Nature & confidence prediction rate
Nature of the Prediction
The FIDES methodology gives reliability predictions that are failures rates, noted λ ("lambda"). Experimental observation shows that plotting the failure rates versus time usually gives the curve below, called a "bathtub curve".
Thus a component's lifetime can be divided into three periods:
- Infant mortality, precocious failures.
- Useful life, failure rates significantly constant.
- Wear out, wear failures.
During its infant mortality period, the failure rates decrease. A component's probability of failure decreases over time. This is period in which failures are caused by process implementation problems and environmental stress screening.
The useful life is represented by a constant failure rate. The probability of failure is independent of the equipment's number of hours in operation (random failures). This period, often non-existent for mechanical goods, is the reference period for electronics.
During the wear out period, the probability of failure increases with the number of hours of operation: the older the equipment, the greater the chance of a failure. This type of behavior is typical of systems subject to wear or other progressive degradation corresponding to climbing failure rates.
The FIDES evaluation model proposes a reliability prediction with constant failure rates. The infant mortality and wear out periods are excluded from the prediction, for the following reasons:
- Firstly, the infant mortality is representative of an equipment or system's end-of development phase. Controlling the rise in reliability in this phase is crucial to achieve good reliability rapidly.
- The wear out period is also excluded from FIDES because it is in principle sufficiently far away as regards the useful life of electronics systems as covered by FIDES. However, it is essential to check during system design that this is the case. For components whose lifetime is insufficient, approaches other than the sole predicted reliability must be used to address this point, such as the definition of preventive maintenance.
- It is true that at microscopic level very few failure mechanisms strictly meet a "constant rate" type law. Nevertheless:
- Many cumulative failure mechanisms (increasing with time), have a dispersion value that makes them similar to a constant for the periods under consideration.
- The multiplicity and diversity of components, even for a single board, make the accumulation close to a constant.
- The age differences between equipment of a single system or a pool produces a constant rate for an observer of the system.
For these reasons, using a constant failure rate remains the most pertinent approach for a system reliability prediction.
Prediction Confidence Rate
The FIDES methodology is intended to predict realistic reliability levels, close to the average values usually observed (by contrast with pessimistic or conservative values).
An essential question when predicting reliability is the degree of confidence in the value. This question is all the more important that the users have no confidence in the raw results produced by previous methodologies, and that reliability control (quantification and engineering) has become essential for all projects.
One of the aims of the FIDES project is to build this confidence. However, obtaining an exact prediction is not the sole purpose of the FIDES methodology: identifying and controlling the factors affecting reliability may be considered even more important.
As a general rule, a prediction of reliability cannot be linked to a confidence interval, as can be done when measuring failure rates from field returns. In the case of FIDES, it might be possible to calculate a confidence interval for some basic failure rates, but it is practically impossible to predict the confidence rate for all adjustment parameters, even for known and widely used physical acceleration laws.
It is important to keep in mind that reliability belongs to the field of probability: in the same way that it is impossible to predict what will be the life of a product, it is impossible to predict exactly when a failure will occur, or why. The physics of failure is used in some cases to give lifetime probabilities (Time To Failure) and this type of prediction is complementary to the reliability prediction.
Note: Using values with several significant digits in the models does not imply the precision of the expected results.
The prediction's representativeness increases with the number of components considered. The predictions do not usually apply at item level. It is better to avoid comparing reliability prediction and observed reliability below PWA level, and it is better still to compare them at equipment level (assembly of PWAs) or above.