Innovative Time to Failure Prediction Utility Meters

Target Objective

• Objective: develop innovative AI and machine learning techniques to predict utility meter failures.
• Data to be used: real-time time series and event meter data
• Importance to industry:
  • Can be used to achieve predictive maintenance.
  • Compare to traditional maintenance approaches, such as “run to failure” and “schedule” maintenance, predictive maintenance has several advantages: less equipment down time, fewer interruptions, longer asset life, and increased efficiency

Project Detail

• Approach
  • Classification: use real-time meter data to predict whether a meter will fail in a fixed time period in the future (e.g. one month)
  • Regression: use meter data to predict the time-to-failure (in months) for a meter
  • Will focus on classification first
• Research plans, milestones, and time line
  • Milestone 1: training and testing meter data (in collaboration with WE Energies)
  • Milestone 2: classification algorithms, software, and simulation results
  • Milestone 3: white paper and other publications

Target Outcomes

• Publications
  • White paper(s)
  • Potential journal/conference publications
• Software
  • Meter data analysis and classification software
  • Training and testing data sets for future research
• Education
  • Training for PhD student(s)
  • Inspire potential projects for machine learning classes