Office of Technology Transfer – University of Michigan

Equipment condition monitoring and decision-making using nonlinear regression

Technology #6591

Questions about this technology? Ask a Technology Manager

Download Printable PDF

Categories
Researchers
Jerome P. Lynch
Managed By
Drew Bennett
Associate Director - Software Licensing 734-615-4004
Publications
Automated Analysis of Long- Term Bridge Behavior and Health using a Cyber-Enabled Wireless Monitoring System
Proc. SPIE 9063, Nondestructive Characterization for Composite Materials, Aerospace Engineering, Civil Infrastructure, and Homeland Security 2014, 90630Y (10 April 2014); doi: 10.1117/12.2045244, 2014

Whenever manufacturing machinery unexpectedly goes offline, considerable money and productivity is lost. Equipment condition monitoring is a way to avoid that by constantly monitoring machine conditions for variations indicating faults that require preventative maintenance. At its base, it involves reading vibration, ultrasound, thermography, and other monitoring data and comparing them against known limits for normal operation. However, the advent of big data and distributed computing now allows for a more effective method of turning monitoring data into actionable directives.

Big data statistics power complex condition recognition

The new method starts with an abundance of low-cost wireless sensors that are easy to deploy and provide multiple sources of complementary data. Then, big data statistical algorithms are applied, using non-linear multivariate regression to classify operation into a variety of normal or abnormal states. This method allows for enormous flexibility in recognizing patterns of operation, such as normal changes during time of day or when asynchronous machines occasionally match operating nodes. The result is adaptable recognition of normal states and tighter bands to better recognize abnormal outliers.

Applications

  • Complex machine condition monitoring
  • Large area structural health monitoring

Advantages

  • Advanced statistical approaches provides better outlier recognition
  • Algorithms optimized for low-power embedded systems
  • Low cost and resource requirements enable greater scalability