Pilot Testing of an AI Algorithm to Identify Fault Category and Fault Cause from DFR Records

When 687 Fault Recorders Generate Too Much Data for Human Analysis

Scottish Power operates a massive transmission network with 687 fault recorders across 202 substations—generating an estimated 300 fault records daily, yet only 15% provide actionable insights. With a six-person analysis team overwhelmed by volume, the utility shifted from proactive analysis of every record to reactive investigation only after control room notifications. But what critical asset degradation signals are being missed in the unexamined 85%?

This groundbreaking paper documents Scottish Power's pilot deployment of machine learning to automatically categorize fault records and identify root causes—solving a problem that defeated previous rule-based automation attempts.

What You'll Discover:

Learn why earlier rule-based automated analysis systems failed—they simply couldn't reliably distinguish between trip events, through faults, voltage dips, switching operations, and other categories, producing too many misclassifications to be useful.

Explore the Random Forest machine learning algorithm trained on 1.5 million historical fault records from multiple utilities, with 45,000 expert-labeled examples. Understand how the team overcame challenges including variable record durations, different sampling rates, multiple channels, and severe class imbalance where some fault types appear once per 10,000 records.

See the impressive results: 97% accuracy across 10 fault categories after iterative refinement, with zero errors detected in 2,100 pilot records analyzed. Witness a correctly identified VT issue that would have been completely missed under reactive manual analysis—demonstrating the algorithm's ability to catch developing problems before they escalate.

Discover the dual capability—not just categorizing fault type but identifying root cause, currently including lightning strikes and VT problems, with vegetation contact next on the development roadmap.

Download this paper to see how AI transforms fault record analysis from an overwhelming manual burden into an intelligent automated system that prioritizes analyst attention on truly critical events.