mesoscopic penetration signal has strong directivity, large volatility, and is easily strongly interfered by surrounding high-amplitude signals. And due to the limitation of macroscopic sampling means, the existing detection methods cannot effectively extract the mesoscopic signals with penetration characteristics. At the same time, since most of the existing data analysis methods are fixed methods such as artificially set filtering, noise reduction, time-domain and frequency-domain analysis, etc., and the deep mapping relationships between different transient distribution characteristics, time-series change characteristics, global change characteristics of the signal and the keyhole behavior and penetration fluctuation process cannot be effectively analyzed, it is difficult to reliably identify the actual penetration state on-line. Summary of the Invention

[0003] In order to overcome the deficiencies of the prior art, the present invention is based on the on-line detection of laser welding penetration by machine recognition of mesoscopic characteristics of the keyhole, and proposes an artificial intelligence on-line detection method that uses array sensing means to collect thermally excited mesoscopic detection signals in the characteristic area inside the keyhole, and establishes a neural network model through machine learning to analyze the signal characteristics and extract the welding penetration quality information.

[0004] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0005] The present invention provides an online detection method for laser welding penetration based on mesoscopic signal array recognition. The present invention provides the following technical solutions:

[0006] A penetration detection device, the device comprising: a housing, a substrate, a sensor, a narrowband filter and a three-dimensional fine adjustment mechanism;

[0007] The substrate is connected to the laser welding head to fix the entire device. There are mounting holes on the substrate to respectively fix the device housing, the three-dimensional fine adjustment mechanism and the narrowband filter carrier. The three-dimensional fine adjustment mechanism is connected to the sensor to adjust the three-dimensional spatial position of the sensor sensing chip. The narrowband filter is located on the optical path between the optical focusing lens group and the sensing surface of the array sensor, and the filtered optical real image can be projected onto the sensing surface of the array sensor.

[0008] An online detection method for laser welding penetration, the method being based on a penetration detection device, the method comprising the following steps:

[0009] Step 1: Select the thermally excited mesoscopic signal in the penetration feature region of the keyhole as the detection object, and obtain the mesoscopic detection signal;

[0010] Step 2: Preprocess the mesoscopic detection signal to obtain data in a type recognizable by a computer;

[0011] Step 3: According to the preprocessed data, after calibration, divide it into a training set, a test set and a validation set;

[0012] Step 4: Establish a neural network model, train the weight parameters of the model with the training set data until the result converges, and then adjust the model hyperparameters through the validation set;

[0013] Step 5: Test the reliability of the model through the test set;

[0014] Step 6: Call the trained recognition model to analyze the signal characteristics online and extract the welding stability information.

[0015] Preferably, the thermally excited state mesoscopic signal of the melt-through characteristic area in the keyhole is selected as the detection object, and the clear real image of the thermally excited state signal at the bottom of the keyhole is projected onto the sensing surface of the array sensor by using the optical focusing imaging and spectral transmission principle to obtain the mesoscopic detection signal. The machine learning method is used to identify the characteristics of the mesoscopic signal, and the model is trained before detection. During detection, the trained model is called to perform real-time analysis to give the detection results and obtain the current melt-through state information.

[0016] Preferably, the thermally excited state signal at the bottom of the keyhole is a near-infrared signal generated when the laser beam enters the base material and the metal at the bottom of the keyhole is rapidly melted and evaporated by intense energy input, accompanied by high-density energy excitation.

[0017] Preferably, the method for collecting mesoscopic detection signals is:

[0018] S1. A high-power optical focusing lens system with a working distance of at least 0.6-1.5 m and a depth of field of 10 mm is used to capture a clear real image of the characteristic area at the bottom of the keyhole from the coaxial optical path of the laser welding head. The depth of field is large enough to capture a clear real image of the characteristic area under the fluctuating state without changing the focal length.

[0019] S2. In the near-infrared spectrum, the welding arc, plume, laser beam, and other large amounts of welding radiation signals above the keyhole are effectively shielded by narrow-band filtering, so that the thermally excited state signals in the characteristic area can be effectively separated, thereby significantly reducing the proportion of invalid signals in the detection signal;

[0020] S3. Project the real image of the thermally excited state signal onto the sensing surface of an array sensor with a large sensing coverage area and high detection accuracy. The sensing area of the array sensor should be ≥ the characteristic area to be measured, and the resolution accuracy should be ≤10μm, thereby obtaining mesoscopic signals at different positions in the characteristic area to be measured.

