Context
Two vehicles of the same model, running the same part, wear it out at different rates. Routes, loads, duty hours, climate: these accumulate, and mileage on its own says little about the state a given part is actually in.
Replacement intervals, meanwhile, are set per model. The interval has to hold for the harshest duty in the fleet, which means most vehicles give up their parts with life still in them.
Widen the interval and roadside failures follow. When a truck stops away from the depot, recovery and the rescheduling around it usually cost more than the repair itself.
Across a large fleet these calls add up. Knowing where each vehicle stands, rather than where the model average stands, changes the decision.
What we tested
The input is the sensor readings the vehicle already logs. No additional instrumentation is required.
The model estimates the remaining life of the component. It is built on a class of lightweight machine-learning techniques (reservoir computing) suited to time-series signals.
Validation used public data covering a large number of vehicles, with operating records and component replacement history. Vehicles held out from training were used for evaluation, with a small amount of data used to fit the model to each one.
What it shows
Remaining life could be estimated per vehicle on units the model had not seen during training. Compared with applying one fleet-wide interval, the estimate reflects how each vehicle has actually been worked.
This may apply to your fleet
If you run a large number of identical vehicles or machines and keep records of the parts you have replaced, the same approach may transfer.
It can support levelling out replacement work, sizing parts inventory to real need, and identifying the units working hardest before they become failures.
A short conversation is usually enough to tell whether your operating data is in a usable form.