Reliability and Availablity

Reliability and Availablity This set of notes is a combination of material from Prof. Doug Carmichael's notes for 13.21 and Chapter 8 of Engineering Statistics Handbook. NIST/SEMATECH e-Handbook of Statistical Methods, http://www.itl.nist.gov/div898/handbook/, 2005. available free from: see: NIST/SEMATECH e-Handbook of Statistical Methods on CD Including and improving reliability of propulsion (and other) systems is a challenging goal for system designers. An approach has developed to tackle this challenge: 1. a design and development philosophy 2. a test procedure for components and total systems 3. a modelling procedure based on test results, field tests and probability (statistics Design and development philosophy recognition that reliability is a product is essentiall the abscence of failures or substandard performance of all critical systems in the design, followed by an examination of the factors leading to failure. Causes of failure: a. loading: (inaccurate estimates of) thermal, mechanical or electriacl including vibrations b. strength: (inaccurate estimates of) the load carrying capacity of the components c. environment: presence of dirt, high temperature, shock, corrosion, moisture, etc. d. human factors: heavy handed operators ("sailor proof"), wrong decisions (operator error), criminal activities (sabatoge), poor design, tools left in critical components, use of incorect replacements e. quality control: or lack thereof; loose control of materials and manufacture, lack of inspection, loose specifications f. accident; act of God, freak accidents, collisions g. acts of war: terorism, war damage designer should recognize these potential causes for failure and try to design devices that will resist failure. Detailed Design Features a. try to account for all possible situations in the design stage and eliminate possible failures. Delivering maximumloads and minimum strengths b. assume that every component can fail, examine the outcome of the failure and try to reduce the risk of damage. Failure Modes and Effects Analysis (FEMA) c. institute strict quality control in manufacture and maintenance d. have cleaarly defined specifications (including material specifications and methods of testing) e. develop technology to meet new challenges. conduct development testing. f. consider possible war damage and ship collision g. carry out development testing in arduous conditions System Design Features a. calculate probability of failures. (reliability and availability analysis b. improve system design by standby or redundant systems c. analyze failures, note trends d. specify clearly all operating procedures (good operating manuals) e. require inspection, maintenance and replacement procedures (trend analysis) Failure testing and analysis from field or laboratory tests on components or systems determine number of operating units as a function of time (life): 12/13/2005 1 set up N_surv nominal survival curve 100 a typical survival curve might look like this: number surviving 80 60 40 20 0 20 40 60 80 100 time define the failure rate at time t as λ= −1⋅ proportion_failing_in_δt = δt d N( t) dt N( t) −δN( t) 1 −1 d ⋅ N( t) ⋅ = N( t) δt N( t) dt - as "rate" > 0 consistent with population decline units are: −1⋅ to make some estimates based on this sample: d N( t) N( t) ⌠ N( t) ⎞ ln⎛⎜ λ ( τ ) dτ ⎟ = −⎮ ⌡ ⎝ N( 0 ) ⎠ 0 = λ⋅dt fail_rate( t , δt) := − N( t + δt) − N( t) N( t)⋅δt fail_rate( 40 , 0.01) = 0.01 fail_rate( 10 , 0.01) = 0.01 −1⋅ N = N( t = 0) define ... I ⎛⎜ ⌠ ⎞⎟ N( t) = NI⋅exp −⎮ λ dτ ⎜ ⌡ ⎟ t ⎝ 0 ⎠ d N( t) dt N( t) =λ set ... or ... calculate for modest δt = 0.01 and t =10, 40, 60, 120 fail_rate( 60 , 0.01) = 0.01 looks like λ = failure rate is a constant, not unusual or ... ⎛⎜ ⌠ t ⎟⎞ N( t) = NI⋅ exp −⎮ λ ( τ ) dτ ⎜ ⌡0 ⎟ ⎝ ⎠ t =λ λ= −1⋅ −1 time fail_rate( 120 , 0.01) = 0.01 −1 d ⋅ N( t) = constant N( t) dt d N( t) dt N( t) t ⌠ integrate from N( t) ⎞ ln⎛⎜ λ dτ ⎟ = −⎮ 0 to t ⌡ ⎝ N( 0 ) ⎠ 0 =λ λ := 0.01 N( t) := NI⋅exp( −λ⋅t) NI := 100 100 N( t) 50 0 0 50 t 12/13/2005 2 100 N.B. failure rate is not necessarily the same as (but can be related to) (in this case it is) the probability of failure see Engineering Statistics Handbook an actual failure rate curve might look like this: set up bath tub nominal failure rate 2 100 three regions are evident: 800 failure rate 0 - 100 early failure period = infant mortality rate 1 0 100 - 800 intrinsic failure period aka stable failure period => intrinsic failure rate 0 200 400 600 800 > 800 wearout failure period - materials wear out and degradation failures occur at an ever increasing rate 1000 for most systems, the failure ratetime is relatively constant except for wer in and wear out. If the failure rate is constant, the component is said to have random failure. Reliability (applies to a particular mission with a defined duration.) defined as the probability of operating without degraded performance during a specific time period. At time t 1 , the number operating is N(t1 ) and NI is the initial number. The reliability is: ( ) R t1 = ( ) ⌠1 ⎛ N( t1 ) ⎞ ln⎜ = −⎮ λ dt ⎟ ⌡ 0 ⎝ NI ⎠ t N t1 −1⋅ since ... NI with λ = constant d N( t) N( t) = λ⋅dt ( ( ) R t1 = exp −λ⋅ t1 ) ( ) and if ... λ*t1 << 1, ( ) and expanding in a series ... R t1 = 1 − λ ⋅ t1 e.g. ⎛ ⌠ t1 ⎞ ⎜ ⎮ ⎟ R( t 1 ) = = exp − λ dt ⎜ ⎟ ⌡ NI ⎝ 0 ⎠ R t1 = 1 − λ⋅t1 λt1 := 0.05 ( ) N t1 2 3 λ⋅t1 ) λ⋅t1 ) ( ( + − + .. 1 − λt1 = 0.95 2! 3! ( ) exp −λt1 = 0.951 Mean Time Between (Operational Mission) Failure (MTB(OM)F with field testing,data is collected in the form of operating time, failures and repair time. During the field operation of a component or a system, there is a total number of operating hours and a total number of failures. MTB(OM)F is defined MBT( OM) F = accumulated_life number_of_failures For random failures, the failure rate if ... t1 MBT( OM) F 12/13/2005 λ= <1 number_of_failures accumulated_life = 1 MBT( OM) F t1 R t1 = 1 − λ⋅t1 = 1 − MBT( OM) ⋅ F ( ) 3 Probability of Failure (Q or F) since probability of success + failure = 1 t1 if ... λ*t1 Q = 1 − R = 1 − exp −λ⋅ t1 = λ⋅t1 = MTBF<< 1 ( R+ Q=1 ) now consider separate components C1 and C2 having R 1 and R2 and Q1 and Q2 . then ... (R1 + Q1)⋅ (R2 + Q2) = 1 (R1 + Q1)⋅ (R2 + Q2) expand → R1 ⋅R2 + R1 ⋅Q2 + Q1 ⋅R2 + Q1 ⋅Q2 R1 ⋅R2 = probability_both_C1_and_C2_operating R1 ⋅Q2 = probability_C1_operating_and_C2_failed R2 ⋅Q1 = probability_C2_operating_and_C1_failed