Blog / Reliability
Reliability · Guide

MTBF for Multiple Machines: Series and Parallel

OL By OEE Lab Editorial |Updated 2 November 2026 |8 min read
Short answer

It depends on the question. Fleet MTBF = total operating hours / total failures. Machines in series: 1 / MTBF = 1 / MTBF1 + 1 / MTBF2 + ..., so the line's MTBF is lower than its worst machine. True parallel (either machine can carry the full load), with constant failure rates and no repair: MTBF x 1.5 for two identical machines, and far more with repair.

Pick the question before the formula

Most wrong answers come from using the right formula on the wrong layout. There are three different questions here.

Fleet: separate jobs ABC Series: any failure stops the line ABC Parallel: the other one carries on A1A2
Three layouts, three different formulas. Sketch your own layout first.

The assumptions behind every formula here

  • Constant failure rate. Failures arrive at a steady rate, with no wear-out trend. This is what makes MTBF enough to describe a machine.
  • Failures do not affect each other. One machine failing does not make another more likely to fail.
  • Repair restores the machine as good as before.
  • No buffers, unless stated. A buffer between machines changes the series result.
  • MTBF counts operating hours, as in the IEC definition.

Under a constant failure rate, the chance of running a time t with no failure is e^(-t / MTBF). So only about 37% of runs reach the MTBF without a failure, which means MTBF is not a promise of failure-free running.

Real machines often break these rules. Treat the results as good planning numbers, and use measured line data where you have it.

1. Fleet MTBF: machines doing separate jobs

Use this when you want one number for a group, such as all your presses. Add the operating hours, add the failures, then divide.

Fleet MTBF = total operating hours / total failures
(1,000 + 1,000 + 1,000) / (5 + 2 + 8) = 3,000 / 15 = 200 h
Pool hours and failures. Do not average the machine MTBFs.

Example: machines A, B and C each run 1,000 hours, with 5, 2 and 8 failures. Their MTBFs are 200, 500 and 125 hours.

The plain average of those three is (200 + 500 + 125) / 3 = 275 hours, and it is wrong. It treats the good machine's long gaps as if they offset the failures elsewhere, so use 200 hours.

Read the unit carefully: 200 is machine hours per failure. With all three running at once, somebody gets a call every 1,000 / 15 = 66.7 clock hours.

A fleet number is good for a trend and for staffing. It hides the worst machine, so always list the machines underneath it.

2. Machines in series: any failure stops the line

On a line with no buffers, failure rates add up. The failure rate is 1 / MTBF, so the line's rate is the sum of its machines' rates.

1 / MTBF line = 1 / MTBF A + 1 / MTBF B + 1 / MTBF C
1/200 + 1/500 + 1/125 = 0.005 + 0.002 + 0.008 = 0.015
MTBF line = 1 / 0.015 = 66.7 h
The line's MTBF is always below its worst machine.

If you have N identical machines in series, the line MTBF is the machine MTBF divided by N. Ten machines at 500 hours give a line at 50 hours.

Now add repair times of 2, 3 and 1 hours. The line's MTTR is the rate-weighted average: (0.005 x 2 + 0.002 x 3 + 0.008 x 1) / 0.015 = 0.024 / 0.015 = 1.6 hours.

Line availability from failures is 66.7 / (66.7 + 1.6) = 97.7%. Multiplying the machines' own availabilities (99.01% x 99.40% x 99.21%) gives 97.6%, nearly the same answer.

The small gap has a reason. Multiplying assumes that machines can still fail while the line is stopped, and in practice they usually sit idle.

A buffer changes this, since a short stop upstream may never reach the bottleneck. Then the formula is too pessimistic, as blocked and starved and upstream and downstream stops explain.

3. True parallel: either machine can carry the load

This is real redundancy: two pumps, where one alone meets full demand. The system fails only when both are down at the same time.

No repair: MTBF system = MTBF x 1.5
200 x 1.5 = 300 h
With repair: MTBF system = (3 x lambda + mu) / (2 x lambda x lambda)
lambda = 0.005, mu = 0.25: 0.265 / 0.00005 = 5,300 h
Two identical machines, both running, each with its own repair. Lambda = 1 / MTBF, and mu = 1 / MTTR.

Without repair, the result is modest, 300 hours from 200. With two running, the first failure comes after 100 hours on average, and the other machine then runs alone for another 200.

With a 4-hour repair the picture changes: 5,300 hours. The first failure rarely matters, because the system only fails if the second machine goes down during that short repair.

Strictly, 5,300 hours is the mean time from both machines running until both are down. Later gaps run a little shorter, about 5,200 hours here, because the system restarts with one machine still under repair.

That makes repair speed part of the design. At a 12-hour MTTR, mu is 0.0833 and the system MTBF falls to (0.015 + 0.0833) / 0.00005 = 1,967 hours.

