APPLICATIONS OF RELIABILITY ANALYSIS TO POWER ELECTRONICS SYSTEMS

Similar documents
Monte Carlo Simulation for Reliability Analysis of Emergency and Standby Power Systems

of Emergency and Standby Power Systems

Additional File 1 - Detailed explanation of the expression level CPD

STOCHASTIC BEHAVIOUR OF COMMUNICATION SUBSYSTEM OF COMMUNICATION SATELLITE

Harmonic oscillator approximation

CHAPTER 9 LINEAR MOMENTUM, IMPULSE AND COLLISIONS

Root Locus Techniques

Designing of Combined Continuous Lot By Lot Acceptance Sampling Plan

Distributed Control for the Parallel DC Linked Modular Shunt Active Power Filters under Distorted Utility Voltage Condition

Specification -- Assumptions of the Simple Classical Linear Regression Model (CLRM) 1. Introduction

Small signal analysis

ELG3336: Op Amp-based Active Filters

OPTIMISATION. Introduction Single Variable Unconstrained Optimisation Multivariable Unconstrained Optimisation Linear Programming

MODELLING OF TRANSIENT HEAT TRANSPORT IN TWO-LAYERED CRYSTALLINE SOLID FILMS USING THE INTERVAL LATTICE BOLTZMANN METHOD

Introduction to Interfacial Segregation. Xiaozhe Zhang 10/02/2015

MULTIPLE REGRESSION ANALYSIS For the Case of Two Regressors

On the SO 2 Problem in Thermal Power Plants. 2.Two-steps chemical absorption modeling

EECE 301 Signals & Systems Prof. Mark Fowler

Chapter 3 Differentiation and Integration

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

728. Mechanical and electrical elements in reduction of vibrations

Variable Structure Control ~ Basics

Chapter 11. Supplemental Text Material. The method of steepest ascent can be derived as follows. Suppose that we have fit a firstorder

Department of Electrical and Computer Engineering FEEDBACK AMPLIFIERS

On multivariate folded normal distribution

Method Of Fundamental Solutions For Modeling Electromagnetic Wave Scattering Problems

Digital Simulation of Power Systems and Power Electronics using the MATLAB Power System Blockset 筑龙网

Absorbing Markov Chain Models to Determine Optimum Process Target Levels in Production Systems with Rework and Scrapping

Confidence intervals for the difference and the ratio of Lognormal means with bounded parameters

KEY POINTS FOR NUMERICAL SIMULATION OF INCLINATION OF BUILDINGS ON LIQUEFIABLE SOIL LAYERS

Prof. Paolo Colantonio a.a

Estimation of Finite Population Total under PPS Sampling in Presence of Extra Auxiliary Information

Scattering of two identical particles in the center-of. of-mass frame. (b)

Alpha Risk of Taguchi Method with L 18 Array for NTB Type QCH by Simulation

Two Approaches to Proving. Goldbach s Conjecture

Control Strategy of Cascade STATCOM Based on Internal Model Theory

No! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Survey Results. Class 15. Is the following possible?

Team. Outline. Statistics and Art: Sampling, Response Error, Mixed Models, Missing Data, and Inference

Chapter 6 The Effect of the GPS Systematic Errors on Deformation Parameters

Electrical Circuits II (ECE233b)

DEADLOCK INDEX ANALYSIS OF MULTI-LEVEL QUEUE SCHEDULING IN OPERATING SYSTEM USING DATA MODEL APPROACH

2.3 Least-Square regressions

Verification of Selected Precision Parameters of the Trimble S8 DR Plus Robotic Total Station

Basic Statistical Analysis and Yield Calculations

Comparative Study on Electromagnetic and Electromechanical Transient Model for Grid-connected Photovoltaic Power System

Not at Steady State! Yes! Only if reactions occur! Yes! Ideal Gas, change in temperature or pressure. Yes! Class 15. Is the following possible?

