[UC56] Asset Inspection Data Quality and Anomaly Detection

The Utility Problem

Field inspection programs are one of the primary mechanisms utilities use to verify the physical condition of their distribution assets. Inspectors visit substations, pole-mounted transformers, switching equipment, and underground infrastructure on a regular cycle, recording observations about insulation type, condition ratings, oil levels, visible defects, and other physical characteristics. These inspection records are then filed — as paper forms, PDFs, photographs, or structured reports — and stored separately from the master asset data held in the GIS.

The central problem is that inspection findings are rarely cross-referenced against the master asset record in any systematic way. A GIS record may indicate that a transformer is air-insulated, while the inspector’s report clearly documents an oil-insulated unit at the same location. A switchgear entry may list one voltage class, while the nameplate photograph taken during the last inspection shows a different rating. These discrepancies accumulate silently over time — eroding the reliability of the GIS as a source of truth, distorting capacity analysis, and introducing risk into engineering decisions that depend on accurate asset attributes.

Compounding the problem, inspection findings themselves are not monitored for internal consistency over time. A condition that was rated as a major defect in one inspection cycle may reappear as minor or absent in the next — a reversal that could indicate a data entry error, a substituted asset that was never recorded in the GIS, or a genuine remediation that was never formally closed out in the work management system. Without automated monitoring of inspection result sequences, these anomalies go undetected.

The root cause lies in the nature of inspection data itself: it is predominantly unstructured. Inspection reports exist as PDF documents, scanned forms, annotated photographs, and free-text technician notes — formats that do not lend themselves to direct comparison against the structured attribute fields of a GIS record. Reconciling the two has historically required manual review by experienced engineers, which is both expensive and difficult to perform at scale across a large asset fleet.

Awesense’s use cases are constantly evolving. Request free access to our Sandbox trial and GitHub repository to see the full use case descriptions and implementations, and to learn how you can create new use cases yourself.