A predictive maintenance pilot project is the single most important decision point in any PdM programme — not because the technology is unproven, but because the difference between a successful pilot that scales to the enterprise and a pilot that stalls out at eight assets is almost never about the prediction algorithm. It is about the structure applied before the first sensor is mounted. The plants that go from pilot to plant-wide deployment in 12 weeks follow a repeatable launch framework: disciplined asset selection, deliberate sensor procurement and network provisioning, structured data validation gates, model training with production-quality labelled data, alert configuration tuned to the maintenance team's response capacity, and hard success criteria agreed before the first prediction reaches a decision-maker's screen. Plants that skip these steps spend 12 months trying to scale a pilot that was structurally designed to never leave the test phase. This is the 50-point pilot project launch checklist for industrial predictive maintenance — covering every phase from asset prioritisation through to formal go/no-go decision, designed for reliability teams using iFactory's AI platform, which fuses vibration sensor telemetry, spindle temperature data, motor current signatures, and Shift Logbook inspection records into machine learning models that forecast bearing failures, tool wear progression, and spindle degradation 2–3 weeks in advance. Book a Demo to see how iFactory structures pilot programmes for rapid scale.
Phase 1: Asset Selection and Prioritisation (Points 1–10)
The pilot's asset scope determines everything that follows — sensor procurement quantities, network capacity requirements, data pipeline volume, model training bandwidth, and maintenance team workload. Selecting the wrong assets — too many, too diverse, too critical to allow any prediction false positive — is the single most common cause of pilot failure. The first ten checklist points ensure the asset scope is optimised for learning velocity, not production coverage.
- Select 8–15 assets representing 2–3 failure modes (bearing, imbalance, misalignment)
- Include assets with known failure history — labelled data accelerates model training
- Exclude assets with planned retirement within 12 months
- Include at least one asset class with existing vibration sensor coverage
- Document baseline operating parameters for each asset — speed, load range, duty cycle
- Score each candidate asset on criticality, data availability, and intervention feasibility
- Confirm maintenance team has authority to intervene on selected assets
- Verify CMMS records are accessible and structured for work order correlation
- Document spare parts availability for selected asset classes
- Obtain production team sign-off that pilot assets can be taken offline for sensor installation
Phase 2: Sensor Procurement and Installation Planning (Points 11–22)
Sensor procurement lead times, mounting hardware compatibility, and network provisioning are the operational bottlenecks that determine whether a pilot starts on schedule or drifts into month three before a single data point is collected. The twelve points in this phase ensure sensor selection, mounting method verification, and network infrastructure are validated before installation begins. Book a Demo to see how iFactory provides pre-configured sensor bundles with validated mounting templates for common industrial asset classes.
| Checklist Area | Points | Key Activities | Validation Gate |
|---|---|---|---|
| Sensor Selection | 11–14 | Select vibration, temperature, and current sensor types per asset failure mode; confirm frequency range and mounting compatibility; verify environmental rating (IP67 for washdown zones) | Sensor spec sheet matched to each asset |
| Mounting Hardware | 15–17 | Order stud mount bases, adhesive pads, and magnetic bases per ISO 5348 guidelines; document mounting method and orientation per asset | Surface prep verified on pilot assets |
| Network & Gateway | 18–20 | Provision wireless gateway or edge device; confirm RSSI coverage at each sensor location; allocate IP addresses and data pipeline bandwidth | RSSI ≥ −90 dBm at all locations |
| Power & Cabling | 21–22 | Verify power availability at gateway location; order cable trays, conduit, and strain relief for wired sensors; label all cables per asset ID | Cable routing plan approved |
Phase 3: Installation, Commissioning, and Data Validation (Points 23–34)
The sensor installation phase is where most pilot projects accumulate schedule risk — surface preparation delays, mounting hardware incompatibility, network connectivity issues, and data pipeline configuration problems surface sequentially rather than simultaneously when installation is not pre-planned. These twelve points cover surface preparation per ISO 5348, sensor orientation per ISO 20816-3, communication verification, and baseline data capture for every pilot asset.
Phase 4: Model Training, Validation, and Alert Configuration (Points 35–44)
Model training is where the pilot transitions from data collection to prediction generation — and where the gap between vendor demo performance and production performance becomes visible. Production-grade models require labelled training data, asset-specific baseline calibration, and alert configuration tuned to the actual maintenance team's intervention capacity. Skipping any of these steps produces a pilot with high false positive rates that erodes operator trust before the model has had time to converge.
Phase 5: Success Criteria, Go/No-Go, and Scale Planning (Points 45–50)
The final six checklist points define the pilot's success criteria, the formal go/no-go decision process, and the scale roadmap. Defining success criteria before the pilot runs — rather than retrofitting them after results are visible — is the difference between a defensible scale-up decision and a subjective conversation that stalls the programme for quarters. Book a Demo to see how iFactory's pilot-to-scale transition framework uses automated model performance metrics to support objective go/no-go decisions.
