Predicting Ammunition Condition Across the Storage Life Cycle Using Machine Learning and Environmental Anomaly Features: A Case Study of Republic of Korea Army

초록

This paper introduces the relationship between storage conditions of ammunition depots and the life cycle of ammunition through a data-driven approach. Ensuring the reliability of long-term stored ammunition is a critical challenge for defence organizations, particularly in countries with prolonged military readiness such as the Republic of Korea. Traditional maintenance approaches based on periodic inspections fail to adequately reflect environment influences and evolving degradation trends. This research proposes a data-driven framework for predicting the condition of ammunition by integrating three types of real-world defense data: historical inspection records, condition transition logs, and environment sensor data from the depots. This research focuses on three ammunition types (81mm mortars, 105mm high-explosives, and 155mm high-explosives), which are mainly used in Republic of Korea Armed Forces. Each ammunition data is composed of unified datasets at the lot level that include manufacturing date, inspection metadata, storage duration, humidity and temperature of depots. It is known to the military field that abnormal fluctuations in temperature and humidity significantly affect functional degradation of ammunition degradation. To analyze these effects, we installed Supervisory Control and Data Acquisition (SCADA)-based sensors in each type of ammunition depot (igloo, above-ground, container) hereby collecting temperature and humidity data and defines anomaly features using interquartile range thresholds. These are incorporated into boosting based supervised machine learning models to life cycle of ammunition. The proposed models achieve approximately 90% accuracy across most datasets. Currently Republic of Korea Army is developing a condition-based ammunition monitoring system by expanding the methodology presented in this paper.

키워드