Evaluation of Artificial Intelligence Implementation in Contaminant Impact Assessments

Every substance transported to the International Space Station (ISS) is rated for its potential to impact human toxicology, flammability, biosafety, and the environmental control life support systems (ECLSS). The results of these ratings are defined in a Hazard Mitigation Summary Table (HMST). The HMST ratings provide the basis for establishing safety guidance for designing for minimum risk during substance transportation, storage, and use. Rating the impact on human toxicology, flammability, biosafety, and ECLSS for every substance utilized on ISS is time-consuming and requires a large team of subject matter experts. This study evaluates using artificial intelligence (AI) to enhance ECLSS HMST assessment productivity. As a case study, the predicted impacts from particulate spills on ECLSS were evaluated. Our HMST application, Safety and EnvironmenTal Intelligence EvaLuator (SENTINEL), implements an AI-powered automated workflow that provides HMST ratings in seconds. SENTINEL employs Amazon Web Services (AWS) Textract and the Anthropic Claude model equipped with custom tool-calling capabilities to autonomously extract critical chemical properties. These properties feed into automated calculations that quantify the impact of particulate contaminants on ISS ECLSS filters. Specifically, the pressure drop across the filter and functional capacity consumption percentages are estimated, corresponding to HMST ratings. The workflow handles complex mixtures by performing proportional mass calculations for trace constituents based on concentration percentages, enabling comprehensive ECLSS impact assessments for multi-component substances. To support NASA engineering verification and eliminate AI misinformation risks, SENTINEL uses PyMuPDF to automatically highlight AI-extracted values directly on the original source portable document format (PDF) documents by employing Textract's spatial geometry data to precisely locate and mark each extracted value with its corresponding confidence score. This AI-driven approach significantly reduces the time required for ECLSS HMST evaluations while maintaining analytical rigor through validated engineering calculations and providing transparent validation mechanisms for subject matter expert review.