To improve MTBF, remove the failure modes that repeat most often. Rank failures by count for one machine, find the cause of the top one, change something so it cannot return, and count again over enough operating hours. MTBF rises fast this way, because each failure removed raises the number by more than the last.
MTBF moves one failure mode at a time
MTBF = operating hours / failures, so the only way up is fewer failures in the same hours. An overall program does not change it until a specific failure stops coming back.
Example: a case packer has 24 failures in 1,200 operating hours, so its MTBF is 50 hours. The log shows four modes.
| Failure mode | Failures | Rate per 1,000 h |
|---|---|---|
| Glue sensor fault | 9 | 7.5 |
| Gearbox seal leak | 6 | 5.0 |
| Drive trip | 5 | 4.2 |
| Other named faults | 4 | 3.3 |
| Total | 24 | 20.0 |
Fix the glue sensor and MTBF goes from 50 to 80 hours. Fix the seal as well and it goes to 133 hours, so the second fix adds more, even with fewer failures removed.
Before you start, separate the short stops. A jam cleared in a minute belongs in mean time between stops, as minor stops vs breakdowns explains.
What it wins in output
Take a mean restore time of 1.5 hours. Availability from failures goes from 50 / 51.5 = 97.1% to 133 / 134.5 = 98.9%.
On 1,200 operating hours, that is 36 hours of failure downtime cut to 13.5 hours. At 85% performance and 99% quality, the gain of 1.8 points of availability is about 1.5 points of OEE.
That lost output is the case for the work. Use the downtime cost calculator to set it in money, and see what is a good MTBF for how to pick the target.
Find out which pattern you are fighting
Machines do not all fail the same way, and the fix depends on the pattern. Look at when each failure happens, in hours since the last repair or replacement.
| What the log shows | Likely meaning | What to try |
|---|---|---|
| Failures soon after a repair | Repair or startup problem | Written repair steps, alignment, a check at restart |
| Failures at random intervals | Conditions or operating stress | Remove the stress, watch the condition |
| Failures at a similar age | Wear-out | Replace on a schedule before that age |
Age-based replacement helps only if the part really wears out. In the Nowlan and Heap study for United Airlines (1978), 89% of the aircraft items analysed had no wear-out zone.
Moubray's Reliability-Centered Maintenance warns that those aircraft frequencies are not necessarily the same in industry.
The lesson that holds is narrower: test whether a part has a wear-out age before you schedule more replacements. See reliability centered maintenance.
A sequence that works on one machine
Write what counts as a failure, and do not change it during the work.
Use the same names every time. A list full of vague labels needs cleaning first.
Count failures that return within 24 hours of a repair. They may be one failure repaired twice.
Use 5 Whys or a fishbone, and bring the operators.
A design, a setting, a part, a standard or a check. A reminder is not a change.
Compare failures per 1,000 hours for that mode, over enough hours.
For the cause work, use root cause analysis. For a machine you are designing around or rebuilding, FMEA helps rank risks before they turn into failures.
Levers, and when each one fits
- Operator care. Cleaning and checks catch loose, dirty or dry parts early. See autonomous maintenance.
- Condition checks. They suit failures with a warning before the stop. See condition based maintenance.
- Scheduled replacement. It suits a clear wear-out age. See preventive vs predictive maintenance.
- Removing the stress. Fix the product variation, setting or environment that causes the failure.
- Changing the design. It costs the most, and it may be the only lasting answer for a chronic fault.
A US Department of Energy guide (2010) cites surveys of predictive programmes showing 35-45% less downtime and 70-75% fewer breakdowns. It covers federal facilities and the surveys are not named, so read those as hopeful, not as a plan.
Did the fix really work?
A few quiet weeks prove little. A failure that comes every 133 hours on average can easily go 200 hours with nothing changed.
Use this rule of thumb for one mode with a constant failure rate. If a mode failed every 133 hours, the chance of 400 hours with no failure by luck alone is e^(-400 / 133) = about 5%.
So wait for about three old intervals with none before you call it proven. For rare failures that takes months, and that is normal.
Traps that raise the number and not the output
- Running slower. Say availability goes from 97% to 98%, while performance falls from 92% to 83%. MTBF is up, and availability x performance has dropped from 89.2% to 81.3%.
- Relabelling. A failure moved to "operational" lowers the count without saving a minute. Keep total stop hours beside MTBF.
- Counting more hours. Using calendar or scheduled time instead of operating time inflates MTBF.
- Ignoring repair speed. Fewer failures with longer repairs may not win anything. See the MTTR reduction plan.
MTBF improvement starts with a clean list of what stopped and why. In Fabrico, a light MES for OEE, operators tag the cause at the line, and OEE per line and shift shows you the gain. See Fabrico, a light MES for OEE.
What people ask next
How do you increase MTBF?
Reduce the failures that repeat most often on that machine. Rank them by count, find the cause of the top one, change something and count again. Run the numbers with the MTBF and MTTR calculator.
Does preventive work always improve MTBF?
Not always. It helps when a part wears out at a known age, and a poorly done intervention can even add failures soon after the work. Check for repeat failures after each stop, and see how to build a preventive program.
How long does it take to see an MTBF improvement?
It depends on how often the failure happened. As a rule of thumb, wait about three of the old intervals with no failure. A mode that used to come monthly needs about a quarter.
What is the difference between improving MTBF and reducing MTTR?
MTBF is how often the machine fails, and MTTR is how long each failure costs you. Both lift availability, so compare the hours each could win. See what is a good MTTR.
Should I improve MTBF on every machine?
No, start where failures cost the most output, usually the bottleneck. A failure on a machine with spare capacity may cost far less. See bottleneck analysis.
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.