You are here

Machine Learning Enabled Near-Real-Time Situational Response for Mechanical Systems



OBJECTIVE: Machine Learning system utilizing real-time analysis of Health and Usage Management Systems data for predicting allowable control parameters enabling mission completion under degraded system performance. 

DESCRIPTION: The Army is seeking novel approaches implemented in a “Black Box”, to receive and analyze heterogeneous real-time sensor data (e.g. unsynchronized data from different types of sensor signals) to enable situational response to maximize in-situ operational performance and to complete mission requirements. The “Black Box” is a device that ingests mission parameters, sensor outputs and/or Health and Usage Management Systems (HUMS) data from a defined system of interest. The output of the “Black Box” are the system executable set of control parameters to meet preset mission requirements. Under normal operation with full system capability the control parameters may be generated for maximum overall performance however upon recognizing a subsystem failure or reduced performance, the “Black Box” will provide modified control parameters minimizing the impact of the affected subsystem. The system monitored by the “Black Box” can be as complex as an unmanned autonomous system or a subcomponent within the system (e.g. transmission, gear box, engine, structure, electronic power). Research and approaches to achieve the objective should include Implementation of new or existing Machine Learning algorithms into an Integrated programmable software-hardware “Black Box”. Technology for this capability may include state-of-the-art data-networking and high performance neural enabled hardware. The proof-of-concept will initially be directed toward monitoring input data from structural or power transfer components (e.g. transmissions or gear sets for rotorcrafts and ground vehicles) and subsequent modification of control when an anomaly is detected. For example, the “Black Box” will operate in a learning capacity during normal state of operation, however upon detection of a failure or precursor to failure the “Black Box” will provide user options to limit the available power applied to a failing gear set to maintain mission effectiveness or safe return. The size, weight, and power requirement of the “Black Box” should not exceed that of a state-of-the-art HUMS for the selected demonstration. The “Black Box” concept is intended to be saleable from a simple to a complex system of Army interest and easily be transitioned to commercial applications in the automotive or aerospace industry. 

PHASE I: Define innovative approaches for enabling near-real-time assessment and prediction of remaining serviceable life of a simple system (e.g. structural, mechanical systems or subsystems relevant to one or more categories of ground, air, and autonomous vehicles). These approaches will utilize Machine Learning hardware and software to evaluate real-time sensor data in conjunction with surrogate (proposed) or historical time-data benchmarks to provide modified control parameters (e.g. for structural or mechanical systems). Hardware, software, and combined approaches should be considered. (e.g. high-throughput CPU’s (central processing units) designed for neural engines, implementations of machine learning algorithms based on novel hardware, etc.) 

PHASE II: Establish and expand the Phase I proof-of-concept through the development, testing, and demonstration of a “Black Box” system that will capture and process near-real time data for a proposed Army system component (e.g. structural, mechanical and propulsion systems on an unmanned air vehicle or a ground combat vehicle). The system will provide modified operational control parameters to adjust the flight or driving patterns to extend usable system life and meet mission objectives. The form factor for the “Black Box” shall not exceed that of a state-of-the-art HUMS on the proposed system component demonstrator. 

PHASE III: Provide an adaptive and saleable “Black Box” capable of real-time monitoring and situational response applicable to air, ground, and autonomous systems and subsystems, (e.g. structural components, mechanical, power transfer, and drive systems relevant to both Army and commercial systems). For example, this technology could be applied as an oil pressure monitoring system in military or commercial vehicles. If a significant drop in pressure is sensed, it will provide the usual driver warning, but will also allow the vehicle to continue operation by recommending or automatically implementing measures determined previously through ML, such as limiting top speed, or redirecting cooling capability to the failing area that are most effective in educed performance from prior machine learning data to reach intended mission without further subsystem failures. 


1: Wade, Daniel R., et al., "Machine Learning Algorithms for HUMS Improvement on Rotorcraft Components", Paper presented at the AHS 71st Annual Forum, Virginia Beach, VA May 5-7, 2015 (Distribution Unlimited per AMRDEC PAO (PR 1608).

KEYWORDS: Machine Learning, Artificial Intelligence, Real-Time Control, Health And Usage Monitoring (HUMS), Sensors, Neural, Central Processing Unit (CPU) 


Eric Mark 

(410) 278-6457 

Dr. Asha Hall 

(410) 278-2384 

US Flag An Official Website of the United States Government