[0021] Preferably, the melt penetration state identification model is obtained specifically as follows:

[0022] During signal preprocessing, normalization is first performed to convert all data into numerical values ranging from 0 to 1. The processed data set is then calibrated with the actual welding penetration state and divided into training set, test set and validation set. A neural network model with four classification tasks of incomplete penetration, slight penetration, moderate penetration and excessive penetration is constructed using a computer. The weight parameters of the model are trained using the training set data until the results converge. The model hyperparameters are then adjusted using the validation set. Finally, the reliability of the model is tested using the test set to obtain the optimal penetration model.

[0023] Preferably, the single-channel / or multi-channel / or all mesoscopic detection signal data collected are first preprocessed through the signal preprocessing and converted into data of a type recognizable by a computer, and then the trained recognition model is called for operation, enabling the computer to identify which category among the four classifications of incomplete penetration, micro-penetration, moderate penetration, and over-penetration the current penetration feature of laser welding belongs to, giving an online diagnosis result or providing a basis for adjusting key process parameters for the welding closed-loop control system, such as the power of the welding laser, the welding speed, and the defocus amount.

[0024] An online detection system for laser welding penetration, the system comprising:

[0025] A data acquisition module, which selects the thermally excited state mesoscopic signal in the penetration feature area within the keyhole as the detection object to obtain the mesoscopic detection signal;

[0026] A preprocessing module, which preprocesses the mesoscopic detection signal to obtain data of a type recognizable by a computer;

[0027] A calibration module, which divides the data after preprocessing into a training set, a test set, and a validation set after calibration;

[0028] A model establishment module, which establishes a neural network model, trains the weight parameters of the model through the training set data, and then adjusts the hyperparameters of the model through the validation set until the result converges, and tests the reliability of the model through the test set;

[0029] An online detection module, which calls the trained model to analyze the penetration state in real time and extract the welding stability information.

[0030] A computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to implement an online detection method for laser welding penetration.

[0031] A computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements an online detection method for laser welding penetration when executing the computer program.

[0032] The present invention has the following beneficial effects:

[0033] The present invention uses the thermally excited state signal in the characteristic region at the bottom of the keyhole as the detection signal. First, the generation position of this signal has a great correlation with the penetration of laser / laser hybrid energy field welding, so it can be used as a direct detection signal, which can avoid the influence of interference factors such as environmental humidity, temperature, and gas flow field when using indirect detection signals. At the same time, the near-infrared spectral band enhancement characteristic of this signal also supports the effective shielding of other harmful signals such as welding arcs and plume in this spectral band, improving the proportion of effective signals in the detection signal and reducing the difficulty of signal analysis.

[0034] The present invention proposes a method of projecting the real image of the thermally excited state signal onto an array sensor chip to obtain the mesoscopic signal, which can achieve full coverage recognition of the characteristic region at the bottom of the keyhole. Since the laser keyhole is always in a fluctuating state during the laser welding process, the position of the penetration characteristic region at the mesoscopic scale will also swing accordingly. Therefore, full coverage recognition of the characteristic region at the bottom of the keyhole is very necessary. One feature of the present invention is that it can adaptively track and recognize the penetration characteristic region, accurately locate the detection signal of the key penetration region, and at the same time, use the high-resolution characteristic of the array sensor to perform high-resolution recognition of the target mesoscopic region, effectively shielding most interference signals and improving the detection reliability.

[0035] The present invention uses the deep learning method in the machine learning method, allowing the computer to autonomously analyze the location of the penetration characteristic region at the bottom of the keyhole through a large amount of data training, effectively identifying the different mesoscopic signal polarization characteristics in the four penetration states of non-penetration, micro-penetration, moderate penetration, and over-penetration, that is, the signal characteristics of the rapid switching between high-amplitude signals and low-amplitude signals formed by the rapid alternation of the opening and closing of the keyhole bottom opening. The duty cycle and switching frequency of the signal switching have a relatively direct correlation with the four penetration states, and through a large amount of data analysis, interference signals are effectively shielded and the influence of signal timing fluctuations is avoided, realizing reliable recognition of the penetration state.

[0036] The present invention proposes that for the transient signal of the penetration signal polarization characteristic, a normalization processing method needs to be adopted during signal preprocessing to eliminate the influence of the dominant data attributes of a large magnitude on the recognition sensitivity of the low-amplitude penetration signal, and it can also improve the problem of slow iterative convergence speed caused by the difference in data magnitude. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0038] Figure 1 Schematic assembly diagram of the detection device; 1. Penetration detection device housing; 2. Substrate; 3. Array or image sensor; 4. Narrowband filter; 5. Three-dimensional fine adjustment mechanism

[0039] Figure 2 Flowchart of the method of the present invention;

[0040] Figure 3 Flowchart of the artificial intelligence detection method. Specific embodiments

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation of the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0043] In the description of the present invention, it should be noted that unless otherwise clearly specified and limited, the terms "installation", "connection", "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. Specific embodiment 1:

[0045] According to Figures 1 to 3 As shown, the specific optimized technical solution adopted by the present invention to solve the above technical problems is: The present invention relates to an on-line detection method for laser welding penetration based on the identification of mesoscopic signal arrays.