OPTIMAL COMPUTING BUDGET ALLOCATION FOR MULTI-OBJECTIVE SIMULATION MODELS. David Goldsman

Solution Methods for Time-indexed MIP Models for Chemical Production Scheduling

Lecture 2 Solution of Nonlinear Equations ( Root Finding Problems )

EE 330 Lecture 24. Small Signal Analysis Small Signal Analysis of BJT Amplifier

Modeling of Wave Behavior of Substrate Noise Coupling for Mixed-Signal IC Design

ENTROPY BOUNDS USING ARITHMETIC- GEOMETRIC-HARMONIC MEAN INEQUALITY. Guru Nanak Dev University Amritsar, , INDIA

Computational Biology Lecture 8: Substitution matrices Saad Mneimneh

A LINEAR PROGRAM TO COMPARE MULTIPLE GROSS CREDIT LOSS FORECASTS. Dr. Derald E. Wentzien, Wesley College, (302) ,

Start Point and Trajectory Analysis for the Minimal Time System Design Algorithm

This appendix presents the derivations and proofs omitted from the main text.

3 Implementation and validation of analysis methods

Resonant FCS Predictive Control of Power Converter in Stationary Reference Frame

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

m = 4 n = 9 W 1 N 1 x 1 R D 4 s x i

A Novel Approach for Testing Stability of 1-D Recursive Digital Filters Based on Lagrange Multipliers

Extended Prigogine Theorem: Method for Universal Characterization of Complex System Evolution

Computer Control Systems

: Numerical Analysis Topic 2: Solution of Nonlinear Equations Lectures 5-11:

Physics 111. CQ1: springs. con t. Aristocrat at a fixed angle. Wednesday, 8-9 pm in NSC 118/119 Sunday, 6:30-8 pm in CCLIR 468.

PROBABILITY-CONSISTENT SCENARIO EARTHQUAKE AND ITS APPLICATION IN ESTIMATION OF GROUND MOTIONS

FEEDBACK AMPLIFIERS. v i or v s v 0

Probability, Statistics, and Reliability for Engineers and Scientists SIMULATION

Chapter 6. Operational Amplifier. inputs can be defined as the average of the sum of the two signals.

A New Inverse Reliability Analysis Method Using MPP-Based Dimension Reduction Method (DRM)

A NOVEL FAMILY OF WEIGHTED AVERAGE VOTERS FOR FAULT-TOLERANT COMPUTER CONTROL SYSTEMS

Comparison of the Population Variance Estimators. of 2-Parameter Exponential Distribution Based on. Multiple Criteria Decision Making Method

Chapter 13: Multiple Regression

8 Waves in Uniform Magnetized Media

Chapter 5 rd Law of Thermodynamics

NUMERICAL DIFFERENTIATION

University of Washington Department of Chemistry Chemistry 452/456 Summer Quarter 2014

Pythagorean triples. Leen Noordzij.

Foresighted Resource Reciprocation Strategies in P2P Networks

ECEN 5005 Crystals, Nanocrystals and Device Applications Class 19 Group Theory For Crystals

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS

Discrete Simultaneous Perturbation Stochastic Approximation on Loss Function with Noisy Measurements

MODELLING OF STOCHASTIC PARAMETERS FOR CONTROL OF CITY ELECTRIC TRANSPORT SYSTEMS USING EVOLUTIONARY ALGORITHM

AP Statistics Ch 3 Examining Relationships

Stability Analysis of Inverter for Renewable Energy

Electric and magnetic field sensor and integrator equations

Analysis of the induction machines sensitivity to voltage sags. F. Córcoles and J. Pedra Ll. Guasch

OPTIMAL CONTROL FOR THREE-PHASE POWER CONVERTERS SVPWM BASED ON LINEAR QUADRATIC REGULATOR

S-Domain Analysis. s-domain Circuit Analysis. EE695K VLSI Interconnect. Time domain (t domain) Complex frequency domain (s domain) Laplace Transform L

