Machine Health Monitoring

This project, in conjunction with ADVIS Inc. is focused on developing a device to cost-effectively bring machine health monitoring (MHM) to a broad spectrum of Department of Defense (DoD) assets (vehicles, pumps, rotating machinery), where the implementation of conventional monitoring systems is cost prohibitive. To meet size (1 cubic inch) and power consumption (battery life of 3 years) requirements, the device utilizes low- power embedded machine learning (ML) models with data from acoustic and vibration sensing systems. Spectral features extracted from a recorded signal can be used to train an embedded ML model to perform tasks such as the detection of anomalies and faults in mechanical systems. As security is a primary consideration for the DoD, the device cannot communicate with any of the vehicle’s electronics or transmit data to the cloud. Instead, faults are detected with distributed sensing elements and machine learning models deployed on near-to-the-sensor hardware. Tre DiPassio (M.S. ‘20, Ph.D. ‘23) and Aaron Bundy (M.S. ‘24) did a lot of great work on this project. We’ve found Edge Impulse to be an outstanding platform to train and deploy neural networks on the edge devices.

Simple Example - Noise Classification

In this simple example, the network is trained to distinguish between white, brown, and pink noise bursts during inference. A different color LED is illuminated depending on which noise burst is detected.

Simple Example - Fault Classification

The above example is expanded to a real dataset. A network is trained using a dataset of labeled air compressor faults and deployed to recognize the healthy state of the device, or identify a specific fault. The dataset includes faults with bearings, pistons, flywheels, belts, and valve leaks, as well as healthy data.