Intelligent Process Control for Additive Manufacturing of Alloys

Thesis Abstract

This study addresses the critical challenge of achieving consistent material properties and defect-free builds in additive manufacturing (AM) of metallic alloys through intelligent process control (IPC) that integrates real-time sensing, data analytics, and adaptive actuation. Despite advances in AM hardware, variability in laser/arc power, scanning strategy, powder characteristics, and thermal histories leads to residual stresses, porosity, anisotropy, and part distortion, limiting industrial adoption. The aim is to develop and validate an IPC framework that leverages in-situ sensor fusion, machine learning-assisted process optimization, and model-predictive control to stabilize thermal cycles and microstructure evolution during laser-based and arc-based AM of nickel-based, titanium-aluminum, and titanium-alloy systems. Specific objectives include (1) characterize process-structure-property relationships under varied process conditions using designed experiments; (2) develop a real-time sensing suite (pyrometric, acoustic emission, melt-ppool imaging, and powder-feeding sensors) and fuse data through Bayesian and Kalman filtering to estimate instantaneous thermal history and predicted microstructure; (3) formulate a model predictive control (MPC) strategy informed by a physics-informed neural network (PINN) surrogate of heat transfer and solidification kinetics; (4) validate the IPC framework on multi-material AM test coupons and functional components with targeted grain size control, reduced porosity (<0.5%), and minimized residual stress (<200 MPa) using neutron diffraction and X-ray diffraction residual stress measurements; (5) assess robustness across machines, powders, and build geometries, and (6) perform techno-economic analysis to quantify performance gains and implementation barriers. The methodology adopts a mixed-methods research design combining experimental, computational, and analytical components. The population comprises powder-bed fusion AM machines across three industrial facilities, with a sample of 60 test coupons and 6 demonstrator components produced under varied process recipes and environmental conditions. Data collection instruments include in-situ pyrometric temperature sensors, high-speed melt-ppool cameras, acoustic emission sensors, high-resolution infrared thermography, powder flow and laser/arc power monitors, and post-build characterization tools (electron backscatter diffraction, scanning electron microscopy, x-ray computed tomography, and microhardness mapping). The analytical framework integrates (i) multivariate regression and Gaussian process regression to model process-to-property mappings, (ii) Bayesian data fusion to synthesize heterogeneous sensor signals, (iii) model-based MPC with a physics-informed neural network (PINN) surrogate for heat transfer, solidification, and phase transformation kinetics, and (iv) statistical hypothesis testing (ANOVA) to evaluate process parameter significance. Validity and reliability are ensured through calibration experiments, repeatability trials, cross-validation of predictive models, and sensitivity analyses. The study employs theoretical grounding in control theory (MPC and robust control), data-driven modeling, and materials science (dendrite arm spacing, grain growth, and phase fraction evolution) to establish a cohesive IPC framework. Expected findings include improved process stability evidenced by reduced coefficient of variation in melt pool temperature, decreased porosity and hot-cracking incidence, and more uniform microstructures across deposited layers. The MPC-PINN model is anticipated to outperform conventional feedback control by achieving tighter control of peak temperatures, faster settling times after disturbances, and more consistent phase fractions in target alloy systems. The research is expected to demonstrate transferability of the IPC framework across nickel-based superalloys and titanium alloys, with a scalable data architecture enabling rapid adaptation to new powders and machines. The study will contribute to knowledge by integrating physics-based heat transfer with data-driven process inference to realize autonomous AM process control, presenting a transferable, modular IPC architecture and a validated surrogate model suite. The theoretical contribution includes empirical validation of the combined Bayesian data fusion and PINN-enabled MPC paradigm for complex, non-linear, time-dependent additive manufacturing processes, extending the literature on intelligent manufacturing in the metallurgical domain. The main conclusion is that intelligent process control can significantly enhance the reliability and performance of AM alloys by real-time fusion of multisensor data, physics-informed predictive modeling, and optimized actuation of process variables. Recommendations include investment in integrated sensing hardware packages, development of industry-standard IPC interfaces for machine interoperability, and the adoption of the proposed MPC-PINN framework as a stepping-stone toward autonomous, closed-loop AM production lines. Further research directions suggest exploring reinforcement learning for adaptive policy refinement, extending the framework to novel alloy systems, and conducting long-term durability assessments of IPC-assisted AM components under service conditions.