Monitoring and assessment of weld penetration condition during pulse mode laser welding using air-borne acoustic signal

Real-time monitoring system is one of the essential criteria in the era of the fourth industrial revolution (Industry 4.0). Among the monitoring systems in laser welding applications, acoustic methods have recently caught the attention of researchers due to their benefits in promoting simple, low-cost, and non-contact systems. However, applying this method in PW mode laser was challenging due to the different characteristic of signal and noise acquired from this process as compared to CW process. Therefore, this particular work aims to investigate the characteristics of acoustic sound signal from PW Fiber laser, develop an appropriate signal processing algorithm to suppress the effect of noise on the extracted sound features, and develop an empirical model for weld depth estimation. To achieve the objectives, a 1.8 mm thick 22MnB5 boron steel plate was welded with varied laser peak power (PP) and pulse duration (PD) levels. Simultaneously, the sound signal was acquired between the frequency of 20 Hz to 12.8 kHz throughout the process. Signal features, such as mean absolute deviation (MAD), standard deviation (SD), kurtosis (K), L-scale (LS), L-kurtosis (LK), bandpower (BP), and sum of synchrosqueezed wavelet coefficient (CSqWCsum) were extracted from the acquired sound. To develop the signal processing algorithm, multi-lag phase space (MLPS) method was adopted in which some modifications on its original algorithm were made by introducing the localized crest factor (CF) thresholding method to reduce the influence of noise. Results showed that the acquired sound recorded transient behaviors with a slight change in its overall amplitudes with respect to the change in the level of weld parameters. Meanwhile, the dominant frequency was found to be fluctuated between 5760 Hz and 7000 Hz without a clear pattern in the case of different levels of weld parameters involved in this study. The results from feature selection analysis show that the combination of SD, L-kurtosis, and modified-MLPS recorded the most significant relation with weld penetration. Furthermore, the combination of these features with the laser peak power and pulse duration recorded a better regression trend with an adjusted R-squared of 0.937. Two empirical models for weld depth estimation were developed from the combination of these sound features and weld parameters using the multiple linear regression (MLR) and artificial neural network (ANN) methods. Through MLR method, the obtained model was DOP = 0.634SD - 0.814LK + 0.0014MLPS + 116.44PD + 0.0014PP - 0.7781. Results from the model validation analysis showed that both models could significantly estimate weld penetration during the PW laser welding process with an estimation error less than 8%. However, the ANN model recorded a more accurate and precise estimation with the lowest estimation error, i.e., 3.3%. The results of the analysis suggest that the acoustic methods can be used to monitor weld penetration on a real-time basis during PW mode laser welding process. Moreover, the methods can also be used to provide a quantitative assessment on weld penetration during the process. This finding gives alternative solution to the development of a real-time process monitoring system in PW mode laser welding, which aligns with the criteria needed in the new era of manufacturing system