Mean time between failures (MTBF)

How long does an asset run, on average, before it fails?

Current value (maintenance case data): 118.0 h - ! Watch. Open the case.

Field	Value
Definition	Total operating hours divided by the number of failures in the same scope and period.
Formula	sum(operating_hours) / sum(failures)
Unit	hours
Granularity	Asset x month; rolled up by asset class.
Direction	Higher is better ↑
Target and thresholds	≥ 150.0 h (amber ≥ 100.0 h)
Parameters	–
Owner	Reliability Engineer
Source	Process data (UCI AI4I 2020) mapped to an asset fleet + work-order generator
Input	table asset_periods, columns operating_hours, failures
Recommended chart	bar - Ranking assets: the lowest bars are the reliability backlog.
Related KPIs	Mean time to repair (MTTR), Availability, Planned maintenance share

How to read it

Higher means more reliable. Compute it as a ratio of sums over the period, never as the average of monthly MTBFs (months with zero failures break that).

Common misreadings

• Averaging per-asset MTBF values instead of sum(hours)/sum(failures) overweights rarely-used assets.
• Using calendar hours instead of operating hours rewards idle assets.
• With very few failures the value is volatile; report the failure count next to it.

Worked example (from the test suite)

asset_id	operating_hours	failures
A	400.0	2
A	300.0	0
B	500.0	1

Expected result: 400.00 h. Ratio of sums: (400 + 300 + 500) / (2 + 0 + 1) = 1200 / 3 = 400 h. Averaging per-row MTBF would fail on the 0-failure row.

This example is asserted in tests/test_formulas.py against the Python function and in tests/test_sql.py against the SQL below.

Implementations

Python function (the single implementation used by the site):

@kpi("mtbf")
def mtbf(df: pd.DataFrame) -> float:
    return safe_ratio(df["operating_hours"].sum(), df["failures"].sum())

Reference SQL (tested on SQLite):

SELECT CAST(SUM(operating_hours) AS REAL) / SUM(failures) FROM asset_periods