From reactive- to condition-based maintenance: Artificial intelligence for anomaly predictions and operational decision-making

The USGS maintains an extensive monitoring network throughout the United States in order to protect the public and help manage natural resources. This network generates millions of data points each year, all of which must be evaluated and reviewed manually for quality assurance and control. Sensor malfunctions and issues can result in data losses and unexpected costs, and are typically only noticed after they occur during manual data checks.  

By connecting internal USGS databases to “always-on” artificial-intelligence applications, we can constantly scan data-streams for issues and predict problems before they occur. By connecting these algorithms to other cloud-hosted services, the system can automatically notify staff about potential issues so prevantative maintenance or repairs can be scheduled, minimizing data loss and malfunctions. This predictive intelligence can help operational decision-making within USGS, improve data quality, and is projected to reduce network maintenance costs. 

This CDI proposal is to 1) conduct a USGS-wide survey to identify opportunities for using artificial intelligence in day-to-day operations and databases USGS-wide, and 2) scale a pilot, cloud-hosted machine-learning application and early alert system for problems in USGS water quality sites and develop broadly applicable algorithms for detection sensor abnormalities.



Principal Investigator : Matthew J Cashman

Co-Investigator : Michelle P Katoski, Todd Lester

Cooperator/Partner : Matthew Kockuk