Simulated Power of the Discrete Cramér-von Mises Goodness-of-Fit Tests

A NUMERICAL MODELING OF MAGNETIC FIELD PERTURBATED BY THE PRESENCE OF SCHIP S HULL

Two-Layered Model of Blood Flow through Composite Stenosed Artery

Image Registration for a Series of Chest Radiograph Images

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Approximating the Sum Operation for Marginal-MAP Inference

Chapter 9: Controller design. Controller design. Controller design

CISE301: Numerical Methods Topic 2: Solution of Nonlinear Equations

bounds compared to SB and SBB bounds as the former two have an index parameter, while the latter two

The gravitational field energy density for symmetrical and asymmetrical systems

Lesson 16: Basic Control Modes

Transcription:

APPLICATIONS OF RELIABILITY ANALYSIS TO POWER ELECTRONICS SYSTEMS Chanan Sngh, Fellow IEEE Praad Enjet, Fellow IEEE Department o Electrcal Engneerng Texa A&M Unverty College Staton, Texa USA Joydeep Mtra, Sr. Member IEEE Department o Electrcal & Computer Engneerng North Dakota State Unverty Fargo, North Dakota USA ABSTRACT Relablty and cot conderaton play an mportant role n the choce o varou alternatve. Thee could be alternate degn to meet an objectve or could be alternate plannng cenaro. To be able to make tradeo between cot and relablty, relablty need to be quanted and there need to be method or predctng the relablty. Th paper decrbe three technque o relablty analy, a technque baed on Markov procee, cut et method and Monte Carlo mulaton. Thee technque are llutrated ung two example. The Markov proce llutrated by applcaton to tudy ault tolerant eature n PEBB baed converter conguraton. Smlarly the Monte Carlo mulaton and cut et technque are llutrated by applcaton to tandby power ytem ncludng unnterruptble power upply ytem. BACKGROUND Th ecton gve a bre ntroducton o the relablty evaluaton method. There are many method avalable n the lterature; here three method wll be revewed. The two analytcal method are the Markov procee and cut et and one orm o Monte Carlo Smulaton alo decrbed. ANALYTICAL METHOD The analytcal method decrbed here ue the concept o Markov chan and Markov cut-et. A detaled treatment o thee technque avalable n [1]; a ummary o the method preented here. Markov Chan A Markov chan a equence o event contng o the tranton o a component or ytem o component amongt a et o tate uch that the tate to whch the component or ytem trant n the uture depend only on the current tate o the component or ytem and not on the tate t ha tranted 1

through n the pat. I a ytem conorm to (or, more realtcally, approxmate) th behavor, then the theory o Markov chan may be appled to t. In th knd o analy, normaton o ntertate tranton rate ued to determne the rate o trantng to alure tate. The tranton rate rom tate to tate j the mean rate o the ytem pang rom tate to tate j. In th paper, the duraton o component tate are aumed exponentally dtrbuted, whch mean that the ntertate tranton rate are contant. Th explaned n the next ecton. The tate probablte can be obtaned by olvng BP = C (1) where B matrx obtaned rom A by replacng the element o an arbtrarly elected row k by 1; A matrx o tranton rate uch that the element a j = j and a = a ; j contant tranton rate rom tate to j; P column vector whoe th term p the teady-tate probablty o the ytem beng n tate ; C column vector wth kth element equal to one and other element et to zero. j j Once the teady-tate probablte have been calculated, the relablty ndce [1] can be computed ung the ollowng relatonhp. Frequency o ytem alure: The relatonhp or the requency o ytem alure [1] gven by Eq. (2) or Eq. (3) = p ( S F) = F j F j p (3) j j ( S F) (2) where the requency o the ytem alure, S the ytem tate pace and F the ubet o aled tate. Mean Down Tme: The expected tme o tay n F n one cycle where r P / = (4) 2

