High-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction

Abstract

The invention provides a high-strength steel laser-arc hybrid welding monitoring method based on acoustic feature extraction. The high-strength steel laser-arc hybrid welding monitoring method comprises the steps that firstly, laser-arc hybrid welding sound signals are collected and preprocessed; step 2, constructing a sound signal feature analysis method; step 3, extracting sound signal features; step 4, feature fusion and dimension reduction; by means of the high-strength steel laser-arc hybrid welding monitoring method based on acoustic feature extraction, frequency-band-divided wavelet noise reduction can be achieved, and noise and effective signals are effectively separated; the multi-frequency-band feature extraction can be carried out for the low-frequency band and the high-frequency band, and the joint feature extraction can be carried out for the cross-frequency band; therefore, the problem of coupling of multi-source acoustic signals of hybrid welding is solved, and meanwhile quantitative evaluation of the laser-arc energy coupling state can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of laser arc hybrid welding of high-strength steel for ships, and in particular to a laser arc hybrid welding monitoring method for high-strength steel for ships based on acoustic feature extraction. Background Art

[0002] In recent years, with the adjustment of the world's economic structure and the continuous advancement of industrial technology, laser welding technology has been widely used and deeply studied in developed countries. As an efficient and high-quality welding technology, laser welding has technical advantages such as high quality, high efficiency, high stability and low deformation. It has been widely used in aerospace, automobile manufacturing, shipbuilding, electronics and medical fields. According to data from the International Ship Association in 2023, more than 90% of large ship structures in the world are welded with high-strength steel, but traditional welding processes have technical defects such as a wide heat-affected zone (about 2-3mm) and a deformation rate as high as 0.8-1.2mm / m. Laser arc hybrid welding technology can reduce heat input by 40%-60%, but the stability control of the welding process has become a key bottleneck restricting its large-scale application. Current mainstream welding quality monitoring technologies (such as visual sensing and spectral analysis) have two major shortcomings: 1. Visual sensors must maintain a relatively fixed position relative to the welding torch, making real-time image acquisition impossible for complex welds. Furthermore, visual information is easily affected by strong welding light, affecting image clarity and resulting in a signal-to-noise ratio below the industry standard of 15dB. 2. Spectral analysis response delays reach 200-500ms, making real-time control impossible. Acoustic signal acquisition, on the other hand, requires less specific direction, angle, or environmental characteristics. The International Institute of Welding (IIW) 2022 Technology Roadmap indicates that patents for acoustic emission monitoring will grow at an annual rate of 28%.

[0003] Arc sound is generated by the vibrations between the molten pool and the air during welding. Although this sound signal is non-stationary and random, it originates from arc energy fluctuations. Therefore, arc sound should provide a sensitive and accurate correlation between weld quality and the state of the weld pool during welding. From the perspective of bionic welders, most experienced welders believe that arc sound provides as rich and useful information about the welding process as visual perception. Arc sound is highly sensitive to changes in the welding process. Changes in process parameters, arc flow, and droplet transfer all affect the characteristic value of the arc sound signal. Research has shown that when the melting electrode and the base material are short-circuited, the metal of the melting electrode separates from the electrode, generating a small amount of spatter. The liquid metal impacts the weld pool. Furthermore, the shielding gas expands rapidly due to the high temperature at the moment of successful arc ignition, causing strong vibrations in the shielding gas and generating arc sound. Furthermore, during welding, changes in the shape and size of the melting electrode and the weld pool, as well as irregular arc explosions, can generate disturbance signals. Therefore, defects such as spatter, weld deviation, weld leakage, and undercut can all be identified through sound monitoring and analysis. In addition, the sound of the welding process also varies significantly under different penetration states. In summary, extracting the characteristic information of the arc sound signal during welding can achieve real-time monitoring and evaluation of the welding quality during the welding process. However, since the welding sound signal received by the microphone used in the experimental process has the characteristics of multiple sound sources and multiple influencing factors, studying how to extract effective sound signal characteristics is a key link in realizing welding quality monitoring.

