In-situ defect detection and material property evaluationin additive manufacturing using acoustic signal and machinelearning

In-situ monitoring is crucial for detecting defects and estimating material properties to ensure the quality of printed parts in additive manufacturing. Acoustic signals produced during the interaction between the laser and material contain critical information about complex physical mechanisms such as crack formation. However, acoustic-based monitoring in laser powder bed fusion (LPBF) has received little attention due to the noisy environment. This study presents an acoustic-based real-time process monitoring method integrated with machine learning for LPBF process. The key contribution lies in developing a feature extraction approach that utilize machine learning models such as random forests and k-nearest neighbors (KNN), and wavelet transform for defect detection and material property classification. Microphone data collected during LPBF experiments capture both laser-material interaction signals and environmental noise, including contributions from the laser, fan, and powder flow. A bandpass filter is applied to isolate relevant signals, followed by wavelet transform in time- and frequency-domains to obtain representation of the laser-material interaction. The results show that the machine learning models achieve an average material property estimation accuracy of 89%, highlighting its effectiveness in enhancing the monitoring process parameters.