An interpretable machine learning framework for the prediction of cracking in additively manufactured high gamma-prime Nickel-based superalloys

Cracking during directed energy deposition limits the processability of components fabricated from high γ′ Ni-based superalloys and demands prediction tools that are both accurate and interpretable. This work develops a lightweight, data-driven framework that correlates in-situ infrared thermal histories to crack susceptibility in IN625–IN100 graded alloy mixtures spanning compositions with varying γ′-forming precursors (Al, Ti). Thirty-three thin walls were deposited across a range of compositions and process parameters, generating diverse thermal histories for analysis. Pixel-level IR traces (9 × 27 grid) from identical locations of each deposit are extracted and pre-processed, with the nine lateral traces within each layer concatenated to produce 891 layer-level profiles. Principal component analysis (PCA) compressed these high-dimensional profiles into low-rank, physically meaningful features (> 90% variance in the first ten PCs). Three input sets, CP-P (composition + process), CPS-P (composition + process + PC scores), and CS-P (composition + PC scores), were benchmarked using various machine learning classifiers. Using standardized inputs for classification, CPS-P with a Gaussian Process classifier (ARDSE kernel) achieved the best performance (accuracy = 0.96; F1 = 0.92; recall = 0.92), while CS-P performed comparably, indicating that thermal histories largely subsume nominal process metadata. PC basis plots capture distinct thermal-behavior signatures across process inputs, and feature-relevance analyses (Random-Forest importances and ARDSE per-feature length-scales) consistently ranked low-order PC scores and composition as most influential. Retraining the best model enabled crack-probability maps over the composition–parameter space, revealing sharp susceptibility transitions near ~ 87.5 wt.% IN100 under no-dwell and strong mitigation with a 10s dwell. The resulting PCA-based pipeline is interpretable and computationally light, potentially supporting process planning and in-situ quality assurance in metal additive manufacturing.