Low-cost, Reduced-noise Machine Health Monitoring

Back to all technologies
Download as PDF
Researchers at Purdue University have developed a system which allows for efficient, low-cost machine health monitoring. Currently, the performances of traditional commercial sensors such as accelerometers and acoustic emission sensors are far beyond the range of interest in some processes. They often require expensive supporting equipment or acquire more data than is necessary, resulting in computational inefficiencies. The Purdue system uses a classic stethoscope as a sensor and analyzes the data with an autoencoder-based framework. This system can detect anomalies without being fed a training set, nearly eliminates the effect of external noise, and is easier and more cost-effective than accelerometers or acoustic emission sensors. The system has been validated by testing on a 6-axis robotic arm with stethoscopes placed on the wrist and at the base. After collection of test data over 10 days and 7 different local conditions and comparing the daily reconstruction errors to show detection of anomalies, the system succeeded in differentiating normal and anomalous groups with at least 90% success.

-Can detect anomalies without being fed training sets
-Nearly eliminates effect of external noise
-Easier and more cost-effective than accelerometers/acoustic emission sensors

Potential Applications:
Feb 11, 2020
United States
Purdue Office of Technology Commercialization
1801 Newman Road
West Lafayette, IN 47906

Phone: (765) 588-3475
Fax: (765) 463-3486
Email: otcip@prf.org