How to choose a maintenance strategy for each machine

The maintenance strategy that fits a machine depends on how its parts fail and what a failure costs. Run cheap, non-critical items to failure, and service parts that wear with use on run hours, strokes or parts made. Keep calendar PM for work owed by date, and watch critical machines whose failures come at random for a measured change.

Most plants need all five strategies at once, one per failure mode. This guide gives a four-question test for choosing, the survey numbers behind each approach, and a sample plan for CNC mills, presses and molds.

The five maintenance strategies#

Each strategy answers one question: what starts the work?

A calendar PM is a recurrence rule on a work order: a quarterly gearbox inspection comes back every 13 weeks. Usage-based maintenance is the preventive variant that counts use instead of days: hours run, strokes, shots or parts made.

References draw these lines differently. The US Department of Energy’s guide counts a schedule based on machine run time as preventive, and NASA’s reliability guide treats predictive maintenance and condition monitoring as two names for one practice.^[2][3] This article separates them by trigger, because the trigger is what you set up.

How failures behave decides the trigger#

A calendar or hour interval works only when a part fails at a predictable age. Many parts don’t. NASA’s guide cites three studies, of United Airlines aircraft in 1968, Swedish equipment in 1973 and US Navy equipment in 1982, in which random failures made up 77% to 92% of all failures and age-related failures the remaining 8% to 23%.^[3]

Those studies covered aircraft and naval equipment, not your plant, so treat the split as a warning rather than a measurement of your machines. The pattern behind it transfers. NASA notes that simple, single-piece items such as tires, brake pads and compressor blades wear in step with age or stress cycles, while most complex items show no marked wear-out age.^[3]

On a plant floor, that sorts components into two groups:

• Parts that wear with use: cutting tools, die punches, mold vents and ejector pins, slide knives, belts, filters and grease. A count of use predicts their end well, so a usage-based interval fits.
• Complex assemblies that fail at random: spindle bearings, gearboxes, main drive motors and hydraulic pumps. A fixed interval is early for some units and late for others, so measure their condition instead.

Opening a complex machine on a schedule carries its own risk. NASA reports that scheduled overhaul can raise the overall failure rate by adding early-life failures to an otherwise stable system.^[3] A spindle rebuilt every two years because the calendar said so can fail sooner than the one left alone.

What the survey data says#

NIST’s Machinery Maintenance Survey of US manufacturers, with 2016 figures, measures how plants split their maintenance work. On average, 45.7% of maintenance was reactive, 31.8% preventive and 17.3% predictive.^[1] The share barely moved with size: plants with 1 to 19 employees and plants with 100 or more both averaged 43.8% reactive.^[1]

The most reactive quarter of plants, at 81.5% reactive work, had 13.0% of planned production time as downtime. The least reactive quarter, at 14.4% reactive, had 4.0%.^[1]

Read three caveats into those numbers:

• It’s an association. NIST notes that the groups are small and many differences aren’t statistically significant. Downtime due to maintenance issues, 6.7% against 0.9% of planned time, is one of two gaps that were, at the 10% level.^[1]
• Less reactive plants spent more. Maintenance cost was 3.0% of shipments in the least reactive quarter against 1.9% in the most reactive.^[1] Planned work costs money up front, and it buys back running time.
• Predictive adds less than its headlines. Among plants under 50% reactive, the top half in predictive work had 15% less downtime than the bottom half.^[1] That gain is real, and it’s far smaller than the gap between reactive and planned work.

The DOE guide sizes the steps the same way. It estimates that a preventive program saves 12% to 18% in cost over a reactive one, and that a well-run predictive program saves a further 8% to 12% over preventive alone.^[2]

Match each strategy to the machine#

Run every failure mode on a machine through four questions, in order. The first yes sets the trigger.

The DOE guide gives the same sorting as a table, adapted from NASA.^[2]

Choose reactive when a failure is cheap, safe and quick to fix, the machine isn’t a bottleneck, and the spare is in stock. NASA’s examples are non-critical motors under 7.5 HP, comfort cooling and restroom exhaust fans.^[3]

Choose time-based preventive when the work is owed by date whatever the machine did: an inspection a regulation, insurer or warranty sets on the calendar. Building a PM schedule that holds covers the calendar side.

