Error correction method and system for laser welding penetration based on OCT (optical coherence tomography)

Abstract

The invention discloses an OCT-based laser welding penetration error correction method and system, and the method comprises the steps: constructing an OCT penetration detection device to collect the keyhole depth information of a workpiece in the laser penetration welding process, and obtaining the keyhole depth information of the workpiece; performing filtering and moving average processing on the keyhole depth information of the workpiece based on a moving average algorithm of percentile filtering to obtain a keyhole depth change curve of the workpiece; and error correction is carried out on the keyhole depth change curve of the workpiece through a long and short term memory neural network model to obtain a corrected keyhole depth change curve of the workpiece. By means of the method and device, interference generated by OCT measurement can be reduced, noise signals are formed, the interpretability of OCT fusion depth detection data is improved, and therefore the precision of laser welding fusion depth online measurement of OCT is improved. The OCT-based laser welding penetration error correction method and the OCT-based laser welding penetration error correction system can be widely applied to the technical field of laser welding penetration detection.

Description

Technical Field

[0001] The present invention relates to the technical field of laser welding penetration detection, and in particular to an error correction method and system for laser welding penetration based on OCT. Background Art

[0002] Laser welding is a method of tightly connecting the workpiece materials by irradiating the laser beam to the surface of the workpiece to heat and melt the material, so that the workpiece materials are fused together. However, due to the shape and size of the welding head, the structural shape of the welding workpiece itself, and the interference of the fixture, the welding head often needs to be deflected in actual engineering applications. Related studies have shown that the deflection angle of the laser welding head will affect the transient behavior of the molten pool and the keyhole. For example, when the laser welding cast is deflected, the key will grow obliquely along the laser incident direction, which will directly affect the forming quality of the weld. The OCT penetration depth detection system is a direct measurement method that forms scattered reflections by the OCT measurement beam entering the keyhole. The stability of the keyhole is crucial to the accuracy of the OCT penetration depth measurement. However, when the laser incident angle is tilted, the keyhole grows obliquely, resulting in the key depth measured by the OCT to be the tilted depth, which causes the penetration depth measurement error. At the same time, when the keyhole grows obliquely, the tip of the keyhole will bend to form a buttstock, so that the OCT keyhole measurement beam cannot reach the bottom of the keyhole tip and measures on the keyhole wall, resulting in a measurement error.

[0003] In the actual production process of laser welding, due to mechanical assembly errors and thermal deformation of the plates, overlapping gaps will inevitably appear between stainless steel plates. The existence of overlapping gaps will affect the appearance, microstructure and mechanical properties of the weld, and will affect the molten pool surge and keyhole morphology during welding, thereby affecting the OCT penetration detection process. The existence of overlapping gaps will affect the energy propagation of the laser welding laser beam, causing the center of mass of the keyhole to move backward during the actual welding process. During laser lap welding, due to the existence of the gap between the plates, the molten metal will flow into the gap and fill the gap. The backward movement of the center of mass of the keyhole will cause the OCT keyhole measurement beam to directly measure the front wall of the keyhole, causing the measurement beam to reflect multiple times inside the keyhole, thereby causing the optical path difference of the measurement light to increase and bring about measurement errors. In addition, the existing laser welding penetration measurement method basically does not have the detection error correction function, and only performs ordinary filtering and noise elimination on the detection signal, so the measurement error is large. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide an error correction method and system for laser welding penetration based on OCT, which can reduce the interference generated by OCT measurement and the formation of noise signals, improve the interpretability of OCT penetration detection data, and thus improve the accuracy of OCT online measurement of laser welding penetration.

[0005] The first technical solution adopted by the present invention is: an error correction method for laser welding penetration based on OCT, comprising the following steps:

[0006] An OCT penetration depth detection device is constructed to collect the keyhole depth information of the workpiece during the laser deep penetration welding process to obtain the keyhole depth information of the workpiece;

[0007] Based on the moving average algorithm of percentile filtering, filtering and moving average processing are performed on the keyhole depth information of the workpiece to obtain a keyhole depth variation curve of the workpiece;

[0008] The error correction of the keyhole depth variation curve of the workpiece is performed through a long short-term memory neural network model to obtain a corrected workpiece keyhole depth variation curve.

