AI-based defect detection and self-healing in metal additive manufacturing

This pilot study develops a process to evaluate in-situ defect detection and self-healing in Ti-6Al-4V fabricated using laser-based powder bed fusion. A tailor-made test specimen was designed and manufactured for the nanofocus tube X-ray computed tomography (XCT) system. In situ optical tomography was used to capture infrared images containing heat signatures of the hot laser interaction zone. Depicting natural process variation, defective regions were seeded using process manipulation (up to ±30%) in proximity of the experimental standard volumetric energy density (VED). The concomitant defects and heat signatures were both spatially and temporally captured. The results indicate that porosity significantly grows from an average value of 27 parts per million (PPM) to a value of 337 PPM comprising defect sizes of <112 µm when the VED increases by 30%. The outcome confirmed that Ti–6Al–4V can self-heal these defective regions by up to 7 ± 1 layers using the standard VED. A convolutional neural network was trained (n = 211) and was verified with XCT. The model demonstrated prediction accuracy of 94% for the six classes of unfamiliar defective regions. This work enables in-situ detection and healing of defective regions caused by process uncertainty that can shift the quality frontier of novel product design and development.