PROCESS SIGNATURE CHARACTERIZATION AND ANOMALY DETECTION WITH PERSONALIZED PCA IN LASER-BASED METAL ADDITIVE MANUFACTURING

During Laser-Based Metal Additive Manufacturing (LBMAM), the interaction between the energy source (i.e., the laser beam) and the material generates spatter tracks along the laser scan path. The spattering behavior is the “process signature” reflecting the process stability and anomalies that potentially lead to defects in the printing results. In-situ thermography captures the spatter patterns in real-time, which are crucial evidence for process monitoring and anomaly detection in LBMAM. A critical aspect of anomaly detection in LBMAM is identifying the unique spatter pattern in a thermal image that differs from the generic spattering behavior in the current layer being deposited (i.e., “outliers"). It is equivalent to separating the “local features” and layer-wise "shared features” from the same thermal image and recognizing the local anomaly, i.e., recognizing the outlier “local features.” In state-of-the-art, thermal-image-based anomaly detection in LBMAM relies on conventional feature extraction procedures, which limit the “view” of information to each thermal image and cannot separate the layer-wise spatter pattern governing these thermal images. Consequently, the features extracted are less interpretable, and the anomaly detection method cannot distinguish the abnormality in the current process signature from the layer-to-layer evolution of spattering dynamics. Targeting the limitation in state-of-the-art literature, this study adopts personalized Principal Component Analysis (PerPCA) as a novel means for process signature characterization and anomaly detection in LBMAM. Designed for heterogeneous data with shared features, PerPCA simultaneously extracts the “local features” and “shared features” from the same image. An outlier detection strategy can be designed upon PerPCA that removes the shared features from a process signature and examines the local features at the current printing time for abnormality. The proposed method is demonstrated with selected data from the 2018 AM Benchmark Test Series from the National Institute of Standards and Technology (NIST). The results show good potential for extending the method to complex geometry built with LBMAM.