Checklist

Predictive Maintenance: Data Requirements & Labeling Checklist (Pilot Planner)

Interactive checklist and pilot planner to ensure PM pilots collect sufficient, consistently labeled, and high-quality data for model training, validation, and evaluation. Includes a short labeling-convention template, a minimal dataset schema example, and structured fields to record readiness, metrics, and next steps.

Interactive Tool

Predictive Maintenance Data & Labeling Checklist

Use this interactive checklist to scope predictive maintenance pilots, capture labeling conventions, and record the minimum dataset, data-quality checks, and pilot evaluation metrics needed to produce repeatable, defensible results. Complete the fields as you plan the pilot; saved responses can be used to compare multiple pilots or handed to data engineers and labeling teams.

Why this matters

Clear failure definitions, balanced labeled examples, synchronized timestamps, and an explicit minimal schema are the difference between a useful pilot and one that can't be evaluated or scaled. This planner helps teams avoid common pitfalls—insufficient negatives, label ambiguity, unlabeled baseline modes, and undocumented sampling rates.

Quick Labeling Convention Template

Labeling document should include:
- Failure class name(s) and short description(s)
- Precise start and end criteria (timestamp rules)
- Label granularity (event-level, cycle-level, per-sample)
- Negative-class rules (what counts as non-failure)
- Edge-case rules and examples
- Labeler qualification and review process
- Inter-annotator agreement target and adjudication method

Sample minimal dataset schema (example)

timestamp (ISO 8601)	machine_id	sensor_id	sensor_type	value	unit	operating_mode	label	labeler_id	label_timestamp

Adapt names to your environment but keep a consistent schema across sensors and systems.

Short, unique name for this pilot (used for exports and comparisons).
Name and role or email of the person responsible for the pilot.
Describe what constitutes a labeled failure. Include precise start/end timestamp rules and observable signals.
Do you have concrete labeled examples (files, annotated traces, or log excerpts) that show the failure?
Best current estimate. If rare, plan for more collection or synthetic approaches.
Important for class balance and realistic evaluation.
List sensors, channel names, sampling rates, units, and the system that provides them.
Calibration records or known offsets improve data reliability and drift analysis.
Specify Hz or sample interval for each channel. Note any downsampling or aggregation rules.
Timestamp drift or timezone mismatches are a common showstopper.
How long raw and processed data will be kept and who can access it. Include privacy or regulatory constraints.
Annotating normal modes reduces false positives and helps context-aware models.
Flags for operating context that affect sensor behavior.
Indicates whether you have a written labeling guideline for labelers.
Set an acceptance threshold (e.g., 80–90%). Specify an adjudication method if below target.
Above this threshold consider sensor maintenance or additional collection.
Select the checks you plan to run during ingestion and validation.
Note any constraints (e.g., PII, GDPR, safety reporting) that affect data sharing or labeling.
Select metrics you will use to judge pilot success.
Specify the portion of labeled data reserved for final evaluation.
Outline how models will be trained, validated, and tested (cross-validation, time-split, backtesting).
Where data will be stored and which teams can access it (roles/permissions).
Capture action items, owners, and deadlines for pilot readiness.
You can explore this tool now. Sign in or create an account to save your responses and return to them later.
Sign In to Save Create an Account
Make this tool part of your work

Save a personal copy, bring it to your team, or tailor the questions and workflow to fit what you are hungry to improve.

Member customization and team collaboration are coming soon.
Discuss Ask the Librarian

Discussion

Comments and conversation will live here.