[0046] A penetration detection device, the device comprising: a housing, a substrate, a sensor, a narrowband filter, and a three-dimensional fine adjustment mechanism;

[0047] The connection between the substrate and the laser welding head serves to fix the entire device. There are mounting holes on the substrate for fixing the device housing, the three-dimensional fine adjustment mechanism, and the narrowband filter carrier respectively. The three-dimensional fine adjustment mechanism is connected to the sensor to adjust the three-dimensional spatial position of the sensor induction chip. The narrowband filter is located on the optical path between the optical focusing lens group and the induction surface of the array sensor, and the filtered optical real image can be projected onto the induction surface of the array sensor.

[0048] A method for on-line detection of laser welding penetration. The method is based on a penetration detection device, and the method includes the following steps:

[0049] Step 1: Select the mesoscopic signal in the thermally excited state in the penetration characteristic area of the keyhole as the detection object, and obtain the mesoscopic detection signal;

[0050] Step 2: Preprocess the mesoscopic detection signal to obtain data in a type recognizable by a computer;

[0051] Step 3: According to the preprocessed data, after calibration, divide it into a training set, a test set, and a validation set;

[0052] Step 4: Establish a neural network model, train the weight parameters of the model with the training set data until the result converges, and then adjust the hyperparameters of the model through the validation set;

[0053] Step 5: Test the reliability of the model through the test set;

[0054] Step 6: Call the trained recognition model to analyze the signal characteristics online and extract the welding stability information. Specific Embodiment 2:

[0056] The difference between Embodiment 2 and Embodiment 1 of this application is only that:

[0057] When selecting the mesoscopic signal in the thermally excited state in the penetration characteristic area of the keyhole as the detection object, specifically:

[0058] Select the mesoscopic signal in the thermally excited state in the penetration characteristic area of the keyhole as the detection object, use the principles of optical focusing imaging and spectral transmission to project the clear real image of the thermally excited signal at the bottom of the keyhole onto the induction surface of the array sensor to obtain the mesoscopic detection signal, and use the machine learning method to identify the mesoscopic signal characteristics to obtain the current penetration state information.

[0059] The specific steps are as follows:

[0060] First, select the mesoscopic signal in the thermally excited state in the keyhole penetration feature area as the detection object. Second, use the principles of optical focusing imaging and spectral transmission to project the clear real image of the thermally excited signal at the bottom of the keyhole onto the sensing surface of the array sensor to obtain the mesoscopic detection signal. Then, use the model trained by machine learning methods to identify the characteristics of the mesoscopic signal and obtain the current penetration state information.

[0061] The thermally excited signal at the bottom of the keyhole is a near-infrared signal generated when the laser beam enters the base material, causing the metal at the bottom of the keyhole to rapidly melt and evaporate through intense energy input and being accompanied by high-density energy excitation. Since there must be a moment when the laser beam penetrates the base material during penetration welding, and at this time, the opening at the bottom of the keyhole will cause the rapid attenuation of the excited state signal, this signal has a good correlation with the weld penetration quality.

[0062] The method for collecting the mesoscopic detection signal is as follows: First, through a high-magnification optical focusing lens group with a shooting working distance of at least 0.6 - 1.5 m and a shooting depth of field of 10 mm, extract the clear real image of the feature area at the bottom of the keyhole from the coaxial optical path of the laser welding head. A large enough depth of field can capture the clear real image of the feature area in the fluctuating state without changing the focal length. Second, in the near-infrared spectral band, effectively shield the welding arc, plume, laser beam, and other a large number of welding radiation signals above the keyhole through narrowband filtering, so that the thermally excited signal in the feature area can be effectively separated, thereby greatly reducing the proportion of invalid signals in the detection signal. Then, project the real image of the thermally excited signal onto the sensing surface of an array sensor with a large sensing coverage area and high detection accuracy. The sensing area of the array sensor should be ≥ the feature area to be measured, and the resolution accuracy should be ≤ 10 μm, so as to obtain the mesoscopic signals at different positions in the feature area to be measured. This method can not only obtain the thermally excited signals at all positions in the area to be measured, accurately analyze the location of the key penetration feature signals, but also further improve the proportion of valid signals in the detection data by directly extracting the mesoscopic signals at the key positions, reduce the data volume of signal analysis, and provide data guarantee for the next step of detection signal analysis.