P = (5) p F Cut Set A cut et a et o component that, removed rom the ytem, reult n lo o power to the load pont. I th et doe not contan any cut et a a ubet, then t called a mnmal cut et. In th denton, component ued n a wder ene o hardware or a partcular ytem condton. In general cut et contanng m component are called m-order cut et. Generally up to econd order cut et are condered and the contrbuton rom hgher order cut et condered neglgble. The ollowng relatonhp are mportant n cut et calculaton. 1) Frequency and duraton o a cut et (a) Frt order cut et [6] : c = (6) r = r (7) c where, r = requency and mean duraton o cut et c c, r = alure rate and mean duraton o component comprng cut et (b) Second order cut et [6] = r + r) (8) c j( j r c = rr j/( r + rj) (9) where, = alure rate o component and j comprng cut et j r, r= mean alure duraton o component and j comprng cut et j (c) Condtonal econd order cut et [6]: Aumng that component al gven component j ha aled, c = r) (10) j ( j r c = rr j/( r + rj) (11) 3

The term alure requency and alure rate are oten ued nterchangeably. Strctly peakng alure rate the mean number o alure per unt o the up tme and alure requency the mean number o alure per unt o the total tme. I the probablty o ytem beng up cloe to unty, then thee two quantte are very cloe. In general, alure requency le than alure rate. 2) Frequency and duraton o nterrupton [6]: = requency o load nterrupton = (12) c r = mean duraton o load nterrupton = r / (13) c c The theory concernng Markov chan and mnmal cut-et dealt wth n greater detal n (4). Thee concept are ued a ollow to analyze the ample ytem. Monte Carlo mulaton Th another method o determnng relablty ndce. Th method decrbed n detal n [1], whle an applcaton o th method to the ample ytem gven n [9]. A ummary o the method preented here. The relablty ndce o an actual phycal ytem can be etmated by collectng data on the occurrence o alure and the duraton o repar. The Monte Carlo method mmc the alure and repar htory o the component and the ytem by ung the probablty dtrbuton o the component tate duraton. Stattc are then collected and ndce etmated ung tattcal nerence. Though there are derent way o mplementng Monte Carlo mulaton, the technque decrbed here the next event method and capable o mulatng dependent event. Th a equental mulaton method whch proceed by generatng a equence o event ung random number and probablty dtrbuton o random varable repreentng component tate duraton. A lowchart gven n Fg. 1. The nput data cont o the alure rate () and mean down tme (r) o every component. The alure rate the recprocal o the mean up tme. The mean down tme the recprocal o the repar rate (µ). The alure and repar rate, and µ, o a component wll be ued to determne how long the component wll reman n the UP tate and the DOWN tate. Smulaton could be tarted rom any ytem tate, but t cutomary to begn mulaton wth all the component n the UP tate. 4

START READ FAILURE RATE AND DURATION DATA FOR ALL COMPONENTS SET INITIAL STATE OF ALL COMPONENTS AS UP FOR EACH COMPONENT DRAW A RANDOM NUMBER AND COMPUTE THE TIME TO THE NEXT EVENT FIND THE MINIMUM TIME AND CHANGE THE STATE OF THE CORRESPONDING COMPONENT; UPDATE TOTAL TIME NO IS THERE A CHANGE IN SYSTEM STATUS? YES UPDATE INDICES SIMULATION CONVERGED? NO YES PRINT OUTPUT STOP Fg. 1. Flowchart or next-event mulaton. The tme to the next event generated by ung the nvere o probablty dtrbuton method. Th explaned a ollow. I the tranton tme o the component are aumed to be exponentally dtrbuted: t ( t) = ρe ρ (14) where ρ the tranton rate. The mean tranton tme, thereore, 0 1 ( t) dt= (15) ρ 5

