8. Assessing Product Reliability
8.3. Reliability Data Collection
8.3.1. How do you plan a reliability assessment test?

8.3.1.3.

Reliability growth (Duane model)

Guidelines for planning how long to run a reliability growth test

A reliability improvement test usually takes a large resource commitment, so it is important to have a way of estimating how long a test will be required. The following procedure gives a starting point for determining a test time:

1. Guess a starting value for , the growth slope. Some guidelines were previously discussed. Pick something close to 0.3 for a conservative estimate (perhaps a new cross-functional team will be working on the improvement test or the system to be improved has many new parts with possibly unknown failure mechanisms), or close to 0.5 for a more optimistic estimate.
2. Use current data and engineering estimates to arrive at a consensus for what the starting MTBF for the system is. Call this M₁. 
3. Let M_T be the target MTBF (the customer requirement). Then the improvement needed on the test is given by

IM = M_T/M₁
4. A first pass estimate of the test time needed is

This estimate comes from using the starting MTBF of M₁ as the MTBF after 1 hour on test and using the fact that the improvement from 1 hour to T hours is just .

Make sure test time makes engineering sense

The reason the above is just a first pass estimate is it will give unrealistic (too short) test times when a high  is assumed. A very short reliability improvement test makes little sense because a minimal number of failures must be observed before the improvement team can determine design and parts changes that will "grow" reliability. And it takes time to implement these changes and observe an improved repair rate. 

Iterative simulation is an aid for test planning

Simulation methods can also be used to see if a planned test is likely to generate data that will demonstrate an assumed growth rate.