[0063] The artificial intelligence detection method utilizes the fact that the thermally excited state signals in the characteristic region at the bottom of the keyhole follow the essence of the welding thermal reaction and have regularities in the trend characteristics of the signals. Through a large amount of data analysis, signal outliers are avoided, and regular characteristics are captured. At the same time, the accuracy of analysis is improved through the composite recognition of signals in multiple regions. The specific method is as follows: The single-channel / or multi-channel / or all mesoscopic detection signal data collected are first converted into computer-recognizable type data through certain data processing, and then the trained recognition model is called for operation to obtain the correlation information of the penetration characteristics of the current laser welding, and an online diagnosis result is given or a basis for adjusting key process parameters for the welding closed-loop control system is provided, such as the power of the welding laser, the welding speed, the defocus amount, etc.

[0064] The method for establishing the penetration state recognition model is to convert the analysis sample data collected by the array sensor or the image sensor into computer-recognizable type data through certain data processing, and then calibrate the analysis samples with the actual welding penetration characteristics and divide them into a training set, a validation set, and a test set. Then, a neural network model for a 4-classification task of incomplete penetration, micro-penetration, moderate penetration, and over-penetration is constructed by a computer. The penetration recognition model is trained with the data in the training set until the result converges, and then the hyperparameters of the model are optimized through the validation set to obtain the optimal penetration recognition model. Finally, the accuracy and effectiveness of the model are evaluated through the test set data.

[0065] The method for establishing the penetration state recognition model is proposed based on the fluctuation characteristics of the penetration signal. Since the penetration signal depends on the generation of a low-amplitude signal due to the lack of a thermally excited state signal at the notch when the laser beam penetrates the base material during penetration welding, and at the same time, since the opening at the bottom of the keyhole does not exist in a normal state but is always in a rapid alternating form of opening / closing, and when it closes, it is a distinct high-amplitude signal, the penetration signal is a transient signal with bipolar characteristics. That is to say, the signal will quickly switch between low amplitude and high amplitude, and there is a large difference between the two values. Therefore, when training the penetration state recognition model, first, during signal preprocessing, normalization processing needs to be done first, that is, all data is converted into values that vary between 0 and 1. The purpose of doing this is to eliminate the influence of data of different magnitudes, because the difference in data magnitudes will cause the attributes with larger magnitudes to dominate, while the recognition of the penetration signal precisely requires being more sensitive to low-amplitude signals, and normalization processing can also improve the problem of slowdown in the iterative convergence speed caused by differences in data magnitudes. Secondly, after calibrating the processed data set with the actual welding penetration state, it is then divided into a training set, a test set, and a validation set. Thirdly, a neural network model with a 4-classification task of incomplete penetration, partial penetration, moderate penetration, and over-penetration is constructed using a computer. The penetration state recognition model is trained with the data in the training set until the results converge, and the model hyperparameters are adjusted through the validation set. Finally, the accuracy of the model and the effectiveness of the prediction results are evaluated through the test set data.

[0066] The real-time detection method for penetration characteristics is to first preprocess the collected single-channel / or multi-channel / or all mesoscopic detection signal data and convert it into data of a type recognizable by a computer, and then call the trained recognition model for calculation, so that the computer can identify which category among the 4 classifications of incomplete penetration, partial penetration, moderate penetration, and over-penetration the current penetration characteristics of laser welding belong to, and give an online diagnosis result or provide a basis for regulating key process parameters for the welding closed-loop control system, such as the power of the welding laser, the welding speed, the defocus amount, etc.

[0067] The present invention uses the thermally excited state signal in the characteristic region at the bottom of the keyhole as the detection signal. First, the generation position of this signal has a great correlation with the penetration of laser / laser hybrid energy field welding, so it can be used as a direct detection signal, which can avoid being affected by interference factors such as environmental humidity, temperature, and gas flow field when using an indirect detection signal for detection. At the same time, the near-infrared spectral band enhancement characteristic of this signal also supports the effective shielding effect on other harmful signals such as welding plume in this spectral band, improves the proportion of effective signals in the detection signal, and reduces the difficulty of signal analysis.

[0068] The present invention proposes a method of projecting a real image of a thermally excited state signal onto the sensing surface of an array sensor to obtain a mesoscopic signal, which can achieve full coverage recognition of the keyhole fluctuation region, and can effectively address the problem that the mesoscopic penetration signal cannot be effectively extracted due to the position change of the keyhole root region caused by the free swing of the keyhole during laser welding. At the same time, the high-resolution characteristics of the array sensor can be used to perform high-resolution recognition of the target mesoscopic region, effectively shielding most interference signals and improving the detection reliability.

