The question usually arrives as either/or: keep the preventive schedule, or move to predictive maintenance. In a heavy fleet it is never either/or. A schedule handles the wear you can count in hours. Reading the signs handles the failures that do not follow the hour meter.

The useful question is which approach each component deserves. This guide gives you plain definitions, the inputs predictive work actually needs, a decision rule per component, and a worked example on one final drive.

Four strategies, defined plainly

  • Run to failure. Fix it when it breaks. Right for cheap, non-critical parts with a spare on the service truck: a work light, a mirror, a wiper.
  • Preventive. Service at fixed hours or dates whether the part needs it or not. Right for fluids, filters, grease and belts, where wear tracks hours closely and the parts are cheap.
  • Condition-based. Inspect or measure, and act when a limit is crossed. Right for undercarriage, tires, bucket teeth and air filters with a restriction indicator.
  • Predictive. Trend data over time to see a failure coming before any limit is crossed. Right for expensive components that give early warning: engines, transmissions, final drives, hydraulic pumps, turbochargers and aftertreatment.

A healthy shop runs all four at once, on different parts of the same machine.

What predictive needs in practice

Predictive maintenance sounds like new sensors and software. In many heavy fleets the first inputs already exist and are simply not collected in one place:

  • Fault codes with their occurrence counts, read from the engine controller. Our guide to J1939 SPN and FMI codes explains how to read them.
  • Engine data: coolant temperature, oil pressure, boost pressure, exhaust temperature and fuel rate.
  • Oil and coolant analysis results, trended by machine and compartment.
  • DPF regeneration frequency on Tier 4 machines.
  • Inspection notes and photos from the daily walkaround.
  • Hours and duty: idle share, load, breaker hours.
  • Repair history: what failed, at what hours, and what the machine reported in the weeks before.

The last item is often the one missing. Without work orders that record what was actually found, nobody can tell which signals came before which failures, and every threshold stays a guess.

Where each one pays

Two questions sort most components: how much does a failure cost, and how much warning does it give?

Reliability engineers call the warning window the P-F interval: the time between the point where a failure becomes detectable (P) and the point where it fails (F). A long P-F interval with a measurable signal is where predictive pays. A short or invisible one is where it does not.

Component Usual strategy Signal to watch
Engine oil and filters Preventive, checked by oil analysis Hours
Air filter Condition-based Restriction indicator
Undercarriage Condition-based Wear measurements against hours
Final drive Predictive Iron trend in oil samples, temperature, seal leaks
Hydraulic pump Predictive Pump case drain flow, oil analysis, slow cycle times
Turbocharger Predictive Boost codes, exhaust temperature, oil consumption
DPF Condition and predictive Regeneration frequency, differential pressure codes
Work lights, mirrors Run to failure The walkaround

A decision rule per component

Ask four questions about each major component, in this order.

  1. What does a failure cost, including downtime, rental of a replacement and any damage to nearby parts?
  2. Does it give warning in data you can collect, and how many hours or days of warning?
  3. Is the wear tied closely to hours?
  4. Can you act inside the warning window, given parts lead time and mechanic availability?

Then apply the rule:

  • If the wear tracks hours and the parts are cheap, use preventive.
  • If the warning window is longer than the time you need to act, use predictive or condition-based.
  • If there is no useful warning and a failure is cheap, run to failure.
  • If there is no useful warning and a failure is expensive, stock the spare and plan the swap, because no amount of data will save you.

The fourth question is where many plans fall down. An alert two days before a failure is useless on a part with a three-week lead time, unless the part is already on your shelf.

Example: one final drive

Round numbers, for illustration only. Use your own.

Say a final drive on a 30-ton excavator costs $18,000 to replace after a failure, and the machine is down six working days. A replacement rental and the delay to the crew cost $1,200 a day. The unplanned failure costs 18,000 + (6 × 1,200) = $25,200.

Now say oil sampling catches the bearing wear early, and a planned seal and bearing repair costs $6,000 with the machine down one day it was scheduled to be in the yard anyway. The difference is $19,200.

Sampling both final drives every 500 hours on a machine that runs 2,000 hours a year is eight samples. At $30 a sample, that is $240 a year. One caught failure pays for 80 years of sampling on that machine. The program does not need to catch every failure. It needs to catch one, now and then, with someone acting on the result.

A single reading rarely means much. A change against the same machine’s own history usually does. These trends are cheap to collect and point at real problems:

  • Coolant temperature creeping up at the same load and similar weather: a cooling stack loading with dirt, a slipping fan drive or a tired water pump.
  • Boost pressure dropping for the same work: an intake leak, a charge air cooler leak or a turbo on its way out.
  • DPF regenerations getting closer together: more soot from injectors, intake restriction or a duty change toward idle.
  • Fuel burned per engine hour rising with no change in the work: drag somewhere, from dragging brakes to a hydraulic system working too hard.
  • Oil top-ups getting more frequent, logged at the daily inspection: rings, turbo seals or a leak nobody has found.
  • Hydraulic cycle times slowing. Time a full boom raise at full throttle with warm oil once a month and write it down. A pump losing efficiency often shows up here early.

None of these needs a new sensor. Most need someone to write the number down in the same place every time.

How to start without new sensors

  1. Keep the preventive schedule. Predictive work is added on top, never swapped in. Intervals and limits come from the manufacturer’s operation and maintenance manual; confirm them there before changing any.
  2. Pick three failure modes that cost you most over the last two years, from your work orders and invoices.
  3. Write the signal and threshold for each. For example: coolant temperature trending above that machine’s own summer baseline, or iron rising across two consecutive final drive samples.
  4. Route each alert to one named person with a defined action and a deadline.
  5. Record what was found when someone acted: confirmed, early sign, or false alarm.
  6. Review monthly. Tighten thresholds that miss, loosen ones that cry wolf.

Three failure modes done well beat thirty done badly. Add the next one when the first three are boring.

Common mistakes

  • Dropping the PM schedule to go predictive. Filters and fluids still need changing on hours.
  • Buying sensors before using the data you already have. Engine data, fault codes and oil analysis come first.
  • Sending every alert to everyone. People stop reading. Route by severity and machine.
  • No closed loop. If nobody records whether an alert was right, the system never gets better.
  • Treating a health score as a diagnosis. A score tells you where to look first, not what is broken.
  • Ignoring lead time. Warning without parts on hand only gives you longer to worry.
  • Baselines from the wrong machine. Compare each machine with its own history first, then with the same model on similar duty.

Questions to ask about any tool

Before you buy predictive software or a health score, ask the vendor, including us:

  • Which inputs does the score use, and can I see why a machine scored low?
  • Does it work across every brand in my fleet, or only machines with one manufacturer’s telematics?
  • Can I set who gets each alert, by severity, machine and job?
  • Does it learn from what my mechanics record on the work order?
  • Can I export the raw data and the scores?

Where Gytix fits

Asset health monitoring collects fault codes, engine data and inspection notes from every machine into one record and gives each one an asset health score, so the shop sees which machines to look at first and why. Preventive maintenance runs the hour-based schedule underneath, with duty factors and work orders that open 50 hours ahead. Both are part of the Shop plan.

If you are comparing systems, our guide to choosing fleet management software lists the questions to ask about data, coverage and contracts before you sign anything.