Real-time laser welding quality evaluation method based on data analysis

Abstract

The invention relates to the technical field of laser welding quality monitoring, in particular to a laser welding quality real-time evaluation method based on data analysis. The method comprises the steps that in the laser welding process, reflected light signals of a welding keyhole area and plasma radiation signals above a molten pool are synchronously collected at the same sampling frequency; performing detrending processing on the two paths of signals respectively, and extracting high-frequency fluctuation components; calculating a cross-correlation function of the two paths of high-frequency fluctuation components based on the sliding time window, and generating a coupling stability index; a dynamic judgment threshold value is generated based on the energy fluctuation variance of the plasma radiation signal, and the larger the energy fluctuation variance is, the lower the threshold value is; and comparing the coupling stability index with a dynamic judgment threshold to judge the keyhole instability defect. Through dynamic coupling analysis of the keyhole reflected light and the plasma radiation signal, real-time evaluation of the welding quality is achieved, and the welding defect can be accurately recognized without complex sensor fusion.

Description

Technical Field

[0001] This application relates to the field of laser welding quality monitoring technology. More specifically, this application relates to a real-time laser welding quality assessment method based on data analysis. Background Technology

[0002] Laser welding, as a high-precision and high-efficiency advanced joining technology, has been widely used in automotive manufacturing, aerospace, electronics, and new energy batteries. With the development of intelligent manufacturing and industrial digitalization, higher demands are being placed on the real-time monitoring and evaluation of laser welding quality. However, existing laser welding quality monitoring technologies still face many challenges in practical applications.

[0003] Existing technical solutions mainly employ multi-sensor fusion schemes or deep learning methods. For example, multimodal fusion schemes collect data from various sensors during laser welding, including visual signals, acoustic signals, thermal imaging data, and spectral information, and then use deep learning methods to fuse this multimodal data, enabling online monitoring of welding quality. Methods based on molten pool image analysis acquire images of the weld pool using high-speed cameras and utilize convolutional neural networks to analyze the molten pool morphology and predict welding quality. Methods based on acoustic signals collect acoustic emission signals generated during the welding process and extract temporal and frequency domain features for quality classification.

[0004] In summary, existing laser welding quality monitoring technologies suffer from the following problems: multi-sensor solutions are structurally complex and costly, and synchronous acquisition and alignment of multi-source data present technical challenges; deep learning methods require a large amount of labeled training data, resulting in long model training cycles and high computational complexity; single signal features are insufficient to comprehensively reflect the welding quality status, the strong light radiation in the molten pool region places extremely high demands on image acquisition equipment, and acoustic signals are easily affected by environmental noise; existing methods have limited adaptability to different welding process parameters, and the models lack generalization ability. Therefore, there is an urgent need to develop a real-time laser welding quality assessment method that can achieve low-cost deployment, efficient real-time processing, and good process adaptability while ensuring evaluation accuracy. Summary of the Invention

[0005] The purpose of this application is to propose a real-time evaluation method for laser welding quality based on data analysis, in order to solve the problems of complex and expensive multi-sensor solutions, poor real-time performance and insufficient generalization ability in the prior art.

[0006] The data analysis-based real-time evaluation method for laser welding quality provided in this application includes: simultaneously acquiring reflected light signals from the welding keyhole region and plasma radiation signals from above the molten pool during laser welding; performing detrending processing on the reflected light signals and the plasma radiation signals respectively to extract high-frequency fluctuation components characterizing the transient welding process; calculating the cross-correlation function between the high-frequency fluctuation components of the reflected light signals and the plasma radiation signals based on a sliding time window to generate a coupling stability index; generating a dynamic decision threshold based on the energy fluctuation variance of the plasma radiation signals within the current time window, wherein the dynamic decision threshold decreases as the energy fluctuation variance increases; and comparing the coupling stability index with the dynamic decision threshold to determine whether a keyhole instability defect has occurred in the current welding process.