[0069] The present invention adopts the deep learning method in the machine learning method, enabling the computer to autonomously analyze the location of the penetration feature region at the bottom of the keyhole through a large amount of data training, effectively identify the different mesoscopic signal polarization characteristics in the four penetration states of incomplete penetration, micro-penetration, moderate penetration, and over-penetration, and effectively shield interference signals and avoid the influence of signal timing fluctuations through a large amount of data analysis, realizing reliable recognition of the penetration state.

[0070] For the transient signal of the penetration signal polarization characteristic, it is proposed that a normalization processing method needs to be adopted during signal preprocessing to eliminate the influence of the dominant data attributes of a larger magnitude on the recognition sensitivity of the low-amplitude penetration signal, and it can also improve the problem of slowdown of the iterative convergence speed caused by the data magnitude difference. Specific Embodiment Three:

[0072] The difference between the third embodiment of this application and the second embodiment is only that:

[0073] The present invention provides an on-line detection system for laser welding penetration, and the system includes:

[0074] A data acquisition module, which selects the thermally excited mesoscopic signal in the penetration feature region in the keyhole as the detection object to obtain a mesoscopic detection signal;

[0075] A preprocessing module, which preprocesses the mesoscopic detection signal to obtain data in a type recognizable by a computer;

[0076] A calibration module, which divides the data into a training set, a test set, and a validation set after calibration according to the preprocessed data;

[0077] A model establishment module, which establishes a neural network model, trains the weight parameters of the model through the training set data until the result converges, then adjusts the model hyperparameters through the validation set, and tests the reliability of the model through the test set;

[0078] An on-line detection module, which calls the trained model to perform real-time analysis of the penetration state and extract welding stability information. Specific Embodiment Four:

[0080] The only difference between the fourth embodiment of the present application and the third embodiment is that:

[0081] The present invention provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to implement an online detection method for laser welding penetration.

[0082] The specific steps are as follows:

[0083] Firstly, the thermally excited state mesoscopic signal in the characteristic melt-through area inside the keyhole is selected as the detection object. Secondly, the clear real image of the thermally excited state signal at the bottom of the keyhole is projected onto the sensing surface of the array sensor by using the principle of optical focusing imaging and spectral transmission to obtain the mesoscopic detection signal. Then, the model trained by machine learning method is used to identify the characteristics of the mesoscopic signal and obtain the current melt-through state information.

[0084] The thermally excited state signal at the bottom of the keyhole is a near-infrared signal generated by the laser beam entering the base material, causing the metal at the bottom of the keyhole to rapidly melt and evaporate due to intense energy input, accompanied by high-density energy excitation. Since there will inevitably be a moment when the laser beam penetrates the base material during full penetration welding, the opening at the bottom of the keyhole will cause the excited state signal to decay rapidly at this time. Therefore, this signal has a good correlation with the penetration quality of the weld.

[0085] The acquisition method of mesoscopic detection signal is as follows: first, a high-power optical focusing lens group with a shooting working distance of at least 0.6-1.5m and a shooting depth of field of 10mm is used to extract a clear real image of the characteristic area at the bottom of the keyhole from the coaxial optical path of the laser welding head. A sufficiently large depth of field can capture a clear real image of the characteristic area under the fluctuating state without changing the focal length. Secondly, in the near-infrared spectrum, the welding arc, plume, laser beam and other large amounts of welding radiation signals above the keyhole are effectively shielded by narrow-band filtering, so that the thermally excited state signal in the characteristic area can be effectively separated, thereby greatly reducing the detection signal. Then, the real image of the thermally excited state signal is projected onto the sensing surface of an array sensor with a large sensing coverage area and high detection accuracy. The sensing area of the array sensor should be ≥ the feature area to be measured, and the resolution accuracy should be ≤10μm. In this way, mesoscopic signals at different positions in the feature area to be measured are obtained. This method can not only obtain the thermal excitation signals at all positions in the area to be measured and accurately analyze the position of the key feature signal of penetration, but also directly extract the mesoscopic signals at key positions to further increase the proportion of effective signals in the detection data, reduce the data volume of signal analysis, and provide data guarantee for the next step of detection signal analysis.

[0086] The artificial intelligence detection method makes use of the fact that the thermal excitation state signals in the characteristic region at the bottom of the keyhole follow the essence of the welding thermal reaction and have regularity in the trend characteristic changes of the signals. Through a large amount of data analysis, signal outliers are avoided and regular features are captured. At the same time, the accuracy of analysis is improved by the composite recognition of signals in multiple regions. The specific method is as follows: The single-channel / or multi-channel / or all mesoscopic detection signal data collected are first converted into computer-recognizable type data through certain data processing, and then the trained recognition model is called for operation to obtain the correlation information of the penetration characteristics of the current laser welding, and an online diagnosis result is given or a basis for the regulation of key process parameters is provided for the welding closed-loop control system, such as the power of the welding laser, the welding speed, the defocus amount, etc.