Th mean that, or ntance, a component UP, then, regardle o how long t ha been n the UP tate, the expected tme to the next alure 1/,.e., the mean up tme. The probablty dtrbuton uncton o the tranton tme T would be F( t) t ρt = P( t T) = ρe dτ = 1 0 e ρt (16) Now F(t) can be regarded a a random varable, unormly dtrbuted between 0 and 1. Th mean that the urvvor uncton S ρt ( T) = 1 F( T) = e (17) alo unormly dtrbuted between 0 and 1. So a random number R generated, n 0 R 1, t can be aocated wth the event that the next tranton occur ater tme t r, gven by n ρt r R n = e, that, ln( Rn) tr = (18) ρ Th method ued to determne the tme to the next tranton or every component, ung or µ or ρ, dependng on whether the component UP or DOWN. At the end o any mulated tme nterval [0, t], where t = total up tme n [0, t] + total down tme n [0, t], the etmate o the relablty ndce are gven a ollow. alure rate: number o alure n [0, t] t = (19) totaluptmen [0, t] mean down tme: totaldown tmen [0, t] r t = (20) number o alure n [0, t] The value o and r at the ntant the mulaton converge are the relablty ndce or the ytem a obtaned rom the Monte Carlo method. The mulaton ad to have converged when the ndce attan table value. Th tablzaton o the value o an ndex meaured by t tandard error, dened a: σ η = (21) n c 6

where σ = tandard devaton o the ndex n = number o cycle mulated c Convergence ad to occur when the tandard error drop below a prepeced racton, ε, o the ndex,.e., when η ε (22) I, or ntance, the mean down tme r choen a the ndex to converge upon, then, ater every ytem retoraton mulated, the ollowng relaton teted or valdty: σ r ε r (23) n c I th crteron ated, the mulaton ad to have converged. Smulaton advantageou n that t not only allow the computaton o ndce at varou pont n the ytem, but alo permt the accumulaton o data pertanng to the dtrbuton o thee ndce, thereby aordng a better undertandng o the ytem behavor. APPLICATIONS The technque decrbed n the paper are llutrated by two applcaton. The rt applcaton to model ault tolerant eature n the PEBB- baed converter propoed n re. [3]. The econd applcaton the tudy o relablty mprovement by ung UPS a tandby power MODELING FAULT TOLERANT FEATURES IN THE PEBB-BASED CONVERTERS [3] In th ecton a modelng approach preented to analyze the ault tolerant behavor o PEBB baed converter propoed n [3]. Fg. 2 (a) how a detaled connecton dagram o combnng three, 3-phae nverter block wth 18 IGBT wtche. The nterconnecton nductance between the module lmt the hgh requency crculatng current. Fg. 2 (b) how the vector dagram o the undamental voltage generated by IGBT wtch par. 1. Fault Tolerant Feature & Relablty Analy: 7

From Fg. 2 (a) and (b) t clear that alure (open crcut) o IGBT wtche 3,6 and 7,10 or 2,5 and 13,16 or 15,18 and 8,11 reult n an open delta tuaton and the modular nverter are tll capable o upplyng the load at a reduced power level. However, the alure o IGBT wtch par 1,4 or 9,12 or 14,17 wll caue a crtcal alure and reult n a ytem hut down. The trade o between relablty and cot can be acheved by the knowledge o cot o nterrupton and contructng relablty model and computng the relablty ndce. The two ndce that are ueul n th tuaton are: 1. Frequency o alure: Th the mean number o alure per year. 2. Mean duraton o alure: Th the mean duraton o a alure event. The cot o nterrupton n ome applcaton depend on the total down tme wherea n other applcaton the requency o nterrupton alo mportant. A central problem n the optmzaton o the overall cot developng a relablty model that relect the actor that eect the relablty. Some approache or contructng th model are dcued n [8-10]. A utable approach would be to ue o Markov Procee to model thee conguraton. In th approach, varou tate reultng rom the alure and repar or replacement o component are dented. Then the ntertate tranton rate are agned that are uncton o the alure rate and repar tme o the component. The tate probablte and other ndce are then calculated ung the mathematcal technque decrbed n [1-2]. 2. State Tranton Dagram: The tate tranton dagram or the topology hown n Fg. 2 (a) hown n Fg. 3. The varou parameter are: = Falure rate o one leg o the nverter. µ = Repar rate when only one leg aled. 1 µ = Repar rate when two leg are aled. 2 µ = Repar rate when three leg are aled. 3 µ = Repar rate when alure detected a a reult o npecton beore the conguraton alure. I In developng th tate tranton model, ollowng aumpton are made: 1. Once the conguraton aled, there are no more component alure a the conguraton wll not be powered. 2. There ome proce o npecton or detecton o alure even beore the conguraton completely down. 3. The conguraton ether up or down. It, however, poble to model degraded mode o alure. 8