[0009] Further, the OCT penetration detection device includes an optical coherence tomography device and a laser welding device, and the optical coherence tomography device and the laser welding device are connected through coaxial coupling, wherein the optical coherence tomography device includes a broadband light source generator, a beam splitter, a measuring arm, a reference arm, a reflector and a spectrometer, wherein:

[0010] The broadband light source generator is used to emit broadband light source;

[0011] The beam splitter is used to split the broadband light source to obtain a first branch light source and a second branch light source;

[0012] The measuring arm is used to collect and process information on the keyhole depth of the workpiece according to the first branch light source to obtain a keyhole depth spectrum signal;

[0013] The reference arm is used to collect information on the upper surface of the workpiece according to the second branch light source to obtain a spectral signal of the upper surface profile of the workpiece;

[0014] The reflector is used to transmit the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal to the spectrometer;

[0015] The spectrometer is used to analyze the interference signal between the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal to obtain the keyhole depth information of the workpiece.

[0016] Further, the step of constructing an OCT penetration depth detection device to collect keyhole depth information of a workpiece during laser deep penetration welding to obtain the keyhole depth information of the workpiece specifically includes:

[0017] Construct an OCT penetration detection device;

[0018] Based on the OCT penetration depth detection device, the keyhole depth information of the workpiece during the laser deep penetration welding process is collected to obtain a keyhole depth spectrum signal and a workpiece upper surface profile spectrum signal;

[0019] The interference signal between the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal is obtained, and a fast Fourier transform is performed to obtain the keyhole depth information of the workpiece.

[0020] Further, the step of filtering and moving average processing the keyhole depth information of the workpiece by the moving average algorithm based on percentile filtering to obtain the keyhole depth variation curve of the workpiece specifically includes:

[0021] Inputting the keyhole depth information of the workpiece into a percentile filter for filtering, and obtaining a filtered keyhole depth variation curve, wherein the filtered keyhole depth variation curve represents a curve composed of several percentile values;

[0022] The filtered keyhole depth variation curve is subjected to low-pass filtering by a moving average algorithm to obtain the keyhole depth variation curve of the workpiece.

[0023] Further, the step of inputting the keyhole depth information of the workpiece into a percentile filter for filtering to obtain a filtered keyhole depth variation curve specifically includes:

[0024] Define the moving center flat window size τ and percentile p of the percentile filter;

[0025] Constructing a keyhole depth data set according to the keyhole depth information of the workpiece, wherein the keyhole depth data set is represented by all measurement values ​​before τ / 2 and after τ / 2;

[0026] Arranging the keyhole depth data set according to a preset sorting rule to obtain an arranged keyhole depth data set;

[0027] The position of the percentile p in the arranged keyhole depth data set is determined, the percentile value is obtained and replaced, and all elements in the arranged keyhole depth data set are traversed to obtain a filtered keyhole depth change curve.

[0028] Further, the expression of the filtered keyhole depth variation curve is specifically as follows:

[0029] D p (t)=F[D(t)],t∈L

[0030] In the above formula, D p (t) represents the percentile value, F[D(t)] represents the percentile function, t represents the time, L represents the keyhole depth dataset, and p represents the percentile.

[0031] Further, the expression of the keyhole depth variation curve of the workpiece is specifically as follows:

[0032]

[0033] In the above formula, D ′ (t) represents the time series of the keyhole depth of the workpiece, T represents the moving window size of the moving average algorithm, △t represents the period between two depth measurements, and t represents the time.