System availability, with a repair crew for each machine, is 1 - (1 - A)². With A = 200 / 204 = 98.04%, that gives 99.96%.

The result needs a clean switchover and no shared failure. A common power feed, control panel or material supply can take both machines out together, and these formulas cannot see that.

Standby is a different case

If the second machine sits idle until needed and cannot fail while waiting, the no-repair MTBF doubles: 400 hours from 200. With a 4-hour repair, (2 x lambda + mu) / (lambda x lambda) = 0.26 / 0.000025 = 10,400 hours.

Those figures assume the switchover always works, so treat them as upper limits.

Two out of three is not what it sounds like

If you need 2 of 3 machines running, the no-repair result is MTBF x 5/6, or 166.7 hours from 200. That is below one machine, yet better than the 100 hours of two machines in series with no spare.

With a 4-hour repair it becomes (5 x lambda + mu) / (6 x lambda x lambda) = 0.275 / 0.00015 = 1,833 hours.

Parallel for capacity is not redundancy

Here is the trap in most factories. Two fillers side by side, each making half the output, are parallel on the floor but not in reliability terms.

When one stops, you lose half the rate at once. Using the 99.96% figure here would badly overstate output.

Two machines, A = 98.04% eachOutput availabilityCost of one failure
True redundancy: one can carry it all99.96%No output lost, if switchover works
Capacity parallel: both needed for full rate98.04%Half the rate until repair
Capacity-weighted availability is (98.04 + 98.04) / 2. For OEE, use this row.

For a capacity layout, use the fleet formula to count events and weight losses by capacity. The OEE side is on OEE for parallel machines and how to average OEE across machines.

Worked example: a mixed line

Example: a feeder (MTBF 500 h), two redundant cappers (200 h each, 4 h repair, so 5,300 h as a pair) and a labeller (125 h). No buffers, so the blocks are in series.

Today: single capper (200 h)66.7 h
Labeller doubled to 250 h90.9 h
Redundant capper pair98.1 h
Both changes161.6 h
Line MTBF by option. Doubling one machine's MTBF does almost as much as buying a second capper.

With one capper: 0.002 + 0.005 + 0.008 = 0.015, so 66.7 hours. With the pair: 0.002 + 0.00019 + 0.008 = 0.01019, so 98.1 hours.

Doubling the labeller alone gives 0.002 + 0.005 + 0.004 = 0.011, so 90.9 hours, with no second machine to buy. Both changes give 0.002 + 0.00019 + 0.004 = 0.00619, so 161.6 hours.

Each option's gain is the failure rate it removes, which is why you rank machines by rate first. Here the labeller has the largest rate, at 0.008.

To see what each option does for output, turn the hours into availability and OEE. The MTBF and MTTR calculator handles one machine at a time, and line OEE vs machine OEE explains how the levels connect.

See it live in Fabrico

Formulas estimate a line, and stop data measures it. Fabrico takes stops live from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and shows OEE per line and shift, so you can test these estimates against what really happened. See how Fabrico tracks stops per machine and line.

What people ask next

How do you calculate MTBF for multiple machines?

For a group of separate machines, divide total operating hours by total failures. For machines in series, add the failure rates and invert the sum. Never just average the MTBFs.

How do you calculate MTBF in parallel?

For two identical machines where one alone can carry the load, with constant failure rates and no repair, multiply MTBF by 1.5. With repair, use (3 x lambda + mu) / (2 x lambda x lambda). If both machines are needed for full output, this does not apply.

Why is my line's MTBF lower than every machine's MTBF?

Because any machine can stop the line, so the failure rates add up. Three machines at 200, 500 and 125 hours give a line at 66.7 hours. Buffers can soften this, as blocked and starved shows.

Does redundancy always improve MTBF?

Not always in the way people expect. Without repair, two machines give only 1.5 times the MTBF, and a 2 out of 3 setup gives 5/6 of one machine's figure. The large gains come from repair during the time the other machine carries on.

What if my machines do not have a constant failure rate?

Then MTBF alone does not describe them, and these formulas are rough. A wear-out pattern needs age-based data or simulation. See reliability centered maintenance for failure patterns.

Should I use MTBF or OEE to compare machines?

Use both, as they answer different questions. MTBF says how often failures happen, and OEE says how much output you lost in total. See how to average OEE across machines.

See OEE per line and shift on your own machines

Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.

Book a Fabrico demo

← More OEE Lab guides · Open the MTBF and MTTR calculator →

OEE essentials: OEE calculator · What is OEE · Six big losses · Availability · TEEP · Cost of downtime · OEE benchmark · Free OEE tracker · Free OEE software · How to track downtime · How to implement OEE · Light MES vs full MES