Reliability Growth Models

[Sergey]

Hi
I have two questions related to the "Parameter Estimation for the N.H.P.P. Model, MLE Approach", described in the Reliability Growth Reference, http://www.weibull.com/RelGrowthWeb/parameter_estimation(nh).htm
1. About N.H.P.P. Model Example 2. This example uses failure times of two systems. It is assumed, that "systems are tested simultaneously and any design changes are incorporated into all test systems". Total test times (T[i]) are used to calculate parameters Lambda and Beta. But in text (Eqn. 124) T[i] is a failure time, isn't it? To calculate failure time isn't hard and it some less, than Total Test Time. Why failure time isn't calculated from initial data and total test time is used ?
2. In the chapter "The N.H.P.P Model" it is assumed, that "Between the times {S[i-1], S[i]} when design changes are made on the system, the failure rate can be assumed "CONSTANT". Sometimes (e.g., car model improvement) we don't know times S[i] and modifications are rather implemented continuously. Is assumption "rate is partially constant" isn't significant for accuracy of MLE estimations, or in this case we have to use some other models?
Thanks beforehand, Sergey.

[tarik]

1- Actually in Eq 124, the first time term T* refers to the end time (either the last failure time if the test is failure terminated, or the test duration if the test if time terminated), the second time term Ti is all the failure times.
2- Dr Crow actually assumes in this model that fixes don’t have to be implemented exactly when the failure happens. The model would still apply for what you described.