[0034] The second technical solution adopted by the present invention is: an error correction system for laser welding penetration based on OCT, comprising:

[0035] The first module is used to construct an OCT penetration depth detection device to collect keyhole depth information during laser deep penetration welding of the workpiece to obtain the keyhole depth information of the workpiece;

[0036] The second module is used to filter and perform moving average processing on the keyhole depth information of the workpiece based on a moving average algorithm of percentile filtering to obtain a keyhole depth variation curve of the workpiece;

[0037] The third module is used to perform error correction on the keyhole depth variation curve of the workpiece through a long short-term memory neural network model to obtain a corrected keyhole depth variation curve of the workpiece.

[0038] The beneficial effects of the method and system of the present invention are as follows: the present invention collects keyhole depth information during the laser deep melting welding process of the workpiece by constructing an OCT penetration detection device, further filters and performs moving average processing on the keyhole depth information of the workpiece based on a moving average algorithm of percentile filtering, and inputs complex initial keyhole depth data into a percentile filter for filtering processing. Since the penetration depth changes more gradually than the keyhole changes, the values ​​after percentile filtering are processed by moving average to obtain a keyhole depth change curve, thereby reducing interference and noise signal formation caused by OCT measurement, and improving the interpretability of OCT penetration detection data. Finally, the keyhole depth change curve of the workpiece is error corrected through a long short-term memory neural network model, thereby improving the accuracy of online measurement of laser welding penetration by OCT. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the steps of an error correction method for laser welding penetration based on OCT of the present invention;

[0040] Figure 2 It is a structural block diagram of an error correction system for laser welding penetration based on OCT of the present invention;

[0041] Figure 3is a schematic diagram of an OCT penetration detection device provided by a specific embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of error correction of laser welding penetration provided by a specific embodiment of the present invention;

[0043] Figure 5 It is a schematic diagram of the structure of a long short-term memory neural network model provided by a specific embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0045] First of all, it should be noted that laser welding, as a new type of connection method, has the advantages of high laser power density, fast welding speed, strong adaptability and high automation level, and has been widely used in the automotive industry. In the actual laser welding process, there are often uncontrollable factors, such as the interference between the fixture and the laser welding head, the need to deflect the welding head, assembly errors and thermal deformation of the workpiece leading to gaps between plates, etc., and various defects may occur in the welded joints. Among them, the biggest influence on the mechanical strength of the welded joint is the insufficient weld depth, especially for the lap weld joint. Therefore, it is of great significance to conduct online detection of laser welding penetration to ensure that the penetration of the welded joint meets the process requirements for actual industrial production.

[0046] Optical coherence tomography (OCT), also known as online coherence imaging or low-coherence interferometry, has the advantages of non-contact deployment, high dynamic range, micron resolution, and insensitivity to all other optical distortions. OCT technology can capture two-dimensional or three-dimensional images with a resolution of microns from the optical reflection and scattering signals of the object. OCT systems usually use a special laser beam different from the welding laser beam, with a wavelength of about 0.85μm or 1.5μm. Among the various laser welding penetration online detection methods, the OCT-based penetration online detection method has its unique advantages. This method emits active measurement light into the bottom of the keyhole to form a reflection, thereby realizing direct measurement of the keyhole depth. However, the violent oscillation of the keyhole and the molten pool during laser deep penetration welding will interfere with the OCT measurement and form a noise signal, resulting in poor interpretability of the OCT penetration detection data. Moreover, in the actual laser welding process, the keyhole depth is not equal to the penetration depth. The change of the keyhole is instantaneous, while the actual penetration depth is lagging and gentle compared to the actual keyhole depth change. Therefore, the OCT raw data needs to be filtered accordingly to extract the accurate melting depth curve.