[0007] This application generates a coupling stability index by simultaneously acquiring reflected light signals and plasma radiation signals and calculating the cross-correlation function. Combined with a dynamic threshold mechanism based on energy fluctuation variance, it can accurately identify keyhole instability defects. This method utilizes the physical causal relationship between keyhole opening and closing and plasma eruption, employing signal correlation analysis instead of traditional amplitude threshold comparison. This effectively solves the problem of misjudgment caused by sensor aging or lens contamination, achieving low-cost deployment and efficient real-time processing while ensuring evaluation accuracy.

[0008] Optionally, the acquisition of the reflected light signal includes: acquiring the reflected light signal characterizing the intensity of reflected light in the welding keyhole area through a coaxial optical path; the acquisition of the plasma radiation signal includes: acquiring the plasma radiation signal characterizing the intensity of plasma cloud radiation above the molten pool through a paraxial optical path.

[0009] This application adopts a dual-path acquisition architecture with coaxial and off-axis optical paths. The coaxial optical path directly acquires the keyhole opening status information, while the off-axis optical path independently acquires the plasma radiation intensity. The two signals are independent of each other but physically related, providing a reliable data foundation for subsequent coupling analysis.

[0010] Optionally, the detrending process includes: applying a high-pass digital filter to the reflected light signal and the plasma radiation signal respectively, or using a moving average subtraction method to remove the DC component and low-frequency trend term of the signal.

[0011] Optionally, the calculation of the coupling stability index includes: within the sliding time window, calculating the cross-covariance of the high-frequency fluctuation component of the reflected light signal and the high-frequency fluctuation component of the plasma radiation signal; calculating the variance of the high-frequency fluctuation component of the reflected light signal and the variance of the high-frequency fluctuation component of the plasma radiation signal, and using the square root of the product of the two variances as a normalization factor; and using the normalization factor to normalize the cross-covariance to obtain a normalized cross-correlation function.

[0012] This application calculates the cross-covariance and performs normalization to obtain a normalized cross-correlation function, eliminating the influence of changes in the absolute amplitude of the signal on the correlation assessment. When sensor aging or lens contamination leads to a decrease in overall light intensity, the waveform similarity, i.e., the correlation, of the two signals remains stable, thereby effectively improving the robustness and industrial applicability of the monitoring method.

[0013] Optionally, the coupling stability index further includes: determining a search interval and setting a sliding window for multiple lag times of the plasma radiation signal relative to the reflected light signal, the search interval covering the estimated maximum physical response delay between the reflected light signal and the plasma radiation signal; calculating the normalized cross-correlation function corresponding to different lag times within the search interval to form a cross-correlation function sequence; searching for the maximum value in the cross-correlation function sequence and determining the maximum value as the coupling stability index, while defining the lag time corresponding to the maximum value as the instantaneous physical delay.

[0014] Optionally, the energy fluctuation variance is inversely proportional to the dynamic decision threshold.

[0015] This application establishes a negative correlation between energy fluctuation variance and dynamic decision threshold, enabling adaptive adjustment of monitoring sensitivity based on welding conditions. In high-power welding mode, where background noise is significant, lowering the threshold can suppress false alarms; in low-power stable mode, increasing the threshold enhances defect detection capabilities, achieving an automatic balance between monitoring sensitivity and false alarm rate.

[0016] Optionally, the calculation of the energy fluctuation variance includes: within the current time window, calculating the statistical variance of the instantaneous power sequence of the plasma radiation signal, and determining the statistical variance as the energy fluctuation variance.

[0017] Optionally, determining the occurrence of keyhole instability defects includes: comparing the coupling stability index with the dynamic decision threshold in real time; if the coupling stability index is less than the dynamic decision threshold, starting a timer or counter; when the duration of the state less than the dynamic decision threshold exceeds a preset time window threshold, outputting a defect determination signal; if the duration does not exceed the time window threshold, determining it as an instantaneous disturbance.

[0018] This application sets a duration determination mechanism, which only outputs a defect determination signal when the coupling stability index is below the threshold for a continuous period of time exceeding the preset time window threshold. This effectively distinguishes between real keyhole instability defects and brief transient disturbances, avoiding frequent false alarms caused by random noise or brief fluctuations.

