Description
Directed Energy Deposition (DED) is a metal additive manufacturing process with applications in demanding industries with stringent requirements. Meeting those requirements is critical to enable the adoption of DED, but detecting and avoiding material defects remains a challenge due to the complexity of the process. In this work, an experimental plan covering several classes of anomalies is conducted and coaxial near-infrared images are collected. This dataset is used in conjunction with expert knowledge of the process and material analysis to build a data-driven model that flags irregular melt pools that could lead to defects.The MiCLAD machine, designed at VUB (Belgium), is equipped with an in-situ hyperspectral near-infrared camera that is monitoring the melt pool during the building process. This camera captures the emitted light from the hot melt pool at 25 different wavelengths. In contrast to greyscale cameras that are typically used for DED in-situ monitoring, the hyperspectral camera essentially captures 3D data, with wavelength as the third axis. Deep Learning techniques enable the analysis of this high-dimensional data.
In this study, an input vector of several consecutive melt pool images is constructed and fed to a Convolutional Neural Network to detect and classify typical anomalies seen in DED (e.g. edge defects, keyhole porosities, lack-of-fusion, etc.). The influence of several hyperparameters (input vector length, amount and size of layers and number of considered input wavelengths) on the detection efficiency is studied and optimized. The novel aspect of this study is the combined use of hyperspectral in-situ data with advanced Convolutional Neural Networks in the context of DED anomaly detection. Anomaly detection is a step towards feedback control of the DED process to avoid defects. In this regard, the inference speed of the optimized model is measured and potential ways to improve it are identified.
| Periode | 7 sep. 2023 |
|---|---|
| Evenementstitel | European Congress and Exhibition on Advanced Materials and Processes |
| Evenementstype | Conference |
| Locatie | Frankfurt, GermanyToon op kaart |
| Mate van erkenning | International |