[0047] Depth of penetration is an important evaluation index of weld quality. Real-time and accurate monitoring of the depth of penetration during laser welding is an important step in obtaining high-quality laser welds. During deep penetration welding of laser welding, due to the high laser power density, the laser irradiates the workpiece material, causing the workpiece material to melt and evaporate and form a piston effect, thereby forming a keyhole, and the keyhole depth can be approximately equal to the depth of penetration of laser deep penetration welding. Due to the advantages of non-contact deployment, high dynamic range, micron resolution, and insensitivity to all other optical distortions, optical coherence tomography provides a solution for directly measuring the depth of keyholes in laser deep penetration welding. However, in actual working conditions, various errors will be introduced into the detection of laser welding depth of penetration. Therefore, the above errors need to be corrected accordingly to achieve accurate measurement of the depth of penetration.

[0048] Based on this, the embodiment of the present invention collects the keyhole depth information in the laser deep penetration welding process in real time through the coaxial OCT online penetration detection system. In view of the problem that the violent oscillation of the keyhole and the molten pool during the laser deep penetration welding process will interfere with the OCT measurement and form a noise signal, resulting in poor interpretability of the OCT penetration detection data, a penetration curve extraction method based on the percentile filter and moving average algorithm is proposed. In view of the deflection of the laser welding head (that is, the processing laser beam is not perpendicular to the processing material) and the existence of an overlap gap during laser lap welding, which brings measurement problems to the OCT online penetration detection system, a correction method for online penetration monitoring based on a long short-term memory network is proposed.

[0049] Reference Figure 1The present invention provides an error correction method for laser welding penetration based on OCT, the method comprising the following steps:

[0050] S100, constructing an OCT penetration depth detection device to collect keyhole depth information of a workpiece during laser deep penetration welding, thereby obtaining keyhole depth information of the workpiece;

[0051] It should be noted that the OCT penetration detection device constructed in the embodiment of the present invention includes an optical coherence tomography device and a laser welding device, and the optical coherence tomography device and the laser welding device are connected through coaxial coupling, wherein the optical coherence tomography device includes a broadband light source generator, a beam splitter, a measuring arm, a reference arm, a reflector and a spectrometer, wherein:

[0052] The broadband light source generator is used to emit a broadband light source; the beam splitter is used to split the broadband light source to obtain a first branch light source and a second branch light source; the measuring arm is used to collect information on the keyhole depth of the workpiece according to the first branch light source to obtain a keyhole depth spectrum signal; the reference arm is used to collect information on the upper surface of the workpiece according to the second branch light source to obtain a workpiece upper surface profile spectrum signal; the reflector is used to transmit the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal to the spectrometer; the spectrometer is used to analyze the interference signal between the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal to obtain the keyhole depth information of the workpiece.

[0053] like Figure 3 As shown, the workpiece to be welded is mounted on a work platform, and the laser beam emitted by the laser is transmitted through optical fiber and lens and focused on the surface of the material. The material absorbs the laser energy to cause melting or even vaporization, thereby achieving the purpose of material processing. In order to achieve the above-mentioned solution to the above-mentioned technical requirements, the present invention proposes an online penetration detection system that coaxially couples an optical coherence tomography system with a laser welding system. The OCT penetration detection system mainly includes components such as a broadband light source (superluminescent diode, SLD), a beam splitter, an object arm (measurement arm), a reference arm and a spectrometer. The measurement principle is as follows Figure 3As shown in the figure, the SLD emits a broadband light source, and a part of the light is guided to the reference path through the beam splitter, which is reflected back to the spectrometer by the reflector. Another part of the light output by the SLD is directed to the processing head, and is focused on the workpiece after being superimposed with the processing laser beam through the optical collimator. In the applied OCT system, two optical reflective lenses with adjustable angles are used to split the measuring light of the measuring arm into two beams. One beam is coaxial with the laser welding beam, and it reaches the bottom of the keyhole with the laser beam to form reflected light and then returns to the original path; the other beam is directed to the upper surface of the welded workpiece by adjusting the reflector angle, and it forms a reflection on the upper surface of the workpiece and returns to the original path. The surface measurement beam plays the role of "keyhole mapping" in the entire OCT penetration detection system, which helps to determine the optimal position of the OCT keyhole measurement beam; the spectral analysis unit decomposes the interference signal, transmits the signal to the computer for data processing, and obtains the weld depth information of the laser processing area through analysis and processing. Through data filtering processing, the penetration curve consistent with the metallographic photo is extracted to realize real-time monitoring of the penetration depth of the laser welding process.

