Electronic – Reliability and MTBF

Must say I've always been highly sceptical of the meaningfulness of MTBF figures. LEDs will gradually reduce brightness over time, but it will be many times their stated life before they cease to be useable as a power indicator. Running at well below the specc'd current will increase lifetime many-fold. And would failure of a power-on LED be counted as a failure of the product?

The term FIT (failure in time) is defined as a failure rate of 1 per billion hours. A component having a failure rate of 1 FIT is equivalent to having an MTBF of 1 billion hours. Most components have failure rates measured in 100's and 1000's of FITs. For components, such as transistors and ICs, the manufacturer will test a large lot over a period of time to determine the failure rate. If 1000 components are tested for 1000 hours, then that is considered to be equivalent to 1,000,000 hours of test time. There are standard formulas that convert the number of failures in a given test time to MTBF for a selected confidence level. For a system of components, one method of predicting the MTBF is to add the failure rates of each component and then taking the reciprocal. For example, if one component has a failure rate of 100 FITs, another 200 FITs and another 300 FITs, then the total failure rate is 600 FITs and the MTBF is 1.67 million hours. For military systems, the failure rates of each component can be found in MIL-HDBK-217. This document includes formulas to account for environmental and usage conditions such as temperature, shock, fixed or mobile equipment, etc. In initial stages of a design, these calculations are useful in determining the overall reliability of a design(to compare with the specified requirement) and which components are most significant in terms of the system reliability so that design changes can be made if deemed necessary. However, component reliability is more of an art than a science. Many components are so reliable that it is difficult to accumulate enough test time to get a good handle on their MTBF. Also, relating data taken at one set of conditions (temperature, humidity, voltage, current, etc.) to another is open to large errors. As already mentioned in the comments, all of these calculations are mean numbers and are useful in predicting the reliability of a large number of components and systems, but not any individual unit.