Vdc/2 Vdc/2 + - + - o 1 3 5 4 6 2 A Vdc/2 + - o + Vdc/2 ' - + Vdc/2 - o'' + Vdc/2-7 9 11 10 12 8 13 15 17 16 18 14 1,4 o B 2, 3,6 5 13,1 7,10 o'' 6 o 14,17 8,11 ' 15,18 9,12 b) C A B C a) Fg. 2. (a) Detaled connecton dagram o the propoed modular nverter topology; (b) Vector dagram wth wtch par number The tate equaton can be wrtten ung the concept o requency balance [1, 2]: 9P1 µ 1P2 µ I P3 µ IP4 µ 2P5 µ 3P6 0 3 P 1+ µ 1P2 0 ( µ ) 3 6 P 1 + I + 8 P 0 ( µ ) 4 = (24) = (25) = (26) P 3 + I + 7 P = 0 (27) 7 P 3+ µ 2P5 0 7 P 4+ µ 3P6 0 = (28) = (29) Where P the probablty o tate. Any ve o the above equaton wth the ollowng equaton can be olved to nd the tate probablte, P 1 + P2 + P3 + P4 + P5 + P6 = 1 (30) The varou relablty ndce can now be ound ung the ollowng relatonhp [1]: Sytem Unavalablty = P 2 + P5 + P6 9

Frequency o ytem alure = 3 P1 + 7P3 + 7P4 Mean Down Tme = Sytem unavalablty / Frequency o ytem alure. µ 2 µ 3 1 µ I ALL COMPONENTS UP 3 (1,4) OR (9,12) OR (14,17) DOWN 2 µ 1 µ I 6 3 (2,5) OR (3,6) OR (7,10) OR (8,11) OR (13,16) OR (15,18) DOWN 5 ANY COMBINATION OF TWO LEGS OTHER THAN IN STATE 4 7 4 [(3,6)&(7,10)] OR [(2,5)& (13,16)] OR [(8,11)&(15,18) DOWN 6 ONE MORE LEG THAN IN STATE 4, DOWN 7 CONFIGURATION FAILURE Fg. 3. State tranton dagram o conguraton decrbed n Fg. 2 (a) STANDBY POWER SYSTEM INCLUDING UPS Sytem Decrpton The tet ytem hown n Fg. 4. Th ytem the ame a the one analyzed n [8, 9]. The power normally uppled rom utlty nput power through the UPS. A ynchronzed bypa and tatc wtch protect the crtcal load n the event o an nverter alure. I voltage lot to the crtcal load, the STS reetablhe voltage n le than one quarter o a cycle. Th condered contnuou power or mot load. When the generator are n tandby mode, ther alure reman undetected except durng perodc npecton. Thereore, whle tartng, there probablty p that a generator may al to tart. Only one generatng unt taken out or planned mantenance. I a generator al whle the other on planned mantenance, t poble to accelerate the mantenance on the econd generator by a actor oα. I power al at bu A, the battery can utan the load or upto 4 hour. 10

