Prepare Signal History for Predictive Maintenance

Predictive maintenance depends on stable equipment identity and clean signal history. Use this recipe to prepare telemetry, events, and maintenance records with DFS before they are used by a predictive maintenance workflow or FactVerse AI Agent predictive maintenance module.

Outcome

At the end of this recipe, you should have:

• one or more connected sources;
• reviewed mappings from source signals to equipment or asset targets;
• sync history showing recent ingestion;
• quality checks for missing, stale, or impossible values;
• a DFS Pro dataset for reusable signal history;
• a steward and validation state for downstream use.

Prerequisites

Prepare:

Requirement	Notes
Equipment list	Stable equipment IDs, names, aliases, and location context.
Signal list	Sensor tags, meter points, event fields, or historian columns.
Source details	Historian, database, CSV export, MQTT topic, REST API, or other source.
Maintenance records	Work orders, inspections, failures, part changes, or operator notes when available.
Review owner	Someone who understands equipment, units, and expected ranges.

Step 1: Define the equipment boundary

Start with the equipment or asset group.

Record:

• equipment ID;
• equipment name;
• aliases used in source systems;
• site, line, area, or system;
• related components;
• expected operating state;
• maintenance owner.

Stable equipment identity is more important than the first model output. If source systems use different names, create a mapping table before using the data downstream.

Step 2: Connect signal sources

Use DFS Lite connectors to connect the source data.

Common source choices:

Source	Typical use
JDBC	Historian tables, data warehouse tables, maintenance databases.
CSV	Exported history, initial data loading, offline model preparation.
MQTT	Streaming equipment or IoT payloads.
REST	Enterprise APIs, asset records, work-order systems.
OPC UA	Equipment or automation telemetry.

Create and test each connector before mapping.

Step 3: Browse and preview source data

For every source:

1. Browse source hierarchy or fields.
2. Preview recent values.
3. Confirm timestamp field.
4. Confirm equipment identifier.
5. Confirm metric or event field.
6. Confirm unit.
7. Check expected value range.

Reject fields that are stale, unexplained, duplicated, or missing required identity.

Step 4: Map signals to targets

Create mappings for the signals that matter.

For each signal, map:

• source path;
• target equipment or asset ID;
• target field;
• unit;
• transform expression when units differ;
• expected range;
• topology or component tag when useful.

Examples:

Source signal	Target use
motor vibration RMS	equipment health feature
bearing temperature	thermal trend feature
current draw	load and operating context
alarm code	event context
work order close date	maintenance outcome context

Review AI-assisted mappings before applying them.

Step 5: Sync and inspect quality

Run sync and inspect:

• read count;
• written count;
• failed count;
• last sync time;
• completeness;
• timeliness;
• accuracy;
• quality trend.

For predictive maintenance, pay special attention to:

• missing intervals;
• timestamp drift;
• unit mismatch;
• source resets;
• flatlined sensors;
• impossible values;
• equipment ID mismatch.

Step 6: Create a DFS Pro dataset

Create a DFS Pro dataset when the signal history needs to be reused.

Recommended dataset fields:

• equipment ID;
• timestamp;
• metric name;
• metric value;
• unit;
• source system;
• quality flag;
• event or work-order reference when available.

Use preview and profile to inspect:

• row count;
• null ratio;
• distinct equipment IDs;
• time range;
• value distribution;
• required columns.

Step 7: Validate and steward the dataset

Assign a steward before using the dataset in production workflows.

Validation checklist:

• equipment IDs match the asset model;
• signal names are understandable;
• units are normalized;
• missing intervals are documented;
• quality issues are reviewed;
• source timestamps are clear;
• maintenance records are joined or linked when needed;
• downstream workflow owner accepts the limitations.

Step 8: Use the dataset downstream

After validation, the dataset can support:

• predictive maintenance analysis;
• anomaly review;
• health summary workflows;
• AI Agent evidence retrieval;
• maintenance planning;
• BI reports;
• model retraining readiness review.

Keep a record of source limitations in the workflow run record or dataset description so reviewers understand what the model can and cannot infer from the data.

Related pages

Page	Use
DFS Lite Connectors	Connect historian, database, API, file, or telemetry sources.
Mapping Source Fields	Map source signals to assets and fields.
Data Quality	Check missing, stale, or invalid values.
DFS Pro Datasets	Create a governed dataset for signal history.
FactVerse AI Agent Predictive Maintenance Workflow	Use prepared data in an AI Agent workflow.