Q1 ⋅Q2 = probability_C1_and_C2_failed Series Systems If it is necessary for all systems to operate, then this termed a series system and is represented as a circuit as: Rseries = R1 ⋅R2 From above; the probability that both are operating is ... e.g. ... ∑ (λ i⋅t1)⎤⎥ = ∏ Ri Rseries = R1 ⋅R2 ⋅R3 .. Rn = exp⎡− ⎢ ⎣ more generally, R1 := 0.9 R2 := 0.9 n R3 := 0.9 ⎦ n 2 components R1 ⋅R2 = 0.81 Rn := 0.9 6 components 6 Rn = 0.531 Parallel Systems If there is redundancy, and either C1 or C2 is required for operation then this is a parallel scheme ... Rparallel = R1 ⋅R2 + R1 ⋅Q2 + Q1 ⋅R2 = 1 − Q1 ⋅Q2 generally ... Rparallel = 1 − Q1 ⋅Q2 ⋅Q3 .. Qn = 1 − ∏ Qi when Qi = Qn Rparallel = 1 − Qi n e.g. ... R1 := 0.9 R2 := 0.9 R3 := 0.9 Qi := 0.1 Rn := 0.9 2 components 12/13/2005 4 2 1 − Qi = 0.99 n R out of N see Handbook of Statistical Methods section 8.1.8.4.R out of N model If a system has n components and reqires any r to be operational; assuming all components have thesame reliability Ri all components operate independent of one another (as far as failure is concerned) the system can survive any (n - r) components failing, but fails at the instant the n - r - 1)th component fails System reliability is given by the probability of exactly r components surviving to time t + the probability of exactly (r + 1) components surviving to time t ... up to all n surviving. These are binomial probabilities: n Rs( t) = ∑ i= r ⎡⎛ n ⎞ ⋅R i⋅ 1 − R n−i⎤ ⎢⎜ ⎟ i ( i) ⎥ ⎣⎝ r ⎠ ⎦ for example (where Ri are not necessarily equal ... n=4 r = 2 i.e. four components of which two are required for operation 2 components R1 ⋅R2 ⋅Q3 ⋅Q4 + R1 ⋅R3 ⋅Q2 ⋅Q4 + R1 ⋅R4 ⋅Q2 ⋅Q3 + R2 ⋅R3 ⋅Q1 ⋅Q4 + R2 ⋅R4 ⋅Q1 ⋅Q3 + R3 ⋅R4 ⋅Q1 ⋅Q2 3 components R1 ⋅R2 ⋅R3 ⋅Q4 + R1 ⋅R3 ⋅R4 ⋅Q2 + R1 ⋅R2 ⋅R4 ⋅Q3 + R2 ⋅R3 ⋅R4 ⋅Q1 R1 ⋅R2 ⋅R3 ⋅R4 n = 4 components sum all these for R s Rs = R1 ⋅R2 ⋅Q3 ⋅Q4 + R1 ⋅R3 ⋅Q2 ⋅Q4 + R1 ⋅R4 ⋅Q2 ⋅Q3 + R2 ⋅R3 ⋅Q1 ⋅Q4 + R2 ⋅R4 ⋅Q1 ⋅Q3 + R3 ⋅R4 ⋅Q1 ⋅Q2 ... + R1 ⋅R2 ⋅R3 ⋅Q4 + R1 ⋅R3 ⋅R4 ⋅Q2 + R1 ⋅R2 ⋅R4 ⋅Q3 + R2 ⋅R3 ⋅R4 ⋅Q1 ... + R1 ⋅R2 ⋅R3 ⋅R4 N.B. a series system is one with r = n i.e. all components must operate. a parallel system is one with r = 1 Standby Systems Standby scenario will be more reliable than parallel as seen in Handbook of Statistical Methods section 8.1.8.5.Standby model Availability Availability is the probability that a component is operational, i.e. it is not being repaired MTTR = mean_time_to_repair = total_time_for_repairs number_of_repairs For every failure there should be a repair, so that the average component is repaired for the average time after it has operated for the average time between failures. Average time between failures is MTBF and for repair MTTR, so assuming component is either operating or being repaired ... availability = A = if ... 12/13/2005 operating_time operating_time + repair_time MTBF > MTTR ( << ) = MTBF MTBF + MTTR which it should be ... 5 A= MTBF MTBF + MTTR =1− probability that it is being repaired is ... Q A MTTR 1 MTBF 1+a = (1 + a) −1 =1−a a<1 MTTR QA = 1 − A = MTBF and as above, availability for series systems would be .. Aseries = A1 ⋅A2 ⋅A3 .. An = ∏ Ai n and parallel ... Aparallel = 1 − Q1 ⋅ Q2 ⋅ Q3 .. Qn = 1 − ∏ Qi n 12/13/2005 6 ( << )