Choose usage-based when a part wears with use and the machine’s hours change from week to week.

Choose condition-based when a failure is costly, comes at no predictable age and gives a measurable warning first.

Choose predictive only when that warning trend is steady enough to plan parts and labor around. You don’t need a forecast on a machine with slack capacity or a standby unit: a limit and a work order cover it.

Here is a sample plan for a plant with mills, a press and molding machines. Replace the intervals with the maker’s or your own history:

The reactive and condition-based rows come from pricing the failure, not from the part’s name. The same arithmetic sorts any pair of failure modes.

Usage-based or calendar PM?#

Choose a usage interval when a part wears only while the machine runs and the machine’s hours change from week to week. Keep the calendar for work that ages whether or not the machine runs, and for any inspection owed on a date, such as the overhead crane in the plan above.

A calendar interval assumes a run rate. For example, greasing every four weeks on a 500-hour interval assumes 125 run hours a week. In a slow month of 320 hours, the grease goes in 180 hours early.

In a busy month of 640 hours, it goes in 140 hours late. A meter on run hours or parts calls for the work at 500 hours every time.

Part counts work the same way. Altium Packaging’s Memphis plant inspects the Slide Knife on a bottling line every 100,000 bottles, because a date-based schedule was always either early or overdue. See the Altium Packaging story.

Usage-based maintenance covers which unit to count, how to set the first interval and how early to warn.

Condition-based or predictive?#

Both start from a measurement against a baseline: what this machine looks like healthy. Condition-based work fires when a reading crosses a limit. Predictive work trends the readings and estimates when the limit will be reached, so the repair lands in a planned stop.

The measurement can come from a route or from a sensor that stays on the machine. On a route, the interval matters. NASA advises cutting the monitoring interval to between a third and a quarter of the previous one once a reading passes its alert limit, so a quarterly vibration route becomes one every three to four weeks.^[3] A sensor that reports continuously removes the interval question, and leaves the limit as the setting that matters.

How much warning you get depends on the failure mode and on how often you measure. A bearing that wears slowly shows a rising trend you can plan around. A CNC tool crash gives no trend at all: it’s a single shock, and the useful response is a work order the moment it happens.

You don’t need a remaining-life forecast to get most of the value. A limit, a baseline and a work order that opens on its own cover most machines. NASA notes that predictive testing doesn’t suit every type of equipment or failure mode, and shouldn’t be the only maintenance a site practices.^[3]

Trigger work orders from the machine#

IoTFlows Maintain opens maintenance work from the same machine data that runs your monitoring. A meter counts runtime hours or parts toward a trigger point. It can email the maintenance team at 85% and open the work order at 100%.

For condition-based work, a machine event rule opens a work order when a health, shock or temperature event fires. Health alerts fire as the vibration health score falls past 60, 40 and 20.

Calendar PM is a recurrence rule on a work order. An operator who classifies a stop can open a work order from the same Classify Downtime panel, which is how RNPSA halved its maintenance response time. See predictive maintenance in IoTFlows.

Build a mixed maintenance plan#

1. Rank machines by what a failure costs

   Price an hour of downtime on each machine, using the method in how to calculate the cost of downtime. Bottlenecks and machines with no standby go to the top.

2. List the failure modes on the top machines

   For each of your top ten machines, write down the ways it has failed and the ways the maker warns it can fail. Your stop history ranked by hours, as in a downtime Pareto, shows which ones cost the most.

3. Run each failure mode through the four questions

   Owed by date, cheap to fix, wears with use, measurable warning. The first yes sets the trigger.

4. Set up each trigger

   Write a recurrence for calendar work, a meter for usage-based work and a limit for condition-based work in your CMMS. For each reactive item, stock the spare.

5. Review the plan every quarter

   Read the work orders each trigger produced. Move a calendar task to a meter when its findings track run hours, and drop a task that keeps finding nothing.