Sensor Training Data for Predictive Maintenance AI

It is the process of labeling equipment measurements so models can learn from operating states and meaningful events. Unitlab supports range and point annotations on CSV time-series recordings, with structured properties and review workflows.

You can annotate CSV recordings with numeric channels such as vibration, temperature, pressure, or current. The X-axis can contain numeric values or supported timezone-aware timestamps.

Yes. Your ontology defines the operating states and event classes. Use ranges for meaningful periods and points for individual samples, retaining the surrounding signal context.

Yes. Inspect channels in separate charts with a shared X-axis window, or combine selected channels in one view. Labels may describe one channel or the recording as a whole.

Use a shared ontology for event classes, annotation properties, recording properties, and relationships. Structured choices and required-property checks help teams follow the same labeling scheme.

Items can move through annotation, review, and rework stages. Reviewers can inspect the chart and anchored comments, while approved benchmarks and consensus support quality checks where configured.

Yes. Use dataset search, metadata, tags, and filters to select relevant assets. Dataset versions help keep prepared data and reviewed releases traceable as the project evolves.

JSONL exports preserve labeled ranges and points, original X-axis values, channel context, and supported properties. Datetime axes use UTC epoch milliseconds; numeric axes retain their source unit.

The reviewed labels become training or evaluation targets in your machine-learning pipeline. Unitlab prepares the annotated data; equipment prediction and deployment remain part of your downstream model workflow.