[Data] Deep learning-based monitoring of laser powder bed fusion process on variable time-scales using heterogeneous sensing and operando X-ray radiography guidance

The proposed Deep Learning (DL) model in this study was trained using signals obtained from a comprehensive time-synced sensing system, incorporating four sensors: back reflection (BR), Visible, Infra-Red (IR), and structure-borne Acoustic Emission (AE). The dataset comprises time-synced signals corresponding to each layer produced by Laser Powder Bed Fusion (LPBF) of SS-316L, capturing three distinct built regimes (Lack of Fusion (LoF), conduction mode, and Keyhole) across nine different parameter sets. Each parameter set uniquely generates one regime, resulting in three parameter sets associated with each regime. These signals from the four sensors are then organized into four distinct running windows (w1, w2, w3, and w4). The durations of these windows are 0.83, 1.65, 2.5, and 3.30 ms, translating to 2500, 5000, 7500, and 10000 data points, respectively, if sampled at 3 MHz. These windows, of varying lengths, are computed across all three processing regimes without overlaps. The naming convention for the .npy file within the .zip file follows this format: channel[a]_[b], where 'a' represents the source channel (0 for back reflection, 1 for Visible, 2 for Infra-Red (IR), and 3 for structure-borne Acoustic Emission (AE)), and 'b' indicates the number of data points in each window (2500, 5000, 7500, and 10000). The dataset is organized based on window length and further grouped by the four channels. Consequently, four datasets are created, each corresponding to one of the four window lengths, for the training of a Convolutional Neural Network (CNN). The operando X-ray imaging analysis ensured that the regimes correlated with the defined process parameters.