Reliability Prediction and Improvement
Reliability prediction is the combination of the creation of a proper reliability model together with estimating (and justifying) the input parameters for this model (like failure rates for a particular failure mode or event and the mean time to repair the system for a particular failure) and finally to provide a system (or part) level estimate for the output reliability parameters (system availability or a particular functional failure frequency).
Some recognized reliability engineering specialists – e.g. Patrick O'Conner, R. Barnard – have argued that too much emphasis is often given to the prediction of reliability parameters and more effort should be devoted to the prevention of failure (reliability improvement). Failures can and should be prevented in the first place for most cases. The emphasis on quantification and target setting in terms of (e.g.) MTBF might provide the idea that there is a limit to the amount of reliability that can be achieved. In theory there is no inherent limit...! Another of their arguments is that prediction of reliability based on historic data can be very misleading, as a comparison is only valid for exactly the same designs, products, manufacturing processes and maintenance under exactly the same loads and environmental context. Even a minor change in detail in any of these could have major effects on reliability. Furthermore, normally the most unreliable and important items (most interesting candidates for a reliability investigation) are most often subjected to many modifications and changes. Engineering designs are in most industries updated frequently. This is the reason why the standard (re-active or pro-active) statistical methods and processes as used in the medical industry or insurance branch are not as effective for engineering. Another surprising but logical argument is that to be able to accurately predict reliability by testing, the exact mechanisms of failure most have been known in most cases and therefore - in most cases - can be prevented! Following the incorrect route by trying to quantify the complete Reliability Engineering problem in terms of MTBF or Probability and using the re-active approach is referred to by Barnard as "Playing the Numbers Game" and is regarded as bad practise.
For existing systems, it is arguable that responsible programs would directly analyse and try to correct the root cause of discovered failures and thereby may render the initial MTBF estimate fully invalid as new assumptions (subject to high error levels) of the effect of the patch/redesign must be made. Another practical issue concerns a general lack of availability of detailed failure data and not consistent filtering of failure (feedback) data or igoring statistical errors, which are very high for rare events (like reliability related failures). Very clear guidelines must be present to be able to count and compare failures, related to different type of root-causes (e.g. Manufacturing-, Maintenance-, Transport-, System induced or Inherent design failures, ). Comparing different type of causes may lead to incorrect estimations and incorrect business decisions about the focus of improvement.
To perform a proper quantitative reliability prediction for systems may be difficult and may be very expensive if done by testing. On part level, results can be obtained often with higher confidence as many samples might be used for the available testing financial budget, however unfortunately these tests might lack validity on system level due to the assumptions that had to be made for part level testing. These authors argue that it can not be emphasized enough that testing for reliability should be done to create failures in the first place, learn from them and to improve the system / part. The general conclusion is drawn that an accurate and an absolute prediction – by field data comparison or testing – of reliability is in most cases not possible. An exception might be failures due to wear-out problems like fatigue failures. In the introduction of Mil. Std. 785 it is written that reliability prediction should be used with great caution if not only used for comparison in trade-off studies.
See also: Risk Assessment#Quantitative risk assessment - Critics paragraph
Read more about this topic: Reliability Engineering
Famous quotes containing the words prediction and/or improvement:
“Recent studies that have investigated maternal satisfaction have found this to be a better prediction of mother-child interaction than work status alone. More important for the overall quality of interaction with their children than simply whether the mother works or not, these studies suggest, is how satisfied the mother is with her role as worker or homemaker. Satisfied women are consistently more warm, involved, playful, stimulating and effective with their children than unsatisfied women.”
—Alison Clarke-Stewart (20th century)
“... till women are more rationally educated, the progress of human virtue and improvement in knowledge must receive continual checks.”
—Mary Wollstonecraft (1759–1797)