UTILITY INPUT POWER SYNCHRONIZED BYPASS ATS BUS A RECTIFIER INVERTER STS BATTERY CRITICAL LOAD BUS G G Fg. 4. Parallel uppled non-redundant unnterruptble power upply TABLE I lt the data pertanng to the ytem. Smulaton Model The lowchart hown n Fg. 1 wa mplemented. In computng the tme to the next event, the method decrbed earler wa ued, but the ollowng dependence were alo ncluded: 1. A generator cannot al whle the utlty upply up 2. A generator cannot be taken out or mantenance the utlty down or the other generator down or under mantenance 3. The UPS cannot be taken out or mantenance unle power avalable at bu A TABLE I: SYSTEM DATA Equpment/Supply (/y) r (h/) Utlty Supply, ngle crcut 0.53700 5.66 Generator (per hour o ue) 0.00536 478.00 Inverter 1.25400 107.00 Recter 0.03800 39.00 ATS 0.00600 5.00 STS 0.08760 24.00 Battery 0.03130 24.00 Equpment Mantenance req (/y) dur (h) Generator 1.00 10.00 UPS 1.00 4.00 Other data: Battery can upply load or 4.0 h Common mode alure o generator ( cm ) 0.0 Acceleraton actor or planned mantenance o (α) 2.0 generator Probablty o alure to tart a generator (p ) 0.015 11

The alure o a generator to tart wa modeled a ollow. A random number z wa generated, 0 z 1. I z p, the generator wa aumed to al. Reult In th ecton the ndce obtaned rom mulaton are compared wth thoe obtaned analytcally, ung cut et analy, n [8]. Cut Set Analy The relablty ndce o the tet ytem have been analytcally evaluated, ung the cut et approach, n [5]. A ummary o the oluton tep and the reult wll be preented here, to provde a ba or valdaton o the mulaton technque. Frt, the combnaton o the utlty upply and the two generator analyzed or alure mode. Fg. 5 how the poble tate th combnaton can aume, and the tranton rate between thee tate. Baed on thee tranton rate, the probablte and requence o occurrence o the aled tate are determned. Th enable computaton o the alure rate and duraton o the utlty-generator ubytem, whch are determned to be p = 0.001576/ y and rp = 5. 443h The next tep nvolve computaton o the rate and duraton o power alure at bu A: A = p+ ATS = 0.007576/ y prp+ ATSrATS ra = = 5. 092h p+ ATS The remander o the cut et analy perormed a hown n TABLE II. The ytem ndce,.e., rate and duraton o power lo at the Crtcal Load Bu (CLB) are determned to be = CLB = 0.005225/ cut et y r r CLB = = 9. 648h 12

µ 9 S, 1 GEN MT 1 GEN READY TO START 2 g 1 r mg 4 S, BOTH GENS READY TO START µ p p _ 1 S, BOTH GENS UP 1 r mg µ g 10 S, 1 GEN MT OTHER DN µ 1 r mg µ g S, 1 GEN DN 5 2 µ g µ _ 2 p p _ p µ g 2 g _ 2 S, 1 GEN DN OTHER UP 2 µ g α r mg _ p _ 7 S, 1 GEN MT OTHER GEN UP µ g S, BOTH GENS DN 6 µ p p p g _ 3 S, BOTH GENS DN cm DOWN g _ 8 S, 1 GEN MT OTHER GEN DN p Fg. 5. State tranton dagram or utlty upply and generator I all dtrbuton are aumed exponental, then the tandard devaton o all the up tme and down tme would equal the correpondng mean up tme and down tme. Th mple that the tandard devaton o the alure rate would alo equal the correpondng mean computed. TABLE II: FREQUENCY AND DURATION OF POWER LOSS AT CRITICAL LOAD BUS (CLB) Cut Set (/y) r (h/) r Power lo at bu A > 4 h 0.003454 5.092 0.017588 Power lo at bu A (0.007576, 5.092) and Falure o [Inverter or battery or STS] (1.3729, 99.812) 0.000125 4.845 0.000606 Mantenance on UPS (1.0, 4) and Power lo at bu A (0.007576, 5.092) Inverter alure(1.254, 107.0) and STS alure(0.0876, 24.0) 0.000003 2.240 0.000007 0.001643 19.603 0.032208 Σ 0.005225 0.050409 Smulaton Reult The mulaton method decrbed earler wa ued to compute the ollowng tattc or the tet ytem: 1. Frequency p and duraton r p o alure o the utlty-generator ubytem (Fg. 4); the tandard devaton o p and r p. 13

