A Computational Framework for Estimating Days of Maintenance Delay of Naval Ships

This work proposes a computational framework for estimating Days of Maintenance Delay (DoMD) for US Navy ships, aiming to improve fleet maintenance planning. The data, containing Controlled Unclassified Information (CUI), is obfuscated and consists of both time-dependent and time-invariant attributes, forming a "fat" tensor with many attributes but few instances (around 200). The solution is a predictive maintenance pipeline that uses obfuscated data for training and then retrains on raw data in the Navy environment without human intervention. The pipeline includes modules for transforming raw data, identifying effective features, selecting machine learning models, and training models robust to outliers and noise. The framework addresses computational challenges and optimization opportunities, and its effectiveness is demonstrated experimentally within the Navy environment.