Ageing Equipment items and the Bathtub Curve
One challenge that every reliability engineer face throughout their career is the lack of reliable reliability data. This issue “forces” most RAM modellers to utilise failure patterns that remain constant over time – the reason being only one: simplicity. The corresponding statistical distribution that represents constant failure rate over time is the random exponential. As many of you know, the exponential distribution only needs one parameter to be defined – the mean time to failure (or MTTF).
This consideration also comes with a number of assumptions:
- It assumes that equipment item is being properly maintained and is operating within its “useful life” phase.
- It assumes that equipment item is considered “as good as new” as long as it is functioning.
If equipment is allowed to deteriorate, or if other factors such as lack of vendor support or obsolescence become significant then wear out may become an issue. The life of an equipment item is frequently divided into three phases:
- Burn in (early life)
- Useful life (constant failure rate)
- Wear out (deterioration and increasing failure rate)
When reliability data are extracted from a reliability database, it is normal practice to assume that burn-in failures are removed by quality testing prior to installation. An example is OREDA where it is specifically stated that failures are removed by quality testing prior to installation for the database.
In addition, most items in the database will be the subject of a maintenance or replacement policy that largely eliminates wear-out failures.
Where equipment is considered to be exhibiting wear out an alternative distribution (such as the Weibull distribution) may be considered.
Whilst it is possible to use Weibull distributions for failure in the simulation models based on assumed wear out factors, this becomes a problem when there is little factual information to derive the shape factor for the equipment under consideration. Therefore an alternative approach (using sensitivity analysis) can be used providing greater transparency of the data values assumed.
For example, taking a normally unmanned installation namely:
- After 5 years equipment failure rate increases by 10%
- After 9 years equipment failure rate increases by 25%
In Maros and Taro, this can be modelled using the transient feature for the failure modes:
Click in Add and change the absolute value for the mean time to failure accordingly:
The results are presented in the figure below:
For the wear out scenario (Case 1), the effect of equipment aging found to be minimal, reducing the through life efficiency by 0.05%. Another noticeable trend is the large performance differential in terms of equipment reliability in later life, there is only a small impact to system performance. This outlines a very important concept – the impact is a combination of capacity, production rate and availability. Thus the relatively small efficiency change is a result of the increasing availability of spare capacity due to production decline as the equipment ages i.e. the decrease in reliability is (in part) offset by increasing redundancy.
The effect of equipment ageing (decreasing reliability) on the above values has been shown to be partly offset by the naturally decreasing production rates of our system that will effectively increase the installed redundancy after production plateau. The field production efficiency in later life is dependent both on the level reduction of equipment reliability and the amount of tail off in production.