[0004] Currently, the primary method for analyzing welding arc sound signals is to extract sound features using a combination of time-domain analysis and statistical methods. This involves analyzing the welding sound signal in the time domain and directly obtaining statistical quantities as features of the sound signal. This method is effective for extracting sound signals from conventional arc welding, but is less effective for analyzing the unique acoustic frequency features of hybrid welding. This is primarily due to the fact that the acoustic signal of laser-arc hybrid welding contains plasma oscillations (high frequency), molten pool flow (low frequency), and ambient noise (medium frequency), making it difficult to distinguish the contributions of different physical processes using traditional single-scale features (such as spectrum peaks). Furthermore, time-domain feature analysis methods are insufficiently capable of characterizing nonlinear dynamics (such as molten pool turbulence), resulting in high feature redundancy and severe information loss. There is a lack of dedicated noise reduction and feature extraction methods for the acoustic signals of laser-arc hybrid welding of high-strength shipbuilding steel.

[0005] Patent CN101719368B discloses a high-intensity directional sound wave transmitter. This device uses an array signal processing circuit to drive a multi-channel speaker array to achieve long-distance, directional transmission of high-intensity sound waves. The signal processing involves three steps: frequency division, band pre-weighting, and beamforming. First, frequency division decomposes the broadband digital signal into multiple sub-band signals. Then, based on a psychoacoustic model and psychoacoustic parameters, band pre-weighting is used to adjust the gain of each sub-band signal. Finally, beamforming is used to adjust the phase of the input signals from each small array, so that the output sounds from each channel are superimposed in the same direction in the far field to improve directivity. However, this device cannot achieve multi-band coordinated noise reduction for high-strength marine steel, and its performance in analyzing the unique acoustic characteristics of composite welding is relatively unsatisfactory. Summary of the Invention

[0006] In view of this, the present invention aims to propose a high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction, so as to solve the problem that the existing sound feature extraction method in the prior art lacks multi-band collaborative noise reduction and cross-scale feature fusion methods for composite welding sound signals, and does not solve the problem of feature coupling caused by the physical mechanism differences between high-frequency (laser plasma) and low-frequency (arc molten pool) sound signals; thereby achieving frequency band wavelet noise reduction and effective separation of noise and effective signals; it can also perform frequency band feature extraction for low-frequency and high-frequency bands respectively, and perform joint feature extraction for cross-bands, thereby improving the flexibility and accuracy of feature extraction; thereby solving the problem of multi-source sound signal coupling in composite welding, and at the same time, it can also achieve quantitative evaluation of the laser-arc energy coupling state.

[0007] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0008] The present invention relates to a high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction, comprising the following steps:

[0009] Step 1: Acquisition and preprocessing of laser-arc hybrid welding sound signals: The required sound signals are collected through the configuration of a multi-band acoustic sensor, and the DC component in the sound signals is removed. Next, the sound signals are subjected to frequency-domain adaptive noise reduction processing, followed by windowing processing.

[0010] Step 2: Construct a sound signal feature analysis method: use time domain analysis and frequency domain analysis to obtain relevant information of the sound signal;

[0011] Step 3: Extracting sound signal features: Extract sub-band features for the low-frequency band and high-frequency band respectively, and extract cross-band joint feature information of the signal at the same time;

[0012] Step 4: Feature fusion and dimensionality reduction: The improved kernel principal component analysis (mKPCA) is used to perform feature fusion and dimensionality reduction on the acquired features of each dimension.

[0013] Furthermore, in step 2, time domain analysis is used to obtain characteristic information of the sound signal that changes over time; and frequency domain analysis is used to obtain the spectrum characteristics of the sound signal by decomposing different frequency components of the sound signal.

[0014] Furthermore, step one includes:

[0015] Step S11: Acquisition and preprocessing of laser-arc hybrid welding sound signals. The required sound signals are collected by setting up a multi-band acoustic sensor: sound sensors of different frequencies are placed near the welding torch and behind the molten pool, and the high-frequency plasma oscillation sound and the low-frequency molten pool flow sound are synchronously collected by the sound signal collector;

[0016] Step S12: removing the DC component in the sound signal: using a method of subtracting the average value of every 20 collected data to remove the DC component of the sound signal;

[0017] Step S13: frequency-domain adaptive noise reduction processing is performed on the sound signal: for the case where the laser-arc hybrid welding sound signal contains high-frequency and low-frequency signals, a frequency-band wavelet noise reduction method is adopted, a Stein unbiased risk threshold method is used for high-frequency signals, and a hard threshold method is used for low-frequency signals;

[0018] Step S14: windowing the sound signal: use a Hamming window to perform windowing on the arc sound signal, with the window length set to 4016 sampling points and the frame shift set to 2008 sampling points.

[0019] Furthermore, step 2 includes:

[0020] Step S21: Constructing a sound signal feature analysis method: using time domain analysis calculation to obtain time domain feature information of the sound signal.