[0054] Furthermore, in the present embodiment, the keyhole depth information of the workpiece during the laser deep melting welding process is first collected based on the OCT penetration detection device to obtain a keyhole depth spectral signal and a workpiece upper surface profile spectral signal; then, the interference signal between the keyhole depth spectral signal and the workpiece upper surface profile spectral signal is obtained, and a fast Fourier transform is performed to obtain the keyhole depth information of the workpiece.

[0055] Specifically, the collected spectral signals are analyzed, and the collected keyhole depth spectral signals and surface profile spectral signals are Fourier transformed, and the calculation is as follows:

[0056]

[0057] The input signal f(x) here is the collected keyhole depth spectrum signal, and F(w) is the frequency domain function of the function f(x) after Fourier transform. After fast Fourier transform, the corresponding depth information can be obtained.

[0058] S200, filtering and moving average processing are performed on the keyhole depth information of the workpiece based on a percentile filtering moving average algorithm to obtain a keyhole depth variation curve of the workpiece;

[0059] S210, inputting the keyhole depth information of the workpiece into a percentile filter for filtering, to obtain a filtered keyhole depth variation curve, wherein the filtered keyhole depth variation curve represents a curve composed of several percentile values;

[0060] Specifically, a moving center flat window size τ and a percentile p of a percentile filter are defined; a keyhole depth data set is constructed according to the keyhole depth information of the workpiece, and the keyhole depth data set is represented by all measurement values ​​before τ / 2 and after τ / 2; the keyhole depth data set is arranged according to a preset sorting rule to obtain an arranged keyhole depth data set; the position of the percentile p in the arranged keyhole depth data set is determined, the percentile value is obtained and replaced, all elements in the arranged keyhole depth data set are traversed, and a filtered keyhole depth change curve is obtained.

[0061] In this embodiment, the corresponding keyhole depth data information is input into the percentile filter. The percentile filter is calculated in the form of a moving center flat window. There are two main parameters that need to be set: the size of the moving center flat window τ and the percentile p. Assuming that at a certain time t, the total measurement value L within the τ / 2 time period before and after t is:

[0062]

[0063] In the above formula, the size of the filter window is τ, and Δt is the period between two depth measurements. The percentile filter function sorts all the measured values ​​D(t) of the set L, which means that the measured depth value at time t is replaced by the value D of the p percentile of the set L. p (t) Replacement, that is, p% of the measurements in the set L are less than D p (t), and finally all the measured values ​​are traversed by moving the central flat window to obtain the percentile filtering curve.

[0064] D p (t)=F[D(t)],t∈L

[0065] In the above formula, D p (t) represents the percentile value, F[D(t)] represents the percentile function, t represents the time, L represents the keyhole depth dataset, and p represents the percentile.

[0066] That is, F[D(t)] is equivalent to a percentile function, which performs percentile filtering on all depth measurements within the τ window (i.e., all elements in the set L).

[0067] S220 , performing low-pass filtering on the filtered keyhole depth variation curve by using a moving average algorithm to obtain the keyhole depth variation curve of the workpiece.

[0068] Specifically, in order to deal with the possible spike interference in the percentile filter curve, the moving average algorithm (MA) is used to perform low-pass filtering on the curve. The moving average algorithm is calculated in a central flat window manner. It can be used to remove noise from the data, reduce high-frequency disturbances in the data, and retain useful low-frequency trends. The parameters that need to be set for the moving average are the moving window size T. Assume that the keyhole value obtained by the percentile filter algorithm at any time t is D p (t), then the keyhole depth value set obtained by the percentile filtering algorithm within T / 2 time before and after time t is expressed as:

[0069]

[0070] In the above formula, D′(t) represents the time series of the keyhole depth of the workpiece, T represents the moving window size of the moving average algorithm, △t represents the period between two depth measurements, and t represents the time.