[0019] Optionally, the keyhole instability defect includes at least one of the following: keyhole collapse defect, porosity defect, lack of fusion defect, or hump weld defect.

[0020] Optionally, the evaluation method further includes: calculating the integral area of ​​the coupling stability index below the dynamic decision threshold and defining it as the instability energy accumulation value; and classifying the current keyhole instability defect into three warning levels: mild, moderate, and severe, based on the instability energy accumulation value and the duration of the instability energy being below the threshold.

[0021] This application divides defects into three warning levels by calculating the cumulative value of instability energy and combining it with the duration. Furthermore, it generates a full weld quality topology map by combining the spatial coordinates of the weld head, realizing the quality mapping from the time domain to the spatial domain, and providing a quantitative basis for post-weld quality traceability and process optimization.

[0022] The beneficial effects of this application are as follows: By constructing a cross-correlation analysis model between reflected light signals and plasma radiation signals, this application elevates the technical characteristics from traditional scalar threshold comparisons to vector relationship analysis. This method is based on the physical law that the keyhole opening size directly regulates the plasma ejection volume during laser deep penetration welding, and characterizes the dynamic interaction between the keyhole and the plasma cloud by calculating the coupling stability index. Compared with existing technologies, this application has the following advantages: accurate quality assessment can be achieved without multi-sensor fusion, significantly reducing equipment costs and deployment complexity; signal correlation analysis is used to combat sensor aging and environmental interference, improving industrial applicability; the dynamic threshold mechanism enables adaptive monitoring under different welding conditions; and the hierarchical early warning and spatial mapping functions support post-weld quality traceability, meeting the urgent need for online monitoring of welding quality in intelligent manufacturing. Attached Figure Description

[0023] Figure 1 This is a flowchart of a data analysis-based real-time evaluation method for laser welding quality according to an embodiment of this application.

[0024] Figure 2 This is a graph showing the changes in the collected parameters of the real-time laser welding quality evaluation method based on data analysis according to an embodiment of this application.

[0025] Figure 3 This is a graph showing the change in the coupling stability index of the real-time laser welding quality evaluation method based on data analysis according to an embodiment of this application. Detailed Implementation

[0026] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Figure 1 The image shows a data analysis-based real-time evaluation method for laser welding quality according to an embodiment of this application.

[0027] S1: Synchronous dual-channel acquisition of welding process signals.

[0028] During laser welding, reflected light signals from the welding keyhole area and plasma radiation signals from above the molten pool are simultaneously acquired at the same sampling frequency. In this embodiment, the sampling frequency is set to 100 kHz to meet the requirement of capturing millisecond-level dynamic changes in the laser welding process.

[0029] The reflected light signal is acquired via a coaxial optical path. This coaxial optical path shares the same optical path as the laser beam, and a beam splitter separates the laser beam from the returning reflected light. A photodetector receives the coaxial light signal reflected by the beam splitter; this signal primarily reflects the change in the intensity of laser reflection from the keyhole opening area. When the keyhole is stably open, the laser energy is absorbed deep within the keyhole, resulting in low reflected light intensity; when the keyhole undergoes periodic opening and closing oscillations or collapses, the reflected light intensity fluctuates drastically. This embodiment uses a silicon-based photodiode as the detection element, with a response wavelength range covering the laser's operating wavelength.

[0030] The plasma radiation signal is acquired via a side-axis optical path. This path is positioned to one side of the laser beam's optical axis, forming a predetermined angle with it. In this embodiment, the angle is set to 45 degrees to obtain the optimal observation angle of the plasma cloud above the molten pool. The plasma cloud is formed by the ionization of metal vapor generated by the laser's action on the metal surface, and its radiation intensity is closely related to the laser power density and the material evaporation rate. A bandpass filter is used to select characteristic bands in the plasma emission spectrum, filtering out interference from reflected laser light and ensuring the purity of the acquired signal. This embodiment uses a bandpass filter with a center wavelength of 500 nanometers and a bandwidth of 20 nanometers.