[0087] The method for establishing the penetration state recognition model is to convert the analysis sample data collected by the array sensor or the image sensor into computer-recognizable type data through certain data processing, and then calibrate the analysis samples with the actual welding penetration characteristics and divide them into a training set, a validation set and a test set. Then, a neural network model for the 4-classification task of incomplete penetration, micro-penetration, moderate penetration, and over-penetration is constructed by a computer. The penetration recognition model is trained with the data in the training set until the result converges, and then the model hyperparameters are optimized through the validation set to obtain the optimal penetration recognition model. Finally, the accuracy and effectiveness of the model are evaluated through the test set data.

[0088] The method for establishing the penetration state recognition model is proposed based on the fluctuation characteristics of the penetration signal. Since the penetration signal depends on the generation of a low-amplitude signal due to the lack of a thermally excited state signal at the notch when the laser beam penetrates the base material during penetration welding, and at the same time, since the opening at the bottom of the keyhole does not exist in a normal state but is always in a rapid alternating form of opening / closing, and when closed, it is a distinct high-amplitude signal, the penetration signal is a transient signal with bipolar characteristics. That is to say, the signal will quickly switch between low amplitude and high amplitude, and there is a large difference between the two values. Therefore, when training the penetration state recognition model, first, during signal preprocessing, normalization processing needs to be done first, that is, all data is converted into values that vary between 0 and 1. The purpose of this is to eliminate the influence of data of different magnitudes, because the difference in data magnitudes will cause the attributes with larger magnitudes to dominate, while the recognition of the penetration signal precisely requires being more sensitive to low-amplitude signals, and normalization processing can also improve the problem of slowdown in the iterative convergence speed caused by the difference in data magnitudes. Secondly, after calibrating the processed data set with the actual welding penetration state, it is then divided into a training set, a test set, and a validation set. Thirdly, use a computer to build a neural network model for 4 classification tasks of incomplete penetration, micro-penetration, moderate penetration, and over-penetration. Train the penetration state recognition model with the data in the training set until the results converge, adjust the model hyperparameters through the validation set, and finally, evaluate the accuracy of the model and the effectiveness of the prediction results through the test set data.

[0089] The real-time detection method for penetration characteristics is to first preprocess the collected single-channel / or multi-channel / or all mesoscopic detection signal data and convert it into data of a type recognizable by a computer, and then call the trained recognition model for operation, so that the computer can identify which category among the 4 classifications of incomplete penetration, micro-penetration, moderate penetration, and over-penetration the current penetration characteristics of laser welding belong to, give an online diagnosis result, or provide a basis for adjusting key process parameters for the welding closed-loop control system, such as the power of the welding laser, the welding speed, the defocus amount, etc. Specific Embodiment Five:

[0091] The difference between Embodiment Five and Embodiment Four of this application is only that:

[0092] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, a laser welding penetration on-line detection method is implemented.

[0093] The specific steps are as follows:

[0094] First, select the mesoscopic signal in the thermally excited state in the keyhole penetration feature area as the detection object. Second, use the principles of optical focusing imaging and spectral transmission to project the clear real image of the thermally excited signal at the bottom of the keyhole onto the sensing surface of the array sensor to obtain the mesoscopic detection signal. Then, use the model trained by machine learning methods to identify the mesoscopic signal features and obtain the current penetration state information.

[0095] The thermally excited signal at the bottom of the keyhole is a near-infrared signal generated when the laser beam enters the base material, causing the metal at the bottom of the keyhole to rapidly melt and evaporate through intense energy input and being accompanied by high-density energy excitation. Since there must be a moment when the laser beam penetrates the base material during penetration welding, and at this time, the opening at the bottom of the keyhole will cause the rapid attenuation of the excited state signal, this signal has a good correlation with the weld penetration quality.