2. Frequency A and duraton r A o power alure at bu A; the tandard devaton o A and r A. 3. Frequency CLB and duraton r CLB o power alure at the crtcal load bu; the tandard devaton o CLB and r CLB. 4. Data or the Probablty Ma Functon o the number o ytem alure per year, N. 5. Data or the Probablty Dtrbuton Functon o the ytem down tme, T. TABLE III compare the ndce obtaned rom mulaton wth thoe obtaned analytcally. TABLE III: COMPARISON OF SIMULATED AND CALCULATED INDICES Smulated Calculated Index Mean SD Mean SD p (/y) 0.001352 0.001497 0.001576 0.001576 r p (h/) 5.028 5.503 5.443 5.443 A (/y) 0.007193 0.006851 0.007576 0.007576 r A (h/) 5.216 5.201 5.092 5.092 CLB (/y) 0.004868 0.004888 0.005225 0.005225 r CLB (h/) 9.7135 13.3312 9.648 9.648 TABLE IV compare the PMF o the number o ytem alure per year. Now or exponentally dtrbuted up tme the alure are Poon dtrbuted,.e., P( N k CLB CLBe = k) = (31) k! Th equaton ued to generate the calculated data or the PMF o N, n TABLE IV. Fg. 6 compare the PDF o the ytem down tme. For exponentally dtrbuted down tme T, the PDF gven by t/ r CLB F( t) = P( T t) = 1 e (32) Th equaton ued to generate the calculated data or the PDF o T, or Fg 5. 14

TABLE IV: PMF OF NUMBER OF FAILURES PER YEAR P( N = k) k 0 1 2 3 4 Smulated 0.9952 0.4825 10-2 0.2167 10-2 0.0000 0.0000 Calculated 0.9947 0.5198 10-2 0.1358 10-2 0.2365 10-7 0.3089 10-10 CONCLUDING REMARKS Th paper ha decrbed how a gven conguraton can be analyzed or producng a quanttatve relablty ndex. The technque decrbed can help to quanty relablty o alternate degn conguraton or the ame degn but wth derent qualty o component, relectng derent alure rate.. Then the relablty can be traded o wth cot to produce conguraton whch have relablty that that conumer prepared to pay or. 1.00 0.80 probablty 0.60 0.40 0.20 0.00 0 20 40 60 80 100 120 down tme (hour) mulated calculated Fg. 6. Probablty dtrbuton uncton o down tme 15

REFERENCES 1. C. Sngh, R. Bllnton, Sytem Relablty Modellng and Evaluaton, Hutchnon Educatonal, London, England, 1977. 2. B. S. Dhllon, C. Sngh, Engneerng Relablty: New Tool and Applcaton, John Wley, New York, 1981. 3. E. Cengelc, P. Enjet, C. Sngh, F. Blaabjerg, J. K. Pederon, New Medum Voltage PWM Inverter Topologe or Adjutable Speed AC Motor Drve Sytem, Proceedng o 1998 Appled Power Electronc Conerence and Expoton, pp. 565-571. 4. IEEE Standard 446-1995, IEEE Recommended Practce or Emergency and Standby Power Sytem or Indutral and Commercal Applcaton. 5. Alexander Kuko, Emergency/Standby Power Sytem, McGraw Hll Book Company, New York, 1989. 6. IEEE Standard 493-1997, Degn o Relable Indutral and Commercal Power Sytem. 7. C. Sngh, A. D. Patton, Relablty Evaluaton o Emergency and Standby Power Sytem, IEEE Tranacton on Indutry Applcaton, vol. 21, no. 2, Mar/Apr 1985. 8. C. Sngh, N. Gubbala, N. Gubbala, Relablty Analy o Electrc Supply Includng Standby Generator and an Unnterruptble Power Supply Sytem, IEEE Tranacton on Indutry Applcaton, vol. 30, no. 5, Sep/Oct 1994. 9. C. Sngh, J. Mtra, Relablty Analy o Emergency and Standby Power Sytem, IEEE Indutry Applcaton Socety Magazne, vol. 3, no. 5, pp. 41-47, Sept./Oct. 1997 16