Predictive maintenance uses machine data to detect developing faults before they cause breakdowns. What can be monitored on moulding machines, how to start practically, and where AI adds value beyond simple thresholds.
A hydraulic moulding machine stopped mid-shift with a failed pump, taking a high-volume tool out of production for two days. Afterwards, the maintenance team reviewed the machine's data. Over the previous three weeks, oil temperature had crept up and the pump had taken slightly longer to build pressure each cycle. The warning signs had been there; nobody was looking.
Predictive maintenance aims to catch those signs and act before failure.
From reactive to predictive
- Reactive maintenance: fix it when it breaks. Unpredictable downtime and often higher repair cost.
- Preventive maintenance: service at fixed intervals. Better, but may replace parts too early or miss failures that happen between services.
- Condition-based and predictive maintenance: monitor equipment condition and intervene when data shows deterioration.
What can be monitored on moulding machines
- Hydraulic systems: oil temperature, pressure, pump performance, filter condition and oil quality.
- Electric drives and motors: current, temperature and vibration.
- Screw and barrel: screw recovery time, torque and pressure trends, indicating wear.
- Clamp and toggle: lubrication, alignment and clamp force consistency.
- Heaters and thermocouples: power consumption and temperature stability.
- Cooling: water flow and temperature differences.
- Process signals: injection pressure, fill time and cushion trends, which can reveal machine or tool problems.
Starting practically
- Choose critical machines: the bottleneck, high-value tools and machines with breakdown history.
- Use data already available from machine controllers before adding sensors.
- Add simple sensors where needed: vibration, temperature, oil condition.
- Set alert thresholds based on normal operating ranges and manufacturer guidance.
- Build a routine: someone reviews alerts and trends and plans interventions.
Where AI adds value
Simple thresholds catch obvious deviations. Machine learning models can:
- Learn the normal behaviour of each machine across products, tools and conditions.
- Detect subtle combinations of changes that indicate developing faults.
- Reduce false alarms by understanding normal variation.
- Estimate remaining useful life of components in some cases.
AI works best when good data, clear failure history and maintenance expertise come together. It is not a substitute for basic maintenance discipline. As with digital OEE systems, data only helps if someone acts on it.
Measuring the benefit
Track unplanned downtime, mean time between failures, maintenance cost and the number of failures predicted and prevented. Compare against the cost of sensors, software and analysis time.
Linking with TPM
Predictive maintenance complements TPM: operators care for equipment daily, maintenance follows planned schedules, and data provides early warning of problems that neither would see. Together they move the plant towards the goal of zero unplanned breakdowns.