[0096] The method for collecting the mesoscopic detection signal is as follows: First, use a high-magnification optical focusing lens group with a shooting working distance of at least 0.6 - 1.5 m and a shooting depth of field of 10 mm to extract the clear real image of the feature area at the bottom of the keyhole from the coaxial optical path of the laser welding head. A large enough depth of field can capture the clear real image of the feature area in the fluctuating state without changing the focal length. Second, in the near-infrared spectral band, effectively shield the welding arc, plume, laser beam, and other a large number of welding radiation signals above the keyhole through narrowband filtering, so that the thermally excited signal in the feature area can be effectively separated, thereby greatly reducing the proportion of invalid signals in the detection signal. Then, project the real image of the thermally excited signal onto the sensing surface of an array sensor with a large sensing coverage area and high detection accuracy. The sensing area of the array sensor should be ≥ the feature area to be measured, and the resolution accuracy should be ≤ 10 μm, so as to obtain the mesoscopic signals at different positions in the feature area to be measured. This method can not only obtain the thermally excited signals at all positions in the area to be measured, accurately analyze the position of the key penetration feature signals, but also further improve the proportion of valid signals in the detection data by directly extracting the mesoscopic signals at the key positions, reduce the data volume of signal analysis, and provide data guarantee for the next step of detection signal analysis.

[0097] The artificial intelligence detection method utilizes the fact that the thermally excited state signals in the characteristic region at the bottom of the keyhole follow the essence of the welding thermal reaction and have regularity in the trend characteristic changes of the signals. By analyzing a large amount of data, individual signal cases are avoided and regular features are captured. At the same time, the accuracy of analysis is improved through the composite recognition of signals in multiple regions. The specific method is as follows: The single-channel / or multi-channel / or all mesoscopic detection signal data collected are first converted into computer-recognizable type data through certain data processing, and then the trained recognition model is called for operation to obtain the correlation information of the penetration characteristics of the current laser welding, and an online diagnosis result is given or a basis for adjusting key process parameters such as the power of the welding laser, welding speed, defocus amount, etc. is provided for the welding closed-loop control system.

[0098] The method for establishing the penetration state recognition model is to convert the analysis sample data collected by the array sensor or image sensor into computer-recognizable type data through certain data processing, and then calibrate the analysis samples with the actual welding penetration characteristics and divide them into a training set, a validation set, and a test set. Then, a neural network model for the 4-classification task of incomplete penetration, micro-penetration, moderate penetration, and over-penetration is constructed by a computer. The penetration recognition model is trained with the data in the training set until the result converges, and then the hyperparameters of the model are optimized through the validation set to obtain the optimal penetration recognition model. Finally, the accuracy and effectiveness of the model are evaluated through the test set data.

[0099] The method for establishing the penetration state recognition model is proposed based on the fluctuation characteristics of the penetration signal. Since the penetration signal depends on the generation of a low-amplitude signal due to the lack of a thermally excited state signal at the notch when the laser beam penetrates the base material during penetration welding, and at the same time, since the opening at the bottom of the keyhole does not exist in a normal state but is always in a rapid opening / closing alternating form, and when it closes, it is a distinct high-amplitude signal, the penetration signal is a transient signal with bipolar characteristics. That is to say, the signal will quickly switch between low amplitude and high amplitude, and there is a large difference between the two values. Therefore, when training the penetration state recognition model, first, during signal preprocessing, normalization processing needs to be done first, that is, all data is converted into values that vary between 0 and 1. The purpose of doing this is to eliminate the influence of data of different magnitudes, because the difference in data magnitudes will cause the attribute with a larger magnitude to dominate, while the recognition of the penetration signal precisely requires being more sensitive to low-amplitude signals, and normalization processing can also improve the problem of the slowdown of the iterative convergence speed caused by the difference in data magnitudes. Secondly, after calibrating the processed data set with the actual welding penetration state, it is then divided into a training set, a test set, and a validation set. Thirdly, a neural network model with a 4-classification task of incomplete penetration, micro-penetration, moderate penetration, and over-penetration is constructed using a computer. The penetration state recognition model is trained with the data in the training set until the result converges, and the hyperparameters of the model are adjusted through the validation set. Finally, the accuracy of the model and the effectiveness of the prediction results are evaluated through the test set data.

[0100] The real-time detection method for penetration characteristics is to first preprocess the collected single-channel / or multi-channel / or all mesoscopic detection signal data and convert it into data of a type recognizable by a computer, and then call the trained recognition model for calculation, so that the computer can identify which category among the 4-classifications of incomplete penetration, micro-penetration, moderate penetration, and over-penetration the current laser welding penetration characteristics belong to, give an online diagnosis result, or provide a basis for regulating key process parameters for the welding closed-loop control system, such as the power of the welding laser, the welding speed, the defocus amount, etc.

