REDI-PRO Real time Engine Diagnostics-Prognostics
Artificial intelligence aims to increase battlefield readiness by diagnosing engine problems in tanks before costly repairs are needed. Researchers at the Department of Energy's Pacific Northwest National Laboratory developed a prototype system to diagnose and predict failures and abnormal operations in the M1 Abrams main battle tank's turbine engine, the AGT1500. Called REDI-PRO, or Real time Engine Diagnostics-Prognostics, this prototype system was developed for the US Army Logistics Innovation Agency as a proof-of-concept. Originally called TEDANN (for Turbine Engine Diagnostics using Artificial Neural Networks), the name was changed to REDI-PRO to reflect the objective of demonstrating real-time onboard predictive maintenance (prognostics) technology.
REDI-PRO used data from 32 existing sensors on the engine and 16 new sensors that were added using a wiring harness. The sensor data are analyzed continuously by a computer that uses artificial neural networks (ANNs), rule-based algorithms, and predictive trending. The ANNs were trained by example to model normal engine performance and to recognize normal and degraded conditions. REDI-PRO prototypes were installed for data collection purposes on tanks at the Washington State National Guard, Yakima Training Center, Yakima, WA and at the US Army's Yuma Proving Ground.
The prognostics technology employed in REDI-PRO has potential to extend the life of mechanical systems, lengthen the time between overhauls, and save hundreds of millions of dollars in maintenance costs over the system's life cycle. Cost-benefit analyses conducted for the Army projected a return on investment on the order of 10:1 for deploying REDI-PRO prognostics on the Army's legacy fleet of tanks with AGT1500 gas turbine engines. After the conclusion of the REDI-PRO project in the 2001 timeframe, the Army decided to replace the AGT1500 engine with a new model. Lessons learned from the REDI-PRO project benefited the design of the new engine and onboard computational resources.