- Model precision ≥ 70% on pilot assets after 90 days of shadow mode data
- False positive rate < 25% for first 90 days; trending downward each month
- At least one verified early detection — prediction confirmed by CMMS inspection finding
- Maintenance team confidence survey score ≥ 3.5 out of 5
- Prediction-to-work-order pipeline operational with > 80% work order acceptance
- Document pilot results in standard format: precision, recall, false positive rate, detections, cost avoidance
- Present formal go/no-go recommendation to steering committee with data from all six criteria
- If go: define scale phase scope (30–80 assets, 3–5 asset classes, 12-week deployment timeline)
- If no-go: document root cause gap and define remediation pilot (8-week maximum before re-decision)
- Archive all pilot data, model configurations, and lessons learned for future reference
The 50-Point Checklist Summary
| Phase | Points | Duration | Deliverable |
|---|---|---|---|
| 1. Asset Selection & Prioritisation | 1–10 | Wk 1 | Pilot asset register with failure mode mapping and criticality scores |
| 2. Sensor Procurement & Planning | 11–22 | Wk 1–2 | Sensor BOM, mounting plan, network topology, and cable routing approved |
| 3. Installation & Commissioning | 23–34 | Wk 3–4 | Sensors mounted, comms verified, baseline data capture initiated |
| 4. Model Training & Validation | 35–44 | Wk 5–10 | Baseline calibrated, alerts configured, shadow mode running with metrics |
| 5. Go/No-Go Decision & Scale | 45–50 | Wk 11–12 | Pilot results documented, go/no-go decision with scale roadmap |
Why Most PdM Pilots Stall — and How the Checklist Prevents Each Failure Mode
The most expensive lessons in predictive maintenance pilot programmes have already been learned by the plants that ran them. The failure modes are well documented across industrial PdM deployments, and each one is preventable when the pilot launch checklist is executed in sequence. The table below maps the four most common pilot failure modes to the specific checklist points that prevent them.
Expert Perspective
I have been involved in more than 40 predictive maintenance pilot evaluations across steel, automotive, chemical, and food processing facilities over the past 12 years. The pilots that scaled to enterprise deployment shared one characteristic that the pilots that stalled did not: a structured launch checklist with hard gates. Not a project plan with optimistic dates — a checklist where each phase had a formal validation gate that had to be passed before the next phase could start. The plants that scaled did not skip sensor procurement validation. They did not start model training before baseline data quality was confirmed. They did not configure alert thresholds without understanding the maintenance team's actual intervention capacity. And they defined the go/no-go criteria — precision target, false positive rate, detection count — in writing before the first sensor was mounted. The pilots that stalled skipped the checklist because it felt like overhead. It was not overhead. It was the structure that made scale possible. The 50-point checklist is not bureaucratic. Every point exists because a pilot somewhere failed by skipping it.
Frequently Asked Questions
8–15 assets across 2–3 failure modes is the proven range for a first-phase pilot. Fewer than 8 assets provides insufficient data diversity for model training. More than 15 assets stretches the installation team, dilutes sensor procurement attention, and delays baseline data collection. The asset set should include at least one asset with known failure history (labelled data) and at least one asset class with existing vibration sensor coverage to accelerate model convergence.
8–12 weeks is the production-grade benchmark. Week 1 covers asset selection and sensor procurement. Weeks 2–4 cover sensor installation, commissioning, and baseline data capture initiation. Weeks 5–10 cover model training, baseline calibration, alert configuration, and shadow mode validation with at least 4 weeks of prediction data logged. Weeks 11–12 cover success criteria review and formal go/no-go decision. iFactory's pre-configured pilot templates reduce the timeline to 8 weeks for facilities with existing sensor coverage on selected assets.
Four criteria should be defined and agreed in writing before the first sensor is installed: (1) model precision target — minimum 70% after 90 days of shadow mode data; (2) false positive rate — below 25% for first 90 days with a declining trend; (3) verified early detection — at least one prediction confirmed by CMMS inspection finding during the pilot period; (4) maintenance team confidence — survey score of at least 3.5 out of 5 indicating willingness to act on predictions. These four criteria provide objective data for the go/no-go decision without requiring subjective interpretation of ambiguous results.
iFactory's platform is designed for pilot-to-scale deployment. The same model training pipeline, alert configuration framework, and Shift Logbook integration that runs on 10 pilot assets scales to 1,000+ assets without re-architecture. Sensor templates, baseline calibration parameters, and alert configuration profiles created during the pilot are reusable across additional assets of the same class. The pilot data — including model performance metrics, false positive classifications, and labelled training events — feeds directly into production model retraining cycles. iFactory provides a standard pilot results document format covering precision, recall, false positive rate, detection events, and cost avoidance that supports the formal go/no-go decision process.
