Condition-based monitoring from the photos your technicians already take

1. A technician takes a photo on the usual round, through your inspection app.
2. StayCharted grades it, on your scale, with a confidence: corrosion 1 to 5, or good / worn / replace soon / replace now.
3. Confident grades are recorded against the asset, with the date.
4. Unsure grades go to the Review Queue on Essentials and Business. An inspector confirms or corrects them, with similar past photos beside each one.
5. Your asset data shows the rate of change: grades per month, and any sudden jump between visits.
6. Corrections can teach the next version. Retrain, evaluate and publish it before using the updated model.

Condition monitoring pays off wherever equipment wearing out causes downtime, repair costs or safety risk. Monitoring tells you the condition; condition-based maintenance uses it to decide when to act. Where the condition is visible, a photo provides evidence for a condition grade.

In every case the grades are your grades, learned from your own inspectors' past decisions. Test on your own photos before relying on it; a pilot shows how well it works on your assets.

A technician's note can go with each photo, and the model learns from both.

The last row is worth a model of its own: rejecting unusable photos at the point of capture keeps the trend clean.

A grade per photo, with the asset ID and date, turns into a trend:

• Current grade: how bad it is now.
• Grades per month: how fast it's changing. This catches the asset that's only a 3 but got there in six weeks.
• Sudden jumps: two grades or more between visits. Check the photo first; a jump is often the wrong asset or a bad angle.

The trend is a simple query on your asset data, in your CMMS, EAM, data warehouse or a spreadsheet. An AI agent can ask StayCharted for each grade and run the query too.

Grades are ordered categories, not equal physical measurements. A move from 2 to 3 need not represent the same amount of deterioration as 3 to 4. Use grade changes to prioritize inspection, not to estimate material loss or a failure date.

StayCharted doesn't replace your EAM, CMMS or asset performance software, such as IBM Maximo, SAP, AVEVA, MaintainX, Fiix or UpKeep. Keep your existing asset records, history, work orders and sensor monitoring in those systems. StayCharted adds a model trained on how your inspectors grade a photo.

A typical flow:

1. A technician photographs a conveyor belt on the round.
2. StayCharted answers "Worn", with a confidence.
3. Your maintenance system stores the grade against the asset, beside the photo.
4. If the confidence is low, an inspector checks it in the Review Queue on Essentials or Business.
5. The inspector, not the model, decides what work is needed.

How it connects: through StayCharted's API, Power Automate, or an AI agent, or with a batch of photos. There are no built-in connectors to these systems; connecting one is an integration your team or your implementation partner builds on the API.

A general AI can see rust in a photo. Without your examples and guidance, it does not automatically know:

• your grading scale, and where your inspectors draw the line between a 3 and a 4;
• your assets, coatings and environment, and what "normal" looks like on them;
• which conditions your team acts on, and which it watches.

Your inspection history already holds those calls, photo by photo. A model trained on them applies a published model consistently, and you can measure how often it agrees with your inspectors before it goes live.

1. Pick one asset type and one scale

One asset type photographed often, with grades your inspectors apply consistently.

2. Gather graded photos

For an initial pilot, collect 20 to 50 photos per grade from your inspection app or CMMS, more for grades that look alike and for the severe ones, which may be the rarest. If past photos aren't graded, have two inspectors grade a few hundred, and settle the rule where they disagree.

3. Upload, check and train

• Upload a ZIP with a folder per grade, or a spreadsheet of photo links with the grade, the technician's note and the asset ID.
• Prepare separate training and test files, keeping every photo of an asset in one set. Test on assets the model never trained on; use the asset ID for grouping and reporting, not as a model input.
• StayCharted checks the photos first: photos that look like another grade, duplicates, near-identical shots, files it couldn't read.
• Train the AI Image Classifier Model first, in about a minute. Try the Dedicated AI Image Model if the differences between grades are small details.

4. Read the report by grade

The report shows how often the model matched your inspectors on photos it never saw, grade by grade, and how many photos would go to an inspector at each confidence cutoff. Read the severe grades first.

5. Connect it to the photos as they arrive

• API: your inspection app or data pipeline sends each new photo and writes the grade and confidence back against the asset.
• AI agents: Claude and ChatGPT through StayCharted's connector, or any agent that can call an API.
• Power Automate for a flow, or a batch of the week's photos as a ZIP or spreadsheet.

6. Review, then retrain

On Essentials and Business, grades below your cutoff wait in the Review Queue. Inspectors' corrections join the next training run. Retrain when something changes: a new camera, a new site, a new coating.

We haven't published a benchmark on industrial assets yet. The closest public results, graded the same way:

• Car photos, damaged or not: 94.0% right. At the default review cutoff, 77% were above the review cutoff, with 98.4% damaged-or-not accuracy in that subset. The benchmark →
• Fruit quality, good against bad: 98.7% right. The benchmark →
• The lesson from both: rare categories can be missed when training examples are scarce. Collect your worst grades on purpose.

Run it on your own photos. Talk to us about a pilot: we'll measure accuracy, how many photos go straight through, and how many need an inspector, on photos of your assets that the model never trained on.

• Maintenance managers and reliability engineers whose technicians photograph assets on every round.
• Inspection managers and heads of asset integrity who need gradings that are consistent across inspectors and sites.
• Industrial inspection service providers who grade photos for many clients, and want every client's grading scale applied the same way each time.
• Asset-management implementation partners who add photo grading to the EAM or CMMS projects they deliver. Partner with us →

It works best where you already have repeat photos of the same assets, a grading scale your inspectors apply consistently, and past inspector decisions to learn from.

• People decide. It grades photos. Repair, replace or shut down stays with your engineers.
• A record of how each model was tested, on which photos and with what result.
• Changes recorded in the workspace's Activity, including changes made through an AI assistant.
• Private to your workspace. Your photos train only your models. Location data and other hidden details are removed from each picture on upload.
• Need it in your own AWS account? Talk to us about a self-hosted Enterprise deployment.

• It doesn't measure. It gives a grade, not a crack length or a corroded area. Bands work as grades.
• It doesn't keep the asset timeline. Your CMMS, EAM or data warehouse does; the model grades each photo.
• It doesn't find where the defect is in the photo. It grades the whole picture.
• It doesn't replace sensors. It covers what the camera can see, including assets that have no sensors.
• It doesn't connect to a CMMS directly. It works through the API, Power Automate, an AI agent or files, so any system that can send a photo link can use it.