Capacity expansion of a deteriorating facility under uncertainty

R EVUE FRANÇAISE D ’ AUTOMATIQUE , D ’ INFORMATIQUE ET DE RECHERCHE OPÉRATIONNELLE . R ECHERCHE OPÉRATIONNELLE C HARLES S. TAPIERO Capacity expansion of a deteriorating facility under uncertainty Revue française d’automatique, d’informatique et de recherche opérationnelle. Recherche opérationnelle, tome 13, no 1 (1979), p. 55-66. <http://www.numdam.org/item?id=RO_1979__13_1_55_0> © AFCET, 1979, tous droits réservés. L’accès aux archives de la revue « Revue française d’automatique, d’informatique et de recherche opérationnelle. Recherche opérationnelle » implique l’accord avec les conditions générales d’utilisation (http://www.numdam.org/ legal.php). Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright. Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques http://www.numdam.org/ R.A.I.R.O. Recherche opérationnelle/Opérations Research (vol. 13, n° 1, février 1979, p. 55 à 66) CAPACITY EXPANSION OF A DETERIORATING FACILITY UNDER UNCERTAINTY (*) by Charles S. TAPIERO O Abstract. — An optimum single capacity expansion of a deteriorating facility is determined when future demandfor the facility is given by an évolution ofprobability distributions. Several probabilistic détérioration processes are considered and the probability distribution ofafacility's extinction isfound. For practical applications, it is suggested that simulation be used in determining the costs and benefits of two or more expansion plans. 1. INTRODUCTION (2) Capacity expansions are strategie décisions involving the highest levels of management. They are designed to fulfill one or both of the following purposes: (1) replace deteriorating facilities or improve the efficiency of production processes by a newer technology; (2) meet anticipated demand growth for production services. Capacity expansion involves present outlays measurable only in probabilistic terms with returns expected in the future. For this reason, problems of capacity expansion require a careful évaluation of the risk implications of a particular expansion. Approaches to capacity expansion have ranged from non-linear programming to heuristic methods. Foremost among these studies is the dynamic programming approach (Manne [7], Manne and Veinott [9], Erlenkotter [2, 3, 4], Sengupta and Fox [11], Tapiero [13]) which provides an optimum séquence of capacity expansion when a future demand is known. These approaches, however, assume a demand growth and non-depreciating facilities. Over time, facilities have a deteriorating productive capacity while demand (*) Reçu novembre 1977. O Hebrew University, Graduate School of Business. (2) This research was supported in part by a grant from the Graduate School of Business of Columbia University. R.A.I.R.O. Recherche opérationnelle /Opérations Research, 0399/0842/1979/55/$ 1.00 © Bordas-Dunod. 56 C.S. TÂPIERO forecasts can at best be characterized by probability distributions. Manne [8], for example, suggested a Brownian motion model for describing a random évolution of demand for a product and obtained an optimum capacity expansion for a facility of the non-depreciating type. The purpose of this paper is to consider a problem of capacity expansion of deteriorating facilities under uncertainty. The foliowing assumptions are made: (1) the détérioration rate is described by a stochastic process; (2) a demand forecast is given by an évolution of probability distributions. Specifically, we assume as given the mean and variance évolutions of the forecasts; (3) only one plant is built at the initial planning time; (4) the riskless discount rate r is given. An optimum capacity expansion is obtained when the planning time T(used in Computing the costs and benefits of the expansion) is given. The cost criterion is assumed to consist of an expansion cost, an excess capacity and a capacity shortage cost. When a facility is to be evaluated over its lifetime, we compute the probability of facility's extinction and use it as a planning time. For simplicity, we have assumed that under such circumstances, the cost criterion consists of the expansion cost and a net return proportional to a facility's productive capacity at time t. If the facility capacity expansion is K and it has a détérioration rate 8, we find that its expected life is approximately (In K)/62. Using this expected life as a planning time, an optimum capacity expansion is determined. For practical purposes, when more realistic and complex assumptions lead to analytically intractable solutions, simulation can be used to compare the economie and risk implications to a firm of two or more expansions. In this case, the problem is not to solve for an optimum capacity expansion but to compare the economie and risk properties of two (or more) competing capacity expansion alternatives. For démonstration purposes, a simulation comparing two such expansions is included. Such a simulation approach can be used only after the theoretical probability structures of capacity détérioration processes has been found. In other words, simulation becomes an essential tool in comparing capacity expansions when the underlying probability processes of capacity and demand are established. The first part of the paper