[0031] The dual-path acquisition architecture, consisting of coaxial and off-axis optical paths, has significant physical implications. The coaxial optical path directly reflects the keyhole's opening state and laser energy coupling efficiency, while the off-axis optical path independently acquires plasma radiation intensity, reflecting the ejection status of metal vapor. Although the two signals are independent, they are physically related; the opening and closing motion of the keyhole directly affects the plasma ejection volume and radiation intensity. This causal relationship lays the physical foundation for subsequent coupling analysis.

[0032] To ensure the synchronization of the two signals, a unified data acquisition card is used to simultaneously trigger sampling of both channels. The inter-channel delay of the acquisition card is less than one microsecond, far smaller than the characteristic time scale of the laser welding process. The acquired raw signals are stored in a circular buffer after analog-to-digital conversion for subsequent real-time processing. Figure 2 This is a graph showing the changes in acquisition parameters of the real-time laser welding quality assessment method based on data analysis according to an embodiment of this application. The graph illustrates the changing trends of the two synchronously acquired signals throughout the welding process. The reflected light signal and plasma radiation signal represent the two signals, respectively. The reflected light signal reflects the dynamic changes in the keyhole opening state, while the plasma radiation signal reflects the intensity of the metal vapor ejection. During normal welding, the two signals maintain a high correlation, and their waveforms change synchronously. The shaded areas mark the keyhole instability defect regions detected by the system. Within these areas, the correlation between the two signals decreases significantly, indicating that the physical coupling between the keyhole and the plasma is disrupted.

[0033] S2: Perform detrending processing to extract high-frequency fluctuation components.

[0034] The reflected light signal and the plasma radiation signal are subjected to detrending processing to filter out the low-frequency baseline drift caused by heat accumulation and extract the high-frequency fluctuation component characterizing the welding transient process.

[0035] During laser welding, as welding continues, the workpiece temperature gradually increases, causing a slow, trending change in the material surface reflectivity and plasma radiation intensity. This low-frequency baseline drift is not related to welding quality and can interfere with the identification of transient fluctuation characteristics. Therefore, detrending processing is needed to separate the useful high-frequency fluctuation components from the original signal.

[0036] The de-trending process can be implemented using a high-pass digital filter. This embodiment uses a Butterworth high-pass filter with a cutoff frequency set to 50 Hz and a filter order of fourth order. The selection of this cutoff frequency is based on the following considerations: the keyhole oscillation frequency during laser welding is typically in the range of hundreds to thousands of Hz, while the baseline drift frequency caused by heat accumulation is usually below 10 Hz. Setting the cutoff frequency to 50 Hz effectively filters out baseline drift while retaining high-frequency information related to keyhole oscillation.

[0037] The detrending process can also be implemented using moving average subtraction. The specific steps are as follows: First, calculate the moving average of the original signal over a certain time window before and after the current moment. This moving average represents the low-frequency trend of the signal. Then, subtract the moving average from the original signal to obtain the high-frequency fluctuation component after removing the DC component and the low-frequency trend. In this embodiment, the moving average window length is set to twenty milliseconds, corresponding to two thousand sampling points. The advantage of moving average subtraction is its simplicity and lack of phase delay, making it suitable for real-time processing scenarios.

[0038] After detrending processing, the high-frequency fluctuation components of the reflected light signal and the plasma radiation signal are obtained. These two high-frequency fluctuation components eliminate absolute amplitude information and retain only the transient variation characteristics of the signal, providing standardized input data for subsequent cross-correlation analysis.

[0039] S3: Calculate the cross-correlation function to generate the coupling stability index.

[0040] Based on a sliding time window, the cross-correlation function between the high-frequency fluctuation component of the reflected light signal and the high-frequency fluctuation component of the plasma radiation signal is calculated to generate a coupling stability index.

[0041] The coupling stability index is used to quantify the dynamic interaction between the keyhole and the plasma cloud. Under normal and stable welding conditions, the opening and closing motion of the keyhole and the plasma eruption exhibit a good phase-locked relationship. The waveforms of the two signals are similar and have a fixed time delay. At this time, the cross-correlation function is high, and the coupling stability index outputs a high value. When keyhole instability occurs, the keyhole motion becomes irregular, the phase relationship with the plasma eruption is disrupted, the correlation between the two signals decreases, and the coupling stability index outputs a low value.

