CHAPTER 12 Reliability & Maintainability McGraw-Hill/Irwin Copyright © 2009 by The McGraw-Hill Companies, Inc. All rights reserved. 12-2 Reliability vs. Quality While product reliability is certainly an important part of quality, the fields treat different problems. Quality: process monitoring, statistical testing of manufactured goods, and product design issues. Reliability: the statistical aspects of the failure of items in the field. 12-3 Reliability of a Single Component Suppose that T is a random variable representing the lifetime of a single component. Then F ( t ) P (T t ) is the cum ulative distribution func tion of T . f (t ) dF ( t ) is the probability density function o f T . dt r (t ) f (t ) 1 F (t ) is the failure rate function of T . 12-4 The Exponential Failure Law The exponential cumulative distribution function and density function are represented mathematically as: F (t ) 1 e f (t ) e t t (The exponential distribution and density function are pictured on the next slide). 12-5 The Exponential Density and Distribution Functions 12-6 Memoryless Property of the Exponential Distribution The exponential distribution is the only continuous distribution possessing this property. Let T be the lifetime of an item. Then it can be shown that if T has the exponential distribution: P (T t s | T t ) P (T s ) In words: the probability that an item will not fail in the next s units of time, given it has not failed up until now (time t), is the same as the probability that a new item will not fail in s units of time. The item “forgets” the t units of time it has been operating. This means that failures occur completely at random. 12-7 Increasing and Decreasing Failure Rate Functions A common probability law for describing the failure characteristics of items in the field is the Weibull distribution: F (t ) 1 e t The Weibull distribution is useful for describing both increasing and decreasing failure rate functions. (See Figure on the next slide). 12-8 Failure Rate Functions for the Weibull Lifetime Distribution 12-9 The Poisson Process in Reliability Suppose that a single item is instantly replaced upon failure, and that successive times between failures are given by independent random variables T1 , T2 , . . . having the exponential distribution. Successive failure times are W 1 T1 W 2 T1 T 2 W 3 T1 T 2 T3 (Refer to the figure on the next slide). 12-10 Realization of a Poisson Process 12-11 Poisson Process (continued) From the figure, we define N(t) as the number of failures that occur up until time t. Then it can be shown that Wn has the Erlang distribution with parameters n and , and N(t) has the Poisson distribution with parameter t. n 1 P (W n t ) e t k! k 0 P ( N (t ) n ) ( t ) e t ( t ) n! n k 12-12 Failures of Components in Series Suppose that N identical components each with failure distribution F(t) are arranged in series (that is, the system fails as soon as a single component fails). Then the cumulative distribution function of the time until failure of the series system, FS(t) is given by: FS ( t ) 1 [1 F ( t )] N 12-13 Failures of Components in Parallel When N identical components are arranged in parallel (where the system fails only when all of the components in the system fail), the cumulative distribution function of the time until failure of the parallel system, FP (t), is: F P ( t ) [ F ( t )] N 12-14 K Out of N systems A K out of N system is one which functions when at least K of the components function. Suppose that R(t) is the probability that a single component functions at time t. Then the probability that the system functions at time t is given by: N! j Nj R K (t ) R ( t ) F ( t ) j K j !( N j ) ! N 12-15 Maintenance Terminology MTBF: Mean Time Between Failures MTTR: Mean Time to Repair Availability = MTBF ____________ MTBF + MTTR 12-16 Deterministic Age Replacement Assumptions: Equipment is operating continuously Ignore downtime for repair and maintenance Planning horizon is infinite All new equipment is identical Only maintenance and replacement costs are considered Objective is to minimize long run costs of replacement and maintenance. The cost of maintaining an item of age u is cu and the replacement cost of a failed item is K. 12-17 Deterministic Age Replacement (continued) The optimal policy is to replace the item when it reaches age t* given by: t* 2K a Extensions: this basic model can be extended to more complex maintenance and salvage value functions, but the optimal solution is often difficult to find. 12-18 Planned Replacement Preventive Maintenance: replace items before they fail to avoid failure during operation. Note: there is no advantage to replacing operating items that fail according to an exponential distribution, because of the memoryless property of the exponential. Planned replacement only makes sense for items with IFR – Increasing Failure Rate - distributions. (See Figure 12-14) 12-19 Successive Cycles for Planned Replacement of a Single Item 12-20 Analysis of Warranties Warranties are a big business for most consumer items such as home electronics, computers, appliances, and automobiles. Two common types of warranties are: Free Replacement Warranty: items that fail are replaced with new items. Pro-rata warranty: A rebate is given proportional to the remaining life of the failed item. 12-21 Free Replacement Warranty Define: T = Lifetime of an item picked at random. = Failure rate of each item. F(t) = Cumulative distribution function of T. C1 = Cost of purchasing a new item with warranty. K = Cost of purchasing a new item with no warranty. W1 = Time that the free replacement warranty is in effect after purchase. (Refer to Figure 12-16 on the next slide). 12-22 Replacement Cycles for Free Replacement Warranty 12-23 Optimal policy for free replacement warranty Assuming that items fail completely at random, (according to an exponential distribution with rate ), the free replacement warranty should be purchased if the cost is less than C1* given by C1* = ( W1 + 1) K 12-24 Optimal Policy for the ProRata Warranty Let W2 be the term of the pro-rata warranty. Then the value at which the consumer is indifferent between purchasing the item with a pro-rata warranty and an item with no pro-rata warranty, say C2* , solves: C2 * KW2 1 e W 2