Reliability Engineering最新文献

Radio-frequency (RF) radiation is capable, in unfortunate circumstances, of causing undesired ignition of flammable gaseous mixtures and undesired initiation of electroexplosive devices (EEDs). All existing safety standards are based on ‘worst-case’ analyses: this implies that if an event is considered realistically possible then protective measures must be applied as if its probability were unity. Effects that are considered very improbable are effectively assigned a probability of zero. This ‘binary’ approach is unsatisfactory for a number of reasons: in particular, the concatenation of apparently realistic worst-case factors frequently leads to unacceptably ‘pessimistic’ conclusions. The solution to such problems is to evaluate the probability of occurrence of each step in the hazard process. The paper reviews the bases of the new British Standards covering these hazards and presents an analysis of some of the probabilistic factors involved.

The Weibull distribution is commonly used to describe the failure rate parameters of electronic components and has the advantage of presenting an ‘easy-to-understand’ picture of component failure modes with respect to time. This allows conclusions to be drawn on whether an increasing, decreasing or constant failure rate is present which is undoubtedly a key factor in making any reliability statements.

This paper describes in detail how a transistor problem on a colour video display unit (VDU) was screened out and the time-to-fail data plotted on Weibull paper to allow conclusions to be drawn. Two different transistors were evaluated in the same application to determine from their respective Weibull plots, the reliability characteristics of each type. As the product under discussion was in its early production stage, the Weibull plots were not used to predict field fallout from the test data, but to understand the failure mechanism.

The paper also describes the way in which weak component failure distributions can be isolated in Weibull plots by using a form of the Bayesian statistical approach. Isolating the weak distribution in this manner allows a more exact calculation of the required burn-in duration which may often have a marked effect on improving product reliability.

A simple method for predicting failure rate of the parallel and series non-repairable systems is presented. Relations for steady-state and time-dependent failure rate are derived. The use of these relations to obtain the steady-state failure rate of complex systems is illustrated.

In recent years a number of availability measures for oil/gas production and transportation systems have been proposed. The most commonly used measure seems to be the expected throughput, expressed as a fraction of the demand (or design) throughput. A number of terms are used for this measure; for example ‘production availability’, ‘production regularity‘, ‘throughput availability’, ‘quantity availability’, ‘productibility’, ‘productiveness’ and ‘deliverability’.

In this paper a suggestion of standardization is made. This includes the ‘throughput availability’ term for the above measure, and several other definitions for related availability measures. Modelling and computing aspects are discussed both for an analytical approach and simulation.

The purchaser of a high integrity mechanical system requires assurance that the manufacturer's claims will be met. Very elaborate, time-consuming and expensive studies could be envisaged to study the stressing, fatigue, wear and other failure mechanisms of every element and justify the design. A classical theory of stress-strength interference, whilst appearing useful in predicting failures, is seen to be less appropriate to the high integrity case. The dilemma that failures are impossible is contested. Finally, it is proposed that a practical, meaningful and economic analysis can be made using a blend of historical data, conventional reliability techniques such as fault tree analysis, physical inspection, experience and common sense.

A recent European Benchmark Exercise has once again provided some evidence that dependent failures can dominate a reliability assessment by three or four orders of magnitude. In any reliability analysis the effect of dependent failures must therefore be assessed. Methods and procedures currently exist which make a dependent failure assessment possible. The best general approach available now, it is contended, is a systems approach which takes account of system defences in evaluating the contribution of dependent failures. A structured procedure for such an approach is presented.

Errors in steady-state availability estimation by 2-state models of one-unit systems which can be represented by 3-state Markovian models are evaluated. It is found that, except for very specific cases, the 2-state models result in unacceptably large error for the case in which degraded systems are not repaired; hence, multistate models should be used. On the other hand, the 2-state models can present a reasonable accuracy for the cases in which degraded systems are repaired. It is recommended that analyst select better models (2-state or 3-state models) by first estimating the model errors as well as the error due to data uncertainty.

Two kinds of dependence are distinguished: stochastic and state-of-knowledge dependence. Models of stochastic dependence include common cause failures and deal with component failures. They are conditioned on a set of parameters whose ranges of values and their correlations are assessed in the state-of-knowledge models. It is argued that the parametric stochastic models represent the class of failure causes that are not explicitly modeled. Three such stochastic models are examined: the Basic Parameter (BP) model, the Multiple Greek Letter (MGL) model, and the Multinomial Failure Rate (MFR) model. Two problems of the MGL model are discussed. The first has to do with the definition of the parameters. It is shown that β, γ, etc., of the MGL model are defined with reference to a specific component and are used improperly in the statistical calculations. The second problem stems from the fact that the MGL parameters are defined in terms of component failures rather than the events that cause their failures. This results in an artificial increase of the strength of the statistical evidence. The multivariate Dirichlet distribution is used as the state-of-knowledge distribution in the MFR model, since it can model the correlations between the parameters and is a conjuga distribution with respect to the multinomial distribution, thus facilitating Bayesian updating. The Dirichlet distribution can also be used with the BP model to represent the analyst's state of knowledge concerning the numerical values of the parameters of this model.