[0042] The coupling stability index is calculated using a normalized cross-correlation function. The specific steps are as follows: First, within the sliding time window, the cross-covariance of the high-frequency fluctuation components of the reflected light signal and the plasma radiation signal is calculated. The cross-covariance is defined as the integral average of the pointwise product of the two signals within the time window. Then, the variances of the high-frequency fluctuation components of the reflected light signal and the plasma radiation signal are calculated. The variances reflect the energy level of the signal fluctuations. Next, the square root of the product of the two variances is calculated as a normalization factor. The normalization factor is used to eliminate the influence of the absolute amplitude of the signal on the correlation calculation. Finally, the cross-covariance is normalized using the normalization factor to obtain the normalized cross-correlation function. The normalized cross-correlation function ranges from -1 to +1, where +1 indicates perfect positive correlation, zero indicates no correlation, and -1 indicates perfect negative correlation.

[0043] Specifically, the formula for calculating the normalized cross-correlation function is as follows: ; In the formula, CSI(τ) represents the coupling stability index function with respect to the lag time τ. Its calculation result is a dimensionless value with a range of [-1,1], which is used to measure the similarity and synchronization of the waveforms of two signals. It represents the high-frequency fluctuation component of the reflected light signal after detrending processing, with the dimension of voltage (V), and characterizes the transient oscillation characteristics inside the keyhole; It represents the high-frequency fluctuation component of the plasma radiation signal after detrending and time shifting, with the dimension of voltage (V), and characterizes the intensity of plasma eruption above the molten pool.

[0044] Regarding the time and window parameters, t represents the current start time of the sliding time window in milliseconds; w represents the length of the sliding window in milliseconds, which determines the temporal resolution of local feature extraction; x is the integration variable, representing the scanning time point within the window; τ represents the lag time of the plasma signal relative to the reflected light signal in milliseconds, which is used to compensate for the physical delay in the transmission of fluctuations from inside the keyhole to changes in the external plasma.

[0045] The numerator of the formula is the cross-covariance integral of the reflected light signal and the plasma signal within the time window. The magnitude of this term reflects the degree of synchronization between the two physical processes in terms of their fluctuation trends. The denominator is the square root of the product of the autocorrelation energy integrals of the two signals, serving as a normalization factor. By using division in the denominator, the difference in absolute signal amplitude caused by variations in welding power or sensor gain can be eliminated, thus ensuring that the final coupling stability index reflects only the essential characteristics of the welding process stability and is not affected by the strength of the signal energy.

[0046] Furthermore, to obtain the optimal coupling stability index, it is necessary to search for the maximum value of the cross-correlation function within a lag time range that can cover the physical delay. Within the sliding time window, the search interval is determined and a sliding window with multiple lag times between the plasma radiation signal and the reflected light signal is set.

[0047] In this embodiment, the search interval is set as a preset time span centered on the zero-lag moment, specifically ranging from -2 milliseconds to +2 milliseconds, with a step size of 10 microseconds. This search interval covers the estimated maximum physical response delay between the reflected light signal and the plasma radiation signal, and also includes minor lead or lag jitter that may occur due to signal transmission or processing.

[0048] The normalized cross-correlation function corresponding to different lag times within the search interval is calculated to form a cross-correlation function sequence. The maximum value is searched within this sequence and determined as the coupling stability index. The lag time corresponding to this maximum value is defined as the instantaneous physical delay. The instantaneous physical delay reflects the time lag between keyhole opening / closing and plasma eruption in the current welding state, and its stability is an important indicator of welding quality.

[0049] In this embodiment, the sliding time window length is set to ten milliseconds, and the window step size is one millisecond, so as to realize the real-time calculation of the coupling stability index by updating it once every millisecond.

[0050] S4: Generate dynamic decision thresholds and perform defect determination.

[0051] Based on the energy fluctuation variance of the plasma radiation signal within the current time window, a dynamic decision threshold is generated, wherein the dynamic decision threshold decreases as the energy fluctuation variance increases, in order to suppress background noise interference under laser welding.

