The value of a warning is the time it gives you to act. Meta Smart Factory runs predictive maintenance inside its CMMS: a vibration or temperature reading past its limit raises a maintenance notification by itself, and the work order that follows reserves the spare part and, where APS is in use, gets a window in the production schedule.
Talk to us about your maintenancePredictive maintenance software collects condition data such as vibration, temperature, current or cycle counts from machines and raises a maintenance job when a reading crosses a limit or drifts from the machine's own normal, before the failure stops production. It only works on failures that give a measurable warning early enough to act on. In Meta Smart Factory it is part of the CMMS, next to preventive plans, so a warning becomes a work order.
Start with what the PLC already holds: drive currents, cycle times, pressures and temperatures cost far less to collect than new sensors. Add instrumentation only where those signals cannot see the failure.
An alert on a dashboard nobody opens changes nothing, so the maintenance module carries it through to the repair. A threshold breach raises a notification with a priority and a trail from the malfunction to closure. Technicians get the work order and the machine history on mobile devices and MSF touch panels. The order reserves its spare part, and raises a purchase order when stock is short.
With APS, the agreed maintenance window reaches the planner as a scheduling constraint. MTBF and MTTR per machine are built from the work order fields (notification time, repair start and finish), so both come from records the technician already filled in.
A failure is worth monitoring when it degrades progressively and gives a measurable signal early enough to act. Vibration and current read rotating mechanical wear well: bearing damage, gear wear, misalignment and imbalance. Control boards and I/O cards fail close to randomly, and a collision or a setup error has no progressive process behind it. Relief valves and interlocks emit nothing until the event they guard against, so the module runs them on interval-based proof tests.
The first useful thresholds come from engineering, such as bearing defect frequencies or a temperature rise over the machine's own normal at the same load, and they need no failure history. Anomaly detection in the AI platform uses an LSTM autoencoder that learns normal sensor behaviour. It needs enough healthy running to cover normal variation, and every model runs in shadow mode before its alert goes live. Remaining-useful-life estimates need run-to-failure examples that most plants do not have.
Preventive maintenance runs plans triggered by dates, operating hours or production counts, whatever condition the machine is in. Predictive maintenance acts on sensor data such as vibration, temperature, current or energy. The module does both, and either trigger ends in the same work order.
Yes. Preventive plans, work orders, spare parts reservations and machine history all run on records alone. The first months build failure coding and a downtime history, which becomes the baseline any later condition monitoring is judged against. Classifying twelve months of your own unplanned stops then shows which assets would justify an accelerometer.
No, it runs standalone, and operators raise notifications from the machine HMI or a browser. The MES adds notifications created from recorded downtime, and APS adds the maintenance window as a scheduling constraint.
Rarely on its own. A warning only helps if it reliably comes earlier than the supplier lead time for that part, and lead-time variability and minimum order quantities are still there. Turn warnings into planned windows first and revisit stock levels after a year of evidence.
Run a Proof of Concept on your own machines and data. It typically takes 2 to 12 weeks depending on the product.