Corrosion Data Analytics for Material Selection, Corrosion Control and Reduced Ship Maintenance

Navy ships are encountering critical issues of corrosion from harsh sea environments, resulting in increased maintenance costs and delays, and ships being decommissioned prior to their expected lifetime.  To mitigate this situation, a machine learning (ML)-based data analytics solution has been envisioned.   In this STTR, CFD Research team will develop an ML-based Corrosion Analytics Toolkit (CoAT) for making informed decision for material selection and corrosion control measures. Phase I will focus on collecting data from open-source reports and journals, extracting knowledge and analyzing the data using various ML techniques.  Our STTR collaborator, the University of Maryland, has developed a Semantically-Structured Knowledge Graph approach for representing highly technical and scientific information. Leveraging principles of Natural Language Processing (NLP) and Large Language Models (LLMs), their approach has proven successful for automatic data extraction and creating knowledge graphs in multiple application areas. In addition to text-derived features, the data will include features such as the alloy composition, salinity, pH, ocean depth, temperature, oxygen concentration, global location, and properties such as corrosion rate, pit depth, and type of corrosion. Additional composition-based features for the alloys will be generated. The underlying relationships in the features and material corrosion properties will be assessed through a combination of unsupervised clustering and supervised property prediction. The key parameters for corrosion will be identified using SHapley Additive exPlanation (SHAP) method with an objective to determine criteria for material selection and corrosion control measures.  A preliminary graphical user interface of CoAT will be developed.  In Phase II, data collection will be expanded to include failure analysis, inspection reports, maintenance records, and coalescence into a unified materials database.  NLP- and LLM-based models for data extraction will be fully developed to create knowledge graphs which can be utilized for informed decision making for corrosion control and reduced ship maintenance.  Comprehensive supervised models for corrosion prediction rate, pit depth, and operational lifetime (where data is available) will be developed and uncertainty analysis will be performed.  These models will be experimentally validated through external collaboration. Predicted corrosion control measures and material selections will be verified against experiments.  A fully functional prototype of CoAT will be developed at the end of the Phase II.