[0052] The calculation of the energy fluctuation variance is performed within the current time window. First, the instantaneous power sequence of the plasma radiation signal is calculated, where instantaneous power is defined as the square of the signal amplitude. Then, the statistical variance of the instantaneous power sequence within the time window is calculated, and this statistical variance is determined as the energy fluctuation variance. The energy fluctuation variance reflects the severity of fluctuations in plasma radiation intensity and is closely related to the laser power density and welding mode. In high-power deep penetration welding mode, the plasma radiation intensity is high and fluctuates violently, resulting in a large energy fluctuation variance; in low-power thermal conductivity welding mode, the plasma radiation is relatively stable, resulting in a smaller energy fluctuation variance.

[0053] The dynamic decision threshold is generated based on the negative correlation between energy fluctuation variance and the threshold. Specifically, an inverse proportional function is used to map the energy fluctuation variance to the dynamic decision threshold. The larger the energy fluctuation variance, the lower the generated dynamic decision threshold, thereby reducing monitoring sensitivity and suppressing false alarms caused by background noise in high-power mode. The smaller the energy fluctuation variance, the higher the generated dynamic decision threshold, thereby improving monitoring sensitivity and enhancing defect detection capabilities in low-power stable mode. In this embodiment, the dynamic decision threshold is calculated using an inverse proportional function, with the upper limit set at 0.9 and the lower limit at 0.5.

[0054] The coupling stability index, calculated in real time, is compared with a dynamic decision threshold to determine whether a keyhole instability defect has occurred during the current welding process. The determination of keyhole instability defect occurrence includes a duration confirmation mechanism. The coupling stability index and the dynamic decision threshold are compared in real time. If the coupling stability index is less than the dynamic decision threshold, a timer or counter is started. A defect determination signal is output only when the duration of the state less than the dynamic decision threshold exceeds a preset time window threshold. If the duration does not exceed the time window threshold, it is determined as a transient disturbance and no alarm is triggered. In this embodiment, the time window threshold is set to five milliseconds, a value determined based on a typical timescale of the laser welding process.

[0055] Figure 3 This is a graph showing the change in the coupling stability index of the real-time laser welding quality assessment method based on data analysis according to embodiments of this application. It illustrates the changes in both the coupling stability index and the dynamic decision threshold. The coupling stability index is obtained by calculating the normalized cross-correlation function of the high-frequency components of the reflected light and plasma signal; a higher value indicates a stronger correlation between the two signals and a more stable welding state. The dynamic decision threshold adaptively adjusts with the energy fluctuation variance, stabilizing at around 0.5. During most of the normal welding time, the coupling stability index is significantly higher than the dynamic decision threshold, indicating a stable welding process. Only in a few time windows does the coupling stability index drop sharply below the dynamic decision threshold.

[0056] The keyhole instability defects include at least one of the following: keyhole collapse defects, porosity defects, lack of fusion defects, or hump weld defects. Keyhole collapse refers to the sudden closure of the keyhole, resulting in ineffective laser energy coupling, often leading to porosity and lack of fusion defects. Porosity defects are formed by gas entrainment and solidification in the molten pool due to keyhole instability. Lack of fusion defects are caused by insufficient energy input, resulting in incomplete fusion of the base material or adjacent weld beads. Hump weld defects manifest as periodic bulges on the weld surface and are related to molten pool flow instability. A common characteristic of these defects is the disruption of the coupling relationship between the keyhole and the plasma, which can be effectively identified by anomalies in the coupling stability index.

[0057] Furthermore, the evaluation method also includes a defect grading function. The integral area where the coupling stability index is below the dynamic decision threshold is calculated and defined as the instability energy accumulation value. The integral area reflects the severity of the defect; a larger area indicates more severe instability. Based on the instability energy accumulation value and the duration of the instability below the threshold, the current keyhole instability defect is classified into three warning levels: mild, moderate, and severe. A mild warning corresponds to short-term, small-amplitude coupling instability, which may not form an actual defect but requires attention; a moderate warning corresponds to more obvious keyhole instability, which may produce defects such as internal porosity; and a severe warning corresponds to severe keyhole collapse, which is highly likely to form obvious welding defects.