[0071] S300, performing error correction on the keyhole depth variation curve of the workpiece by using a long short-term memory neural network model to obtain a corrected keyhole depth variation curve of the workpiece.

[0072] Specifically, the keyhole depth time series D′(t) is used as the input layer neuron, and the actual melting depth time series obtained by metallographic test is used as the output layer neuron to train the melting depth measurement error correction model based on long short-term memory network. The structure of long short-term memory network is as follows: Figure 5 shown.

[0073] The calculation of the long short-term memory network is as follows, the input variable of the input gate is i t , the output variable f of the forget gate t , the state of cell renewal c t , the output variable o of the output gate t They are:

[0074] i t =σ(ω i ·[x t , H t-1 ]+b i )

[0075] f t =σ(ω f ·[x t , H t-1 ]+b f )

[0076] c t =f t c t-1 +i t tanh(ω c·[x t ,h t-1 ]+b c )

[0077] o t =σ(ω0·[x t , H t-1 ]+b o )

[0078] H t =o t tanh(c t )

[0079] In the above formula, ω i ,ω f ,ω c , ω0 are the weights of the input gate, forget gate, cell state update and output gate, respectively, b i , b f , b c , b o are the weight terms and bias terms of the input gate, forget gate, cell state update and output gate, σ is the sigmoid activation function, and tanh is the double full tangent activation function.

[0080] In summary, if Figure 4 As shown, the embodiment of the present invention uses weak coherent optical tomography to coaxially couple with the laser welding system, and Fourier transforms the spectral signals to obtain the keyhole depth data through the mutually interfering spectral signals, and inputs the complex initial keyhole depth data into the percentile filter for filtering. Since the change of the penetration depth is more gentle than the change of the keyhole, the value after the percentile filter is processed by moving average to obtain the keyhole depth change curve; the sequence values ​​of the keyhole depth are used as the input neurons of the long short-term memory network, and the time series values ​​of the penetration depth obtained by the metallographic test are used as the output layer neurons of the long short-term memory network for model training. Through this training model, the measurement error caused by the deflection of the welding head and the overlap gap in the OCT penetration depth online detection system can be effectively corrected.

[0081] Reference Figure 2 , an error correction system for laser welding penetration based on OCT, comprising:

[0082] The first module 201 is used to construct an OCT penetration depth detection device to collect keyhole depth information of a workpiece during laser deep penetration welding to obtain keyhole depth information of the workpiece;

[0083] The second module 202 is used to filter and perform moving average processing on the keyhole depth information of the workpiece based on a moving average algorithm of percentile filtering to obtain a keyhole depth variation curve of the workpiece;

[0084] The third module 203 is used to perform error correction on the keyhole depth variation curve of the workpiece through a long short-term memory neural network model to obtain a corrected keyhole depth variation curve of the workpiece.

[0085] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0086] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. An error correction method for laser welding penetration based on OCT, characterized in that: The following steps are involved: An OCT penetration depth detection device is constructed to collect the keyhole depth information of the workpiece during the laser deep penetration welding process to obtain the keyhole depth information of the workpiece; Based on the moving average algorithm of percentile filtering, filtering and moving average processing are performed on the keyhole depth information of the workpiece to obtain a keyhole depth variation curve of the workpiece; The error correction of the keyhole depth variation curve of the workpiece is performed through a long short-term memory neural network model to obtain a corrected workpiece keyhole depth variation curve.