[0021] Step S22: Use frequency domain analysis to obtain frequency domain feature information of the sound signal: analyze the spectrum of the arc sound signal by simulating the human auditory characteristics to obtain the Mel-frequency cepstral coefficients (MFCCs); and extract the energy and standard deviation feature information of the sound signal in each frequency band through wavelet packet decomposition.

[0022] Furthermore, in step S21, the time domain feature information includes any one or more feature information of zero-crossing rate, sound energy, standard deviation, root mean square, average value, skewness, enthalpy value and kurtosis factor.

[0023] Furthermore, in step S22, the sound signal is decomposed into at least one scale using a wavelet packet decomposition method.

[0024] Furthermore, wavelet packet decomposition decomposes each frequency band into four segments, and the energy sum and standard deviation of each segment are calculated.

[0025] Furthermore, step three includes:

[0026] Step S31: Extracting sound signal features: extracting low-frequency sound signal features through time domain analysis and frequency domain analysis;

[0027] Step S32: extracting the sound signal features in the high frequency band by using time domain analysis and frequency domain analysis methods respectively;

[0028] Step S33: Extracting cross-band joint features: The cross-band joint features are obtained by calculating the high-frequency-low-frequency energy ratio and combining the time-frequency synchronization.

[0029] Furthermore, the frequency range of the low frequency band is between 100 Hz and 10,000 Hz; the frequency range of the high frequency band is between 10,000 Hz and 20,000 Hz.

[0030] Furthermore, in step S33, the time-frequency synchronization is obtained by analyzing the correlation between the two frequency band signals through cross wavelet transform.

[0031] Compared with the prior art, the high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction described in the present invention has the following beneficial effects:

[0032] Through the setting of the method, it is possible to achieve frequency band wavelet denoising, use Stein's unbiased risk threshold for high-frequency signals, and use the hard threshold method for low-frequency signals to effectively separate noise from valid signals; it is also possible to perform frequency band feature extraction for low-frequency and high-frequency bands respectively, and perform joint feature extraction for cross-frequency bands at the same time, thereby improving the flexibility and accuracy of feature extraction. Specifically, by introducing frequency band features driven by physical mechanisms, the problem of multi-source acoustic signal coupling in composite welding is solved, and cross-frequency band coherence analysis is added to achieve quantitative evaluation of the laser-arc energy coupling state. In addition, by focusing on the multi-scale feature extraction and fusion mechanism of acoustic signals, a highly robust feature engineering solution is provided for the quality control of laser-arc composite welding, avoiding reliance on complex monitoring systems and being more technically universal. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0034] Figure 1 This is a schematic diagram of the MFCC feature parameter extraction flow chart;

[0035] Figure 2 Schematic diagram of the detection method flow chart. DETAILED DESCRIPTION

[0036] The inventive concepts of the present disclosure will be described below using terms commonly used by those skilled in the art to convey the essence of their work to other persons skilled in the art. However, these inventive concepts can be embodied in many different forms and should not be considered limited to the embodiments described herein.

[0037] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0038] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0039] In order to solve the problem that the existing sound feature extraction methods in the prior art lack multi-band collaborative noise reduction and cross-scale feature fusion methods for hybrid welding sound signals, and do not solve the problem of feature coupling caused by the physical mechanism differences between high-frequency (laser plasma) and low-frequency (arc molten pool) sound signals; this embodiment proposes a high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction, the method comprising the following steps:

[0040] Step 1: Acquisition and preprocessing of laser-arc hybrid welding sound signals: The required sound signals are collected through the configuration of a multi-band acoustic sensor, and the DC component within the sound signals is removed. Next, the sound signals are subjected to frequency-domain adaptive noise reduction processing, followed by windowing processing.

[0041] Step 2: Build a sound signal feature analysis method: Use time domain analysis and frequency domain analysis to obtain relevant information about the sound signal. Time domain analysis is used to obtain characteristic information about the sound signal as it changes over time; frequency domain analysis decomposes the sound signal into different frequency components to obtain the spectral characteristics of the sound signal.

[0042] Step 3: Extracting sound signal features: Extract sub-band features for the low-frequency band and high-frequency band respectively, and extract cross-band joint feature information of the signal at the same time.

[0043] Step 4: Feature Fusion and Dimensionality Reduction: The improved kernel principal component analysis (mKPCA) is used to fuse and reduce the acquired features in each dimension. By introducing local-preserving projection constraints, the local structure of the feature space is preserved, addressing the local information loss caused by global projection in traditional KPCA.