shows how such probability processes can be found and how they can be manipulated to yield an optimum capacity expansion when the cost function is a simple one. 2. THE CAPACITY OF A DETERIORATING FACILITY Assume that at time f = 0 the present time, a capacity expansion K is implemented whose cost is G(K). Let A- be the capacity at a future time t and R.A.l.R.O. Recherche opérationnelle/Opérations Research CAPACITY EXPANSION 57 assume that the facility détériorâtes randomly where S is the détérioration rate. If P(x, t) is the probability of a capacity x at time t, and if the détérioration process is a stochastic process described by a random walk where A t§x = probability of losing to détérioration one unit of capacity in a small time interval At, then, a probabilistic évolutions for the capacity x at time t is given by (3): -P(x, t)5(x), dt (2.1) P(x = K, to)=l. When 8 is a constant, (2.1) describes a simple death process whose solution is a binomial probability distribution P(x,t) = (2.2) x p(5, t) = e -et. The probability generating function of (2.1) is well known and given by (4): 6(2, 0 = {! + ( * - ! ) « - * } * (2-3) with Therefore, if we expand a productive capacity by K units at time t = 0, and if the détérioration rate ô is constant, the probability distribution of having a capacity of x units at time t is given by the binomial distribution (2.2). When the détérioration rate 5 is probabilistic, it is necessary to randomize the random walk (2.1). However, since (2.1) has a solution given by (2.2), a randomized détériorations process as in (2.1) has a probability given by the mixture distributions of (2.2). (3) This is in fact the welî known death-process in queueing theory, with initial condition given by the K-the expansion, and a death rate equalling the capacity détérioration rate. (4) The probability generating function is defined by Q{z,t)= t P(x,t)zx x= 0 and for the binomial probability distribution (2.2) is well known. vol. 13, n° 1, février 1979 58 C. S. TAPIERO Specifically, \Gtf(e~~bt) be the probabüity distribution of exp( — ), then the probabüity distribution of the capacity x at time t is given by (2.4) Wjo' A gênerai approximate solution for (2.4) is given by Hald [5], When we assume that exp( —St) has a bèta probabüity distribution, we can show, that the probabüity distribution of x at time t has a Polya — Eggenberger distribution [1, 10]. That is, if (2.5) a(t)>Q, b{t)>0. Then inserting (2.5) into (2.4) and solving for P(x, t) yields (6): which has a mean given by: (2.7) E{x(t)}=Ka(t)/[a and a variance given by: K(K-l)a(t)b(t) (2.8) Thus, the capacity on hand at time t has a probabüity distribution given by (2.6) whose mean and variance is given by (2.7) and (2.8). When S is a constant, and when K is large (K>30), a normal approximation to (2.2) is accurate, and the probabüity distribution of a capacity x at time t is given by: (2.9) This approximation is useful in Computing an optimum capacity expansion when the demand for products is also normally distributed. (5) That is, a probabüity distribution whose parameter is also given by a probabüity distribution. (6) Such commutations are straightforward if we note that (2.5) and (2.4) yield a Beta intégral. A solution to this problem is, however, pointed out in Johnson and Kotz [6]. R.AXR.O. Recherche opérationnelle/Opérations Research 59 CAPACITY EXPANSION When a facility's détérioration rate is also function of its age, ô is a time variant and (2.1) is a non-homogeneous stochastic process, whose probability generating function can be shown to be (2.10) where which corresponds to a binomial probability distribution whose probability parameter is J'miX (2.11) TABLE I Probability P.G.F. Variance a 2 (t) Mean \i(t) Assumption about détérioration: ö = Ke~ht(l-e-hc) Ke~ht Assumption about détérioration: f {e r 5r )~Beta Ka(t)b(t) Ka(t) a(t) + b(t) K(K-l)a(t)b(t) Assumption about détérioration: normal approximation (large K) Ke~ Assumption about détérioration: Ô (x) = time variant Xe vol. 13, n° 1, février 1979 S(T)rfT f - 8(T)dTJ 60 C. S. TAPIERO when S(t) is time invariant, then (2.11) reduces to exp(-ôt) which is the case of (2.2). Equations (2.2), (2.5), (2.9) and (2.11) provide four stochastic models for capacity détérioration which may be used to compute the probability distributions of a facility's productive life. These distributions will be used to establish upper bounds on the economie planning time of a facility's expansion. Furthermore, when the future demand for a product is probabilistically given, probability distribution of excess and capacity shortage can be computed. To simplify our présentations the probability results are summarized in table 1. 