2. The error correction method for laser welding penetration based on OCT according to claim 1, characterized in that: The OCT penetration detection device includes an optical coherence tomography device and a laser welding device, wherein the optical coherence tomography device and the laser welding device are connected via coaxial coupling, wherein the optical coherence tomography device includes a broadband light source generator, a beam splitter, a measuring arm, a reference arm, a reflector and a spectrometer, wherein: The broadband light source generator is used to emit broadband light source; The beam splitter is used to split the broadband light source to obtain a first branch light source and a second branch light source; The measuring arm is used to collect and process information on the keyhole depth of the workpiece according to the first branch light source to obtain a keyhole depth spectrum signal; The reference arm is used to collect information on the upper surface of the workpiece according to the second branch light source to obtain a spectral signal of the upper surface profile of the workpiece; The reflector is used to transmit the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal to the spectrometer; The spectrometer is used to analyze the interference signal between the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal to obtain the keyhole depth information of the workpiece.

3. The error correction method for laser welding penetration based on OCT according to claim 2, characterized in that: The step of constructing an OCT penetration depth detection device to collect keyhole depth information during laser deep penetration welding of a workpiece to obtain the keyhole depth information of the workpiece specifically includes: Construct an OCT penetration detection device; Based on the OCT penetration depth detection device, the keyhole depth information of the workpiece during the laser deep penetration welding process is collected to obtain a keyhole depth spectrum signal and a workpiece upper surface profile spectrum signal; The interference signal between the keyhole depth spectrum signal and the workpiece upper surface profile spectrum signal is obtained, and a fast Fourier transform is performed to obtain the keyhole depth information of the workpiece.

4. The error correction method for laser welding penetration based on OCT according to claim 3 is characterized in that: The step of filtering and moving average processing the keyhole depth information of the workpiece by the moving average algorithm based on percentile filtering to obtain the keyhole depth variation curve of the workpiece specifically includes: Inputting the keyhole depth information of the workpiece into a percentile filter for filtering, and obtaining a filtered keyhole depth variation curve, wherein the filtered keyhole depth variation curve represents a curve composed of several percentile values; The filtered keyhole depth variation curve is subjected to low-pass filtering by a moving average algorithm to obtain the keyhole depth variation curve of the workpiece.

5. The error correction method for laser welding penetration based on OCT according to claim 4 is characterized in that: The step of inputting the keyhole depth information of the workpiece into a percentile filter for filtering to obtain a filtered keyhole depth variation curve specifically includes: Define the moving center flat window size τ and percentile p of the percentile filter; constructing a keyhole depth data set according to the keyhole depth information of the workpiece, wherein the keyhole depth data set is represented by all measurement values ​​before τ / t and after τ / t; Arranging the keyhole depth data set according to a preset sorting rule to obtain an arranged keyhole depth data set; The position of the percentile p in the arranged keyhole depth data set is determined, the percentile value is obtained and replaced, and all elements in the arranged keyhole depth data set are traversed to obtain a filtered keyhole depth change curve.

6. The error correction method for laser welding penetration based on OCT according to claim 5, characterized in that: The expression of the filtered keyhole depth variation curve is specifically as follows: D p (t)=F[D(t)],t∈L In the above formula, D p (t) represents the percentile value, F[D(t)] represents the percentile function, t represents the time, L represents the keyhole depth dataset, and p represents the percentile.

7. The error correction method for laser welding penetration based on OCT according to claim 6, characterized in that: The expression of the keyhole depth variation curve of the workpiece is specifically as follows: In the above formula, D′(t) represents the time series of the keyhole depth of the workpiece, T represents the moving window size of the moving average algorithm, △t represents the period between two depth measurements, and t represents the time.

8. An error correction system for laser welding penetration based on OCT, characterized in that: Includes the following modules: The first module is used to construct an OCT penetration depth detection device to collect keyhole depth information during laser deep penetration welding of the workpiece to obtain the keyhole depth information of the workpiece; The second module is used for filtering and moving average processing the keyhole depth information of the workpiece based on the moving average algorithm of percentile filtering to obtain the keyhole depth change curve of the workpiece; The third module is used to perform error correction on the keyhole depth variation curve of the workpiece through a long short-term memory neural network model to obtain a corrected keyhole depth variation curve of the workpiece.

Images