[0044] Through the setting of the method, frequency band wavelet denoising can be achieved, and Stein's unbiased risk threshold is used for high-frequency signals, and the hard threshold method is used for low-frequency signals, so as to effectively separate noise and valid signals; frequency band feature extraction can also be performed for low-frequency bands and high-frequency bands respectively, and joint feature extraction can be performed for cross-frequency bands at the same time, thereby improving the flexibility and accuracy of feature extraction.

[0045] Step one includes:

[0046] Step S11: Acquisition and preprocessing of laser-arc hybrid welding sound signals. The required sound signals are collected by setting up a multi-band acoustic sensor: sound sensors of different frequencies are placed near the welding torch and behind the molten pool. The high-frequency plasma oscillation sound and the low-frequency molten pool flow sound are synchronously collected by the sound signal collector to facilitate the distinction between the effects of the laser and the arc on the sound signals.

[0047] Step S12: removing the DC component in the sound signal: using a method of subtracting the average value of every 20 collected data to remove the DC component of the sound signal;

[0048] Step S13: Perform frequency-domain adaptive noise reduction on the sound signal: When the laser-arc hybrid welding sound signal contains both high-frequency and low-frequency signals, a frequency-band wavelet noise reduction method is used. A Stein unbiased risk threshold method is used for high-frequency signals, and a hard threshold method is used for low-frequency signals to separate noise from valid signals.

[0049] Step S14: windowing the sound signal: use a Hamming window to perform windowing on the arc sound signal, with the window length set to 4016 sampling points and the frame shift set to 2008 sampling points.

[0050] Windowing refers to multiplying a signal by a specific window function to highlight the characteristics of the signal within a specific time period, while reducing discontinuities at the truncation edges and improving the accuracy of spectrum analysis. Frequency domain analysis involves converting a signal from the time domain (time-amplitude) to the frequency domain (frequency-energy) to reveal the distribution characteristics of different frequency components.

[0051] By setting step S12, the problem that due to DC bias, the collected sound signal is prone to DC components, which is not conducive to the subsequent feature extraction of the arc sound signal can be solved, and the influence of the DC component on the sound signal can be effectively eliminated. By setting step S13, the influence of external factors such as the environment that are easily affected when collecting sound signals during the welding process can be effectively avoided. Since any one or more noises such as the welding power supply, transformer and motion mechanism noise, electromagnetic noise, and shielding airflow noise may interfere with or even drown out the arc sound signal, by performing frequency domain adaptive noise reduction processing on the sound signal in step S13, it is beneficial to denoise the noise of the welding sound signal before feature extraction, so as to reduce the difficulty of subsequent signal analysis. In addition, due to the interaction between the laser, arc and molten pool oscillations, the sound signal of laser arc hybrid welding is a non-stationary time-varying signal. By setting step S14, a smooth transition between frames can be achieved, which is convenient for subsequent frequency domain analysis such as Fourier transform.

[0052] Step 2 includes:

[0053] Step S21: Constructing a sound signal feature analysis method: using time domain analysis to obtain time domain feature information of the sound signal. The time domain feature information includes any one or more feature information of zero crossing rate, sound energy, standard deviation, root mean square, mean value, skewness, enthalpy value and kurtosis factor;

[0054] Step S22: Use frequency domain analysis to obtain frequency domain feature information of the sound signal: by simulating the human auditory characteristics to analyze the spectrum of the arc sound signal, so as to obtain the sound frequency domain feature parameters that better reflect the welding quality, namely the Mel-frequency cepstral coefficients MFCC; and through wavelet packet decomposition, extract the energy and standard deviation feature information of the sound signal in each frequency band.

[0055] In step S21, the calculation formula for the zero-crossing rate is: Z(n) refers to the zero-crossing rate of the nth frame; x(n) refers to the signal variable; s(n-1) refers to the sound signal of the nth frame; w(n) refers to the window function for weighted signal processing.

[0056] The formula for calculating sound energy is: En refers to the energy value of the nth frame; x(m) refers to the sound of the mth frame; N-1 refers to the frame length.

[0057] The formula for calculating the standard deviation is: St refers to the standard deviation of the sound signal; n refers to the total number of all possible symbols in the sound source; xi refers to the value of the i-th signal point; Me refers to the average value of the n signal points.

[0058] The formula for calculating the root mean square is: RMS refers to the root mean square of the sound signal.

[0059] The formula for calculating the average value is: me is the same as Me. Both me and Me refer to the average value of n signal points.

