All learning series · Suite application · Part 1
Maximo Visual Inspection
Follow images from labeling to evaluation and deployed inference.
Before you start
MAS learning path. Match the component documentation to your installed release and maintenance build before configuring it.
- Visual Inspection available with suitable training resources.
- A permitted synthetic or training image collection with clear labeling rules.
IBM sources + Assetize exercise · Not lab-verified
01 · Basics — Understand the building blocks
Visual Inspection provides a workflow for creating and training models to inspect assets and products. Training a model and deploying it for use are separate steps.
A useful dataset includes realistic variation in lighting, angles and defects. Near-identical images across training and evaluation can inflate apparent performance.
Topics in this series
- Inspection objective
- Images and labels
- Model training
- Evaluation
- Deployment and inference
02 · Configuration — Work through the setup
Use a training environment and match the actions to the IBM guide for your installed build.
- Define one inspection question and label categories consistently.
- Create an image dataset and review annotations before training.
- Hold out an independent evaluation set, separating near-duplicates from the training set.
- Train using a supported model workflow and inspect mistakes on the held-out images.
- Deploy a selected model and send a new image for inference; retain model identity and the observed response.
03 · Practical example
Fictional teaching example
A teaching dataset distinguishes visibly damaged from intact housings. A model that recognizes the background instead of the defect can appear accurate until the camera location changes.
04 · Practice and verify
Compare inference on a normal image, a defect image and an unfamiliar lighting condition.
Expected result
The deployed model identity is known, and incorrect or uncertain detections are visible in the evaluation record.
Common mistakes
- Reusing training images as proof of generalization.
- Equating a confidence score with guaranteed correctness.
05 · Advanced concepts
Explore confidence thresholds, human review and model replacement. Treat Edge deployment as a separate device/runtime validation when applicable.
IBM references
Product explanations link to IBM sources. Scenarios and exercises are original Assetize guidance, not claims of executed lab procedures. Reviewed 2026-10-01.
Maximo ecosystem overview · Manage learning paths · Practice data in Workbench