[0101] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples. Furthermore, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of such features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined. Any process or method description represented in the flowchart or described in other ways herein can be understood to represent a module, segment, or part of code including one or more N executable instructions for implementing a customized logic function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions can be executed in a manner that is not in the order shown or discussed, including in a substantially simultaneous manner according to the functions involved or in the reverse order, which should be understood by those skilled in the art to which the embodiments of the present invention pertain. The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or N wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM).In addition, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing if necessary, and then stored in a computer memory. It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0102] The above description is only a preferred embodiment of an on-line detection method for laser welding penetration based on mesoscopic signal array recognition. The protection scope of an on-line detection method for laser welding penetration based on mesoscopic signal array recognition is not limited to the above embodiments. Any technical solutions falling within this concept belong to the protection scope of the present invention. It should be noted that for those skilled in the art, several improvements and variations made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An on-line detection method for laser welding penetration, the method being based on a penetration detection device, the device comprising: Housing, substrate, sensor, narrowband filter and three-dimensional fine-tuning mechanism; The substrate is connected to the laser welding head to fix the entire device. The substrate has mounting holes for fixing the device housing, the three-dimensional fine-tuning mechanism, and the narrow-band filter carrier respectively. The three-dimensional fine-tuning mechanism is connected to the sensor to adjust the three-dimensional spatial position of the sensor sensing chip. The narrow-band filter is located in the optical path between the optical focusing lens group and the array sensor sensing surface. The filtered optical real image can be projected onto the array sensor sensing surface. The method is characterized in that: Step 1: Select the thermally excited state mesoscopic signal of the melt-through characteristic area in the keyhole as the detection object and obtain the mesoscopic detection signal; Step 2: Preprocess the mesoscopic detection signal to obtain computer-recognizable type data; Step 3: Based on the preprocessed data, the data is calibrated and divided into training set, test set and validation set; Step 4: Build a neural network model, train the model's weight parameters using the training set data until the results converge, and then adjust the model hyperparameters using the validation set; Step 5: Test the reliability of the model using the test set; Step 6: Call the trained recognition model to analyze signal characteristics online and extract welding stability information; The acquisition method of mesoscopic detection signal is: S1. A high-power optical focusing lens system with a working distance of at least 0.6-1.5 m and a depth of field of 10 mm is used to capture a clear real image of the characteristic area at the bottom of the keyhole from the coaxial optical path of the laser welding head. The depth of field is large enough to capture a clear real image of the characteristic area under the fluctuating state without changing the focal length. S2. In the near-infrared spectrum, the welding arc, plume, laser beam, and other large amounts of welding radiation signals above the keyhole are effectively shielded by narrow-band filtering, so that the thermally excited state signals in the characteristic area can be effectively separated, thereby significantly reducing the proportion of invalid signals in the detection signal; S3. Project the real image of the thermally excited state signal onto the sensing surface of an array sensor with a large sensing coverage area and high detection accuracy. The sensing area of the array sensor should be ≥ the characteristic area to be measured, and the resolution accuracy should be ≤10μm, thereby obtaining mesoscopic signals at different positions in the characteristic area to be measured.

2. The method according to claim 1, wherein: The thermally excited state mesoscopic signal in the characteristic melt-through area in the keyhole is selected as the detection object. The clear real image of the thermally excited state signal at the bottom of the keyhole is projected onto the sensing surface of the array sensor using the principle of optical focusing imaging and spectral transmission to obtain the mesoscopic detection signal. The machine learning method is used to identify the characteristics of the mesoscopic signal and detect the current training model. During the detection, the trained model is called to perform real-time analysis to give the detection results and obtain the current melt-through state information.

3. The method according to claim 2, wherein: The thermally excited state signal at the bottom of the keyhole is a near-infrared signal generated when the laser beam enters the base material and the metal at the bottom of the keyhole rapidly melts and evaporates through intense energy input, accompanied by high-density energy excitation.

4. The method according to claim 3, characterized in that: The specific steps for obtaining the recognition model are: During signal preprocessing, normalization is performed first to convert all data into values that vary between 0 and 1. After calibrating the processed dataset with the actual weld penetration state, it is then divided into a training set, a test set, and a validation set. A neural network model for a 4-classification task of incomplete penetration, slight penetration, moderate penetration, and excessive penetration is constructed using a computer. The weight parameters of the model are trained with the training set data until the results converge. Then, the hyperparameters of the model are adjusted through the validation set. Finally, the reliability of the model is tested through the test set to obtain the optimal penetration model.

5. The method according to claim 4, characterized in that: The single-channel / or multi-channel / or all mesoscopic detection signal data collected is first preprocessed and converted into data of a type recognizable by a computer, and then the trained recognition model is called for operation to enable the computer to identify which category among the 4-classifications of incomplete penetration, slight penetration, moderate penetration, and excessive penetration the current laser welding penetration feature belongs to, and an online diagnosis result is given or a basis for regulating key process parameters is provided for the welding closed-loop control system.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by a processor to implement the method according to any one of claims 1-5.

7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, the method according to any one of claims 1-5 is implemented.

Images