On a nut harvester, a breakdown during the short harvesting window can compromise the entire season. Predictive maintenance does not guess the future: it continuously compares the machine’s real behaviour to its normal behaviour, and flags drift before the breakdown.
Monitoring the right parameters, at the source
It all starts with the raw data read on the CAN bus and the sensors: engine speed and temperature, hydraulic pressures, consumption, command states, location. This is exactly what AMB Rousset does on its harvesters, or ALAMO Group on its electric range. Reading these parameters at the source, rather than through a proprietary gateway, ensures access to the variables that are genuinely predictive of the machine’s health, at the right frequency.
From measurement to anomaly: detecting drift
A single temperature or pressure value says little. Predictive value comes from relating several parameters and their evolution over time. The approach is to build a normal reference behaviour for the machine, then continuously measure the gap between current operation and that reference. Multivariate anomaly detection, combined with statistical thresholds, tells a normal variation in usage apart from a drift that signals an incoming failure. This shifts the logic from a fixed-threshold alarm, often too late, to genuine anticipation.
Closing the loop with the intervention: connected CMMS
Detecting is not enough: you have to act. A detected drift triggers a remote diagnosis, then if needed a maintenance operation planned at the right time, before downtime. Connected to a CMMS, the information feeds directly into the machine’s history and informs future interventions. This is what lets an after-sales team cut travel, arrive with the right part, and capitalise knowledge to better train technicians. Data only delivers its full value when it ends in a maintenance action.
Frequently asked questions
What is the difference between preventive and predictive maintenance?
Preventive maintenance acts at fixed intervals (every X operating hours), whether or not there is real wear. Predictive maintenance relies on the machine’s actual data to act at the right moment, based on its effective condition. It avoids both unplanned failures and premature replacements.
What data is needed to perform predictive maintenance?
The operating parameters read on the CAN bus and the sensors: temperatures, pressures, speeds, consumption, command states, all time-stamped and historised. The closer the reading is to the source and the more generic it is, the wider the range of usable parameters, whatever the machine’s manufacturer.