[0060] The calculation formula for skewness is: Sk refers to the skewness value of the sound signal; Refers to the average value of n sampled sound signals.

[0061] The calculation formula for enthalpy is: H(X) refers to the entropy of the sound source, that is, the statistical average of the amount of information contained in each symbol in the sound signal source; P(xi) refers to the probability of the i-th sound symbol appearing in the sound source, where 0≤P(xi)≤1.

[0062] The calculation formula of the kurtosis factor is: C refers to the kurtosis factor of the sound signal, which is used to measure the blockage of the sound signal, that is, the sharpness of the signal distribution; xmax refers to the maximum value of the sound signal; xmin refers to the minimum value of the sound signal.

[0063] In step S22, unlike traditional Fourier transform methods, which cannot simultaneously depict time-frequency components or perform comprehensive analysis in the time-frequency domain, the present application uses wavelet packet decomposition to decompose the sound signal at at least one scale. This decomposes the sound signal into wavelet packet coefficients of different frequency bands, quantifies the plasma oscillation energy distribution, and thus extracts energy sum and standard deviation features for the sound signal in each frequency band. Preferably, in this embodiment, wavelet packet decomposition typically decomposes each frequency band into four segments, and calculates the energy sum and standard deviation of each segment.

[0064] The method for extracting Mel-frequency cepstral coefficients (MFCCs) in step S22 includes the following steps:

[0065] Step S221: Pre-processing: pre-emphasize the sound signal to be extracted to enhance the high-frequency part of the sound signal and compensate for the high-frequency loss;

[0066] Step S222: Fourier transform: convert the time domain signal into a frequency domain signal through fast Fourier transform to obtain a spectrum;

[0067] Step S223: Calculation of spectral line energy: square the amplitude of each frequency component of the frequency domain signal to obtain an energy spectrum, which provides energy distribution data for subsequent Mel filtering;

[0068] Step S224: Mel filtering energy: using a triangular filter to map the linear spectrum to a Mel nonlinear spectrum based on human auditory perception;

[0069] Step S225: DCT cepstrum: first take the logarithm of the Mel filter energy, and then perform discrete cosine transform (DCT) to obtain MFCC feature parameters;

[0070] Through the settings in step 2, the frequency band characteristics driven by physical mechanisms are introduced to solve the problem of multi-source acoustic signal coupling in composite welding. At the same time, cross-band coherence analysis is added to achieve quantitative evaluation of the laser-arc energy coupling state.

[0071] Step three includes:

[0072] Step S31: Extracting sound signal features: extracting sound signal features in the low frequency band by using time domain analysis and frequency domain analysis methods respectively, wherein the frequency range of the low frequency band is between 100 Hz and 10,000 Hz.

[0073] Step S32: extracting the sound signal features of the high frequency band by using time domain analysis and frequency domain analysis methods respectively, wherein the frequency value range of the high frequency band is between 10000 Hz and 20000 Hz.

[0074] Step S33: Extracting cross-band joint features: The cross-band joint features are obtained by calculating the high-frequency-low-frequency energy ratio and combining the time-frequency synchronization.

[0075] In step S33, time-frequency synchronization is obtained by analyzing the correlation between the two frequency band signals through cross-wavelet transform. Furthermore, the high-frequency-to-low-frequency energy ratio can reflect the energy coupling state between the laser and arc. Furthermore, the characteristic information of the sound signal obtained in step 3 comprises 36 dimensions. For both the low-frequency and high-frequency bands, eight dimensions of feature information, namely zero-crossing rate, sound energy, standard deviation, root mean square (RMS), mean value, skewness, enthalpy, and kurtosis factor, are extracted through time-domain analysis. For both the low-frequency and high-frequency bands, one dimension of Mel-frequency cepstral coefficients and eight dimensions of energy sums and standard deviations of the four different scales of the same frequency band, obtained through wavelet packet decomposition, are extracted through frequency-domain analysis. A cross-band joint feature analysis method is used to obtain two dimensions of feature information, namely the high-frequency-to-low-frequency energy ratio and video synchronization. Details are shown in Table 1 below.

[0076] Table 1 Dimensions of sound signal feature extraction

[0077]

[0078] Through the setting of step three, on the one hand, it is possible to distinguish the acoustic characterization of the laser and arc physical processes through frequency band feature extraction, solving the feature confusion problem of traditional methods; on the other hand, it is also possible to integrate time domain, frequency domain and cross-band features to improve the sensitivity of welding status.

