Machine Learning to optimize durability of TBC columnar structure by Suspension Plasma Spray

This work presents a methodology for data-driven artificial intelligence (AI), probabilistic machine learning (ML), and uncertainty quantification algorithms to maximize the durability of columnar suspension plasma spray (SPS) thermal barrier coating (TBC). The material informatics framework targets SPS process optimization to improve TBC durability.

First, the SPS process parameter window for columnar microstructures was identified using a Latin Hypercube design of experiments. A Gaussian process classifier was then applied and validated to predict the columnar processing domain doubling the proportion of columnar conditions sprayed.

Thermal gradient tests were performed on a subset of the columnar samples to assess TBC durability. Multiple SPS columnar coatings demonstrated greater than the target 40% life increase. Durability data and synthetic data from the columnar classifier were used to train a Bayesian Hybrid Model (BHM)/IDACE model, recommending new process conditions to maximize columnar TBC coating life. These IDACE generated highest probability SPS processing conditions were fabricated and durability tested.

This material is based upon work supported by the U.S. Department of Energy’s Office of Energy Efficiency and Renewable Energy (EERE) under the Advanced Materials and Manufacturing Technologies Office DE-FOA-0002553 AMO Multi-Topic FOA Award Number DE-EE0010213.