Happy Friday, FME Community! Welcome back to TGIF: Thank Goodness It’s FME!
This week, we hav Crystal, one of our Support Specialists, here to share a tutorial on using FME to automate quality assurance for field asset inspections with a little help from AI. The workflow puts AI to use as a second set of eyes on field asset inspections by automatically catching condition mismatches, enriching data, flagging problem assets, and delivering a polished weekly report straight to your reviewers' inboxes.
The problem
Field inspections generate a lot of data, from ratings to map points. Manually reviewing it all for accuracy and consistency is tedious, and mistakes could easily slip through.
The solution
Every time an inspector submits a form, FME fetches the photo, sends it to an AI vision service for a condition assessment, and compares it against the inspector's rating. Discrepancies are automatically flagged and stored in a database. Every week, a PDF report of all flagged assets is automatically generated and emailed to reviewers.
The result is an automated QA process on every inspection, reducing manual review workload. Now let’s move onto the exciting part, the actual FME workflows used to carry this out!
The Workspaces
Part 1: Real-Time Inspection Pipeline
FME Flow Automation
- A webhook listens for incoming Survey123 submissions.
- A Run Workspace action is then triggered.
FME Form Workspace
- The raw JSON payload from Survey123 is parsed, pulling out survey attributes, geometry, and attachment information.
- The inspection photo is fetched from ArcGIS Online and extracted into a raster BLOB.
- The raster is sent to an AI vision service (we used Google Gemini in the workflow), which rates the asset condition (new to critical/unsafe). A structured JSON response is returned containing a rating, confidence score, detected asset type, and reasoning.
- If the image is not clear or depicts an asset with a severe rating, then the record is flagged. All of the results are joined back to the survey record and written to a SpatiaLite database.
Part 2: Weekly Report
FME Flow Automation
- A Schedule trigger runs every week, automatically kicking off the report workspace.
- The output PDF report is emailed to reviewers.
FME Form Workspace
- Reads the enriched inspection records with the AI-enhanced attributes, only keeping the records that the AI has flagged
- These records contain images of assets that were deemed critical, broken, unsafe, severely damaged, or photographed at a poor quality.
- Dynamically calculates the past week's date range at runtime and filters to keep only records added within the past week.
- Creates a PDF report that highlights each flagged asset, showing the inspection photo, the inspector's rating, and the AI's rating with reasoning.
Read the full article here!
Automated QA for Field Asset Inspections Using AI, Webhooks, and FME