[0079] In step 4, we use an improved kernel principal component analysis (mKPCA) for feature fusion and dimensionality reduction on the 36-dimensional features. We introduce local-preserving projection constraints to preserve the local structure of the feature space and address the local information loss caused by global projections in traditional KPCA. The final output fused feature dimension is ≤ 5, and ≥ 90% of the original information is retained.

[0080] By focusing on the multi-scale feature extraction and fusion mechanism of acoustic signals, a highly robust feature engineering solution is provided for the quality control of laser-arc hybrid welding, avoiding dependence on complex monitoring systems and making the technology more universal.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction, characterized in that: The following steps are involved: Step 1: Acquisition and preprocessing of laser-arc hybrid welding sound signals: The required sound signals are collected through the configuration of a multi-band acoustic sensor, and the DC component in the sound signals is removed. Next, the sound signals are subjected to frequency-domain adaptive noise reduction processing, followed by windowing processing. Step 2: Construct a sound signal feature analysis method: use time domain analysis and frequency domain analysis to obtain relevant information of the sound signal; Step 3: Extracting sound signal features: Extract sub-band features for the low-frequency band and high-frequency band respectively, and extract cross-band joint feature information of the signal at the same time; Step 4: Feature fusion and dimensionality reduction: The improved kernel principal component analysis (mKPCA) is used to perform feature fusion and dimensionality reduction on the acquired features of each dimension.

2. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 1 is characterized in that: In the step 2, the time domain analysis obtains the characteristic information of the sound signal changing with time; the frequency domain analysis obtains the spectrum characteristics of the sound signal by decomposing the different frequency components of the sound signal.

3. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 1 is characterized in that: The step one comprises: Step S11: Acquisition and preprocessing of laser-arc hybrid welding sound signals. The required sound signals are collected by setting up a multi-band acoustic sensor: sound sensors of different frequencies are placed near the welding torch and behind the molten pool, and the high-frequency plasma oscillation sound and the low-frequency molten pool flow sound are synchronously collected by the sound signal collector; Step S12: removing the DC component in the sound signal: using a method of subtracting the average value of every 20 collected data to remove the DC component of the sound signal; Step S13: frequency-domain adaptive noise reduction processing is performed on the sound signal: for the case where the laser-arc hybrid welding sound signal contains high-frequency and low-frequency signals, a frequency-band wavelet noise reduction method is adopted, a Stein unbiased risk threshold method is used for high-frequency signals, and a hard threshold method is used for low-frequency signals; Step S14: windowing the sound signal: use a Hamming window to perform windowing on the arc sound signal, with the window length set to 4016 sampling points and the frame shift set to 2008 sampling points.

4. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 1 is characterized in that: The second step includes: Step S21: constructing a sound signal feature analysis method: using time domain analysis calculation to obtain time domain feature information of the sound signal; Step S22: Use frequency domain analysis to obtain frequency domain feature information of the sound signal: analyze the spectrum of the arc sound signal by simulating the human auditory characteristics to obtain the Mel-frequency cepstral coefficients (MFCCs); and extract the energy and standard deviation feature information of the sound signal in each frequency band through wavelet packet decomposition.

5. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 4 is characterized in that: In step S21, the time domain feature information includes any one or more feature information of zero-crossing rate, sound energy, standard deviation, root mean square, average value, skewness, enthalpy value and kurtosis factor.

6. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 4 is characterized in that: In step S22, the sound signal is decomposed into at least one scale using a wavelet packet decomposition method.

7. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 6 is characterized in that: The wavelet packet decomposition decomposes each frequency band into four segments, and calculates the energy sum and standard deviation of each segment.

8. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 1 is characterized in that: The step three includes: Step S31: Extracting sound signal features: extracting low-frequency sound signal features through time domain analysis and frequency domain analysis; Step S32: extracting the sound signal features in the high frequency band by using time domain analysis and frequency domain analysis methods respectively; Step S33: Extracting cross-band joint features: The cross-band joint features are obtained by calculating the high-frequency-low-frequency energy ratio and combining the time-frequency synchronization.

9. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 8 is characterized in that: The frequency range of the low frequency band is between 100 Hz and 10,000 Hz; the frequency range of the high frequency band is between 10,000 Hz and 20,000 Hz.

10. The high-strength steel laser arc hybrid welding monitoring method based on acoustic feature extraction according to claim 8, characterized in that: In step S33, the time-frequency synchronization is obtained by analyzing the correlation between the two frequency band signals through cross wavelet transform.

Images