Page 341 - Improving Machinery Reliability
P. 341
Life Cycle Cost Studies 301
So, where did the failure rate for the pump housing come from? Use experience or
other sources. One-stop shopping for failure rates is not possible!
Select cost data from local plant experiences or proposed cost structures for new
plants.
Step 11: Select Preferred Course of Action Using LCC. The selection of a par-
allel/redundant strategy using ANSI pumps is the most attractive alternative out of
the three proposed because it avoids process failure and thus reduces the high cost of
unreliability. Buy equipment that is electrical-power-efficient, yet reliable and cor-
rectly sized. Ascertain high hydraulic efficiency to make substantial reductions in
electrical power consumption, which is usually a hidden cost item but clearly identi-
fied by LCC as a vital element.
Aidding Uncertainty to the LCC Results
Each element in the above LCC computation is uncertain. Nothing fails on sched-
ule. Nothing is repaired in exactly the same time interval. Seldom are costs for
acquiring goods and services the same price each time. Furthermore, experience tells
us that knowledge of failure modes for equipment is required to make best use of
reliability-centered-maintenance (RCM) strategies. Uncertainty requires the use of
statistical distributions in addition to the usual arithmetic.
Most engineers know about normal (Gaussian) statistical distributions that employ
a mean value, x-bar, to describe central tendencies and a standard deviation, G, to
describe scatter in the data. A better statistical distribution for explaining the life and
repair times for equipment are Weibull distributions with a shape factor, p (similar to
o), and a characteristic life, q.
Statistical distributions give a different value every time data is drawn for solving
spreadsheet problems because of chance selections. Thus Monte Carlo simulation
techniques are used to join probability distributions and economic data to solve prob-
lems of uncertainty using spreadsheet techniques. Monte Carlo simulation tech-
niques use random numbers to generate failure data and cost data considering the
statistical distributions. Monte Carlo results are similar to real life because the results
have variations around a given theme.
Monte Carlo results are used with common spreadsheet programs such as ExcelTM
or LotusTM. Specialized add-in programs such as @RiskTM can add uncertainty to the
calculations. Instead of producing a single answer, the Monte Carlo results provide a
central trend while providing an idea about the expected variations that may result
from many interactions. Ideas about the variations in results are obtained by repeat-
ing the Monte Carlo trials many times and studying the end results. With fast PCs on
almost every engineers desk, it is possible to conduct 10,000 iterations of a compli-
cated spreadsheet in only a few minutes at a very low cost.
A flag was raised in the Alternative #1, Do-nothing Case, section about exponen-
tial failure distributions. With the exponential distribution, the chance for failure is
uniform for each period and this does not conform to equipment expectations where
wear-out failure modes may predominate with their increasing failure rates as equip-