A factory manager does not need another screen saying that everything is green. The useful warning is the one that arrives early enough, reaches the right person and helps avoid a problem that matters.

That is a better starting point for predictive maintenance than asking how many machines can be connected in the first month. For a small or medium-sized manufacturer, one well-chosen pilot can reveal whether the approach fits the business.

Understand What You Are Trying to Predict

NIST distinguishes predictive maintenance, which uses observed information such as temperature, noise or vibration to anticipate failure, from scheduled preventive maintenance and repair after a breakdown. Its manufacturing research also emphasises the economic consequences of maintenance decisions. NIST: Manufacturing Machinery Maintenance

The practical framework below is TechFullNews editorial guidance, not a claim that a particular sensor or model will deliver a guaranteed saving.

Start by writing down the failure you want to address. “Improve maintenance” is too broad. “Give the maintenance team useful warning of a recurring bearing problem on this production-critical machine” is something a team can investigate.

Choose a Machine That Creates a Real Business Problem

Look for an asset with a documented failure history, meaningful downtime consequences and a team that can act on an alert. Avoid choosing a machine solely because it is new or already has an attractive monitoring screen.

A hypothetical workshop might select the one machine that regularly holds up final dispatch. Another might focus on equipment whose faults create expensive rework. Neither should assume that the most expensive machine is automatically the best pilot.

Before buying anything, review the maintenance log with the operator and technician. Ask what happened before previous failures and whether that information was recorded. Their experience helps define what the monitoring project needs to learn.

Build a Baseline Before Judging the Alerts

Record the operating conditions that make a reading meaningful. A measurement taken under a heavy load may not be comparable with one taken while the machine is idle. Note product changes, maintenance work and sensor adjustments alongside the readings.

Agree on a short list of outcomes before the pilot begins:

  • Alerts that led to a useful inspection or repair.
  • Alerts that created work but revealed no relevant issue.
  • Failures that occurred without useful warning.
  • Time spent maintaining the monitoring system itself.

This prevents the project from being judged only by the number of notifications it produces. Set the pilot duration around the failure pattern and production cycle; a quiet week is not proof of success.

Decide Who Acts on a Warning

Imagine an alert arriving during a night shift. Who sees it? What are they authorised to do? When should a qualified technician inspect the machine, and how is the outcome recorded?

Write that route down before connecting the first sensor. Keep the existing maintenance and safety procedures in place during the evaluation. Monitoring can support a decision; it does not replace the expertise needed to assess equipment safely.

A useful alert record links the warning to the inspection, the finding and the action taken. Over time, that record helps the team understand whether warnings are trustworthy and whether thresholds need review.

Count the Work Behind the Software

The budget should include installation, integration, staff time, recurring fees and ongoing checks on data quality. Compare these with conservative estimates of avoidable losses rather than assuming every future breakdown will disappear.

At the end of the pilot, make a deliberate decision: expand, adjust or stop. A project that reveals poor data or an unworkable response process has still produced useful information. The goal is a better maintenance decision, not a larger dashboard.