An Innovative Approach to Leverage System Safety MBSE Model Information Using AI/ML

During Phase II, the Penta-INTUITIVE team plans to expand the Model Based Systems Engineering (MBSE) model established in the Phase I Base effort and further embrace automation to improve the efficiency of the Navy munition and missile system safety process. The team will focus on using AI/ML techniques such as Graph Databases, Large Language Models (LLMs), or Retrieval-Augmented Generation (RAG) models to extract requirements from government standards and place those into the MBSE model. The team will also focus on using the same AI/ML techniques to take System of Interest data (either historical report data or raw test data) and import that information into the MBSE model. In parallel, the team would work on expanding the MBSE model and creating the necessary traceabilities for the additional requirements and test data that will be added. The Penta-INTUITIVE team plans to accomplish two main goals during the Phase II Base periodĀand plans to accomplish an additional goal during the Phase II Option period, if awarded.ĀĀ During the Phase II Base the Team plans to develop and implement AI/ML techniques to automatically vectorize and import standardized requirements for the Navy munition and missile system safety process, as well as develop and implement AI/ML techniques to automatically cultivate, categorize, structure, and import Test Data into the MBSE model. Then, the Team will establish traceability within the MBSE model between test data and corresponding requirements to determine if the test data satisfies the thresholds established in the requirements. Once the requirements and test data have been imported, and the traceability has been established. The Team plans to develop a Mitigation Recommendation Model that will review historical FMEA/FMECA reports to determine potential comparable failure mode(s) to the userÆs current System of Interest (SoI). Once this data has been reviewed and added into the AI/ML model, it will undergo matrix factorization. Then, the K-Nearest Neighbors (KNN) method will be utilized to recommend mitigations along with predictions for how acceptable/favorable each mitigation will be evaluated.Ā Finally, during the Phase II Option effort the Team plans to train a classification model with test data of historical munition and missile systems to create a prediction model of munition and missile system test data. Then the Team will develop a similarity classifier using the K-nearest neighbors algorithm (KNN) to locate comparable test outcomes in the matrix and report the fitness of each outcome. Finally, to test the validity and robustness of this prediction model, the Team plans to systematically remove known test points of available components, munition systems, and tests to quantify modelÆs sensitivity to test gaps.