Quantum Kernel Models for Rapid Fatigue‐Life Estimation From In Situ Melt‐Pool and Acoustic Emission Signals

In safety‐critical manufacturing domains such as laser powder bed fusion and directed energy deposition, fatigue‐life prediction for aerospace brackets, biomedical implants, turbine repair parts, and structural coupons must be performed accurately and rapidly because undetected degradation can cause catastrophic failure. Existing, data‐driven fatigue prediction methods use postprocess inspection or single‐modal sensing and classical kernel methods, which have limited temporal resolution, decreases prediction accuracy with nonstationary process and reduces scalability for real‐time use. To overcome these limitations, the present study proposes a QK‐FEF that combines quantum kernel learning with multimodal in situ sensing to predict fast fatigue‐life. It incorporates time‐resolved melt‐pool thermography and broadband acoustic emission signals that are coded into structured feature embeddings by way of physics‐friendly temporal coding and frequency‐energy descriptors. On an eight‐qubit variational circuit, a parameterized quantum feature map is generated and a nonclassical kernel is created that provides class separability and nonlinear regression over the traditional radial basis and polynomial kernel. The model was assessed with 4567 synchronized build‐cycle records of a stainless steel LPBF system, including 2914 fatigue tested specimens and 1653 validation samples with fatigue life measured in cycles to failure. Experimental results show a coefficient of determination of 0.942, an absolute mean of 3870 cycles, and 21.6% less predictivity error than with optimal Gaussian process regression. The proposed method is scalable for high‐end manufacturing environments and can provide near‐real–time fatigue‐life estimation.