Accelerating the Development of CMAS and Sulfate-Induced Hot Corrosion Resistant Materials through Machine Learning

For decades gas turbines have been a reliable source of propulsion for a variety of marine-based vehicles. Since their adoption as the go-to solution for vehicles requiring a higher power density than conventions diesel engines can produce, they have suffered from corrosion of their alloyed components. Long believed to be the primary source of the corrosion, great efforts have been made to reduce the Na2SO4 and CaSO4 contamination in the fuel sources. Despite recent efforts to reduce sulfur content to near zero levels (<15 ppm), sulfate induced hot corrosion of alloyed components in marine gas turbine engines still occurs. In fact, hot corrosion can be caused by the silicate and sulfate species in the air that these engines need to operate. Turbine engines also suffer from CMAS attack on components which have thermal barrier coatings (TBCs) and environmental barrier coatings (EBCs) applied. These coatings are subject to degradation when debris composed of calcium-magnesium alumino-silicates (CMAS) is ingested into the engine, melts in the turbine hot-section, and deposits on the coated components. The CMAS reacts with the coating and degrades the mechanical properties of the coating during temperature cycling which occurs during normal engine operation. Models linking the thermochemical and thermomechanical degradation of the TBCs due to CMAS are needed to understand life of the coatings and to identify best strategies for developing improved coating systems. One of the largest challenges in developing calcium-magnesium-alumino-silicate attack (CMAS) and calcium sulfate hot corrosion resistant coatings is the high number of variables that characterize the material behavior. Different phases form as the corrosion progresses which makes conventional modeling tools ineffective to capture the phenomena that drive coating failure. Typical ICME approaches require problem simplification to dissect the problem into discrete parts that can be modeled and then uses them in conjunction to predict behavior. These approaches are most effective with a low number of inputs which can be characterized in relationship to one another. Machine learning is the science of programming computers to learn from data without having to identify underlying relationships or underlying behavior. Machine learning algorithms excel when there is a large number of inputs that are difficult to characterize by leveraging modern computing power to extract underlying trends. Developing corrosion resistant coatings is an ideal application for machine learning due to a large number of different factors that drive performance. The proposed program seeks to employ machine learning algorithms to accelerate the development of CMAS and sulfate hot corrosion resistant materials. In addition to the ML modeling, MR&D will enhance an existing CMAS durability model to provide additional data for the ML models.