[0058] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this application, based on the technical solution and inventive concept of this application, should be within the scope of protection of this application.

Claims

1. A real-time evaluation method for laser welding quality based on data analysis, characterized in that, The evaluation method includes: During laser welding, reflected light signals from the welding keyhole area and plasma radiation signals from above the molten pool are collected simultaneously. Detrending processing is performed on the reflected light signal and the plasma radiation signal respectively to extract the high-frequency fluctuation component characterizing the welding transient process; Based on a sliding time window, the cross-correlation function between the high-frequency fluctuation component of the reflected light signal and the high-frequency fluctuation component of the plasma radiation signal is calculated to generate a coupling stability index. Based on the energy fluctuation variance of the plasma radiation signal within the current time window, a dynamic decision threshold is generated, wherein the dynamic decision threshold decreases as the energy fluctuation variance increases; The coupling stability index is compared with the dynamic decision threshold to determine whether a keyhole instability defect has occurred during the current welding process.

2. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The acquisition of the reflected light signal includes: The reflected light signal, which characterizes the intensity of reflected light in the welding keyhole area, is acquired through a coaxial optical path. The acquisition of the plasma radiation signal includes: acquiring plasma radiation signals that characterize the intensity of the plasma cloud radiation light above the molten pool through a paraxial optical path.

3. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The detrending process includes: High-pass digital filters are applied to the reflected light signal and the plasma radiation signal respectively, or moving average subtraction is used to remove the DC component and low-frequency trend term of the signal.

4. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The calculation of the coupling stability index includes: Within the sliding time window, the cross-covariance between the high-frequency fluctuation component of the reflected light signal and the high-frequency fluctuation component of the plasma radiation signal is calculated. Calculate the variance of the high-frequency fluctuation component of the reflected light signal and the variance of the high-frequency fluctuation component of the plasma radiation signal, and use the square root of the product of the variances of the high-frequency fluctuation components of the reflected light signal and the plasma radiation signal as a normalization factor. The cross-covariance is normalized using the normalization factor to obtain the normalized cross-correlation function.

5. The real-time laser welding quality evaluation method based on data analysis according to claim 4, characterized in that, The coupling stability index also includes: A search interval is determined and a sliding window is set for multiple lag times of the plasma radiation signal relative to the reflected light signal. The search interval covers the estimated maximum physical response delay between the reflected light signal and the plasma radiation signal. Calculate the normalized cross-correlation function corresponding to different lag times within the search interval to form a cross-correlation function sequence; The maximum value is searched in the cross-correlation function sequence and determined as the coupling stability index. The lag time corresponding to the maximum value is defined as the instantaneous physical delay.

6. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The energy fluctuation variance is inversely proportional to the dynamic decision threshold.

7. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The method for calculating the variance of energy fluctuations includes: Within the current time window, the statistical variance of the instantaneous power sequence of the plasma radiation signal is calculated, and the statistical variance is determined as the energy fluctuation variance.

8. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The determination of the occurrence of keyhole instability defects includes: The coupling stability index is compared with the dynamic decision threshold in real time. If the coupling stability index is less than the dynamic decision threshold, start a timer or counter; When the duration of a state less than the dynamic decision threshold exceeds a preset time window threshold, a defect determination signal is output; if the duration does not exceed the time window threshold, it is determined as an instantaneous disturbance.

9. The real-time laser welding quality evaluation method based on data analysis according to claim 8, characterized in that, The keyhole instability defect includes: At least one of the following: keyhole collapse defect, porosity defect, lack of fusion defect, or hump weld defect.

10. The real-time laser welding quality evaluation method based on data analysis according to claim 1, characterized in that, The evaluation method also includes: Calculate the integral area where the coupling stability index is lower than the dynamic decision threshold, and define it as the instability energy accumulation value; Based on the accumulated instability energy and the duration of instability below the threshold, the current keyhole instability defect is classified into three warning levels: mild, moderate, and severe.

Images