3. OPTIMUM CAPACITY EXPANSION OF A DETERIORATING FACILITY We shall first assume that the planning time used to compute the economie value of a capacity expansion is given by T (where Tis smaller than the facility's life). The capacity expansion size is chosen to minimize the expected cost of / G(K) + T \ T' if x(t)-B(t)^o, (3.1) T x(t)-D(t)<Q. r=n where: G (K), the capacity expansion cost; x(t), the capacity at time t — a random variable; D {t)} the demand at time t — a random variable; r, the riskless discount rate; T, the planning time; cx, cost per unit excess capacity; c2, cost per unit capacity shortage. This criterion will be simplified further when we consider a probabilistic planning time. The expected value of (3.1) is given by: G(X)+ £ r)-' f" f z{t)g(z{t))dz(t) C l (l+r)-' t =Q JJ 0 z{t)g{z{t))£{t)} where z(t) = x(t)-D{t) and g(z(i)) is the probability distribution of z(t). If we assume 8 constant and use the normal approximation to (2.2) and if the demand is normally distributed with mean \i(t) and variance <J2(t), then z(t) has also a normal probability distribution. Specifically, the time variant mean and variance of this distribution is: R.A.I.R.O. Recherche opérationnelle/Opérations Research CAPACITY EXPANSION 61 Also, using the normal distribution equality: ,(0). (3-4) where ƒ and F are the standard and cumulative standard normal distribution. Since where 2 f" 2 erf(tt)= — e" f dt. s/n J o But since erf( — u)^ — erf(w), we obtain, in équation (3.4): r a z 2 (0}. (3.5) Equation (3.2) together with (3.5) can be written dropping the time subscript, as: a, • T + X c 2 (l+r)-^ z . (3.6) t=o A derivative of (3.6) with respect to K yields one équation in one unknown (3.7) which may be solved numerically for K, or by trial and error. + ê?e-*n*\„ r^_J, nAl-e-^ >e-^0. vol. 13, n° 1, février 1979 (3.7) 62 C. S. TAPIERO For example, assume a normally distributed demand with time variant mean and variance given by |a(0=1004-10t + 5$in(nt/3) and u2(t) = l0t, also G(K)= 10KOHO, cx = 1.0 and c2 = 3.0, r = 0.06 and T-30. An optimum capacity expansion is then given by K* (5 = 0.10): K* = 300 units From further numerical computations (see table II) we note that large détérioration rates do not imply smaller expansions. The size of the expansion is as indicated in (3.6) a function of knowledge of the demand forecast [i. e. the magnitude of a 2 (*)]. When the forecasted demand is not known, a 2 (t) is very large and a Taylor's series approximation to the exponentials in (3.6) is acceptable and yields: TABLE II Optimum Capacity Expansions r ô 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 0.14 0.15 370 430 480 520 590 "— — 590 560 540 530 520 500 480 460 340 370 410 440 460 460 450 450 450 440 430 420 400 380 370 300 330 350 360 360 370 370 370 370 370 360 350 340 330 320 270 280 290 300 310 310 320 310 310 300 290 280 270 260 260 230 250 260 260 270 270 270 270 260 250 250 240 240 240 230 210 220 230 240 240 240 230 230 230 220 220 220 220 210 210 200 210 210 210 210 210 210 200 200 200 200 200 200 200 200 Also erf (e) « 0 for E very small. Thus (3.6) can be written as: t=o 2% . (3.8) Subtracting (3.8) from (3.6) we obtain: - ^ 2 / a 2 ) - l ] 1, (3.9) J R.A.I.R.O. Recherche opérationnelle/Opérations Research 63 CAPACITY EXPANSION which is the value (in an expected cost sense and given the détérioration rate) of a demandas for ecast. Since exp(-u^/a^)< 1, the larger the variance the larger the value of the demand forecast. When the planning time T is not specifically given, we can compute the capacity's expansion cost over the facilities' life time. That is, the planning time T is a random variable expressing the probability that x, the capacity, equals zero. For the four cases considered in this paper [équations (2.2), (2.6) and (2.9)] the probability that the capacity is equal to zero at time T is given below: TABLE III Probabilities of capacity extinction Probabilities {l-e-*T}K ö-Const r(a(T))r(K + b(T)) r{K + a(T) + b(T)) /(e- 5 7 )-Bèta Normal approximation. . (KC~&T{1 B(û(nKlÔ(r)) 2 2ôr /Ke- &r a-e- Sr )} C -8î))l/2e X D{-fl/2)K e - y/ÏÜ l1- ° Ô (x)-time variant . } Consider for simplicity the case of a constant détérioration rate and compute the Laplace transform of the probability of extinction. Then, (3.10) = (l/5)B(s/8, K+i), where B(a, P) is the Bèta function B(s/b, vol. 13, n° 1, février 1979 (3.11) 64 C. S. TAPIERO The mean extinction time is clearly given by: 1) Ô d (3.12) 2 The mean is therefore (3.13) For K large, an approximation to (3.13) is: ^ £ ( 1 / 0 - f (1/S 2 )^j= ^-lnK. 0 i=l J1 (3.14) ° Using the mean extinction time as the facility's planning time, we consider a cost criterion given by (3.15) t=Ö where p is a net return per unit capacity x (t ) at time t. If we further approximate the problem by considering the continuo us form of (3.15) we note that the expected value of the facility is given by: G{K)~r [^ [ px(t)e~rtdtf(x(t))f{T)dx(t)dT. (3.16) Jo Jo Jo Replacing the expected value of capacity, Ke~dt, and integrating yields: \e-(r + roo à)T_^i G(K)+\ pKlJo (r + o) } f(T)dT. Since E { e~{r+b)T} is the Laplace transform of T, we obtain an expected value given by: The optimum capacity expansion is thus found by solving for K in 3G p r 1 g(r + 6 / § , K + l ) " | ] +t pK 8 g An analytical solution to this équation is very difficult, numerical results are however, fairly easy to obtain. Simplification can also be used in obtaining less R.A.I.R.O. Recherche opérationnelle /Opérations Research CAPACITY EXPANSION 65 précise but analytically tractable results. Namely, using the mean extinction time as the planning horizon, we obtain an optimum capacity expansion given by a solution of: —f - ! - _ L W - ( 5 + r ) / ô 2 - - ^ - (3 19) If the détérioration rate is small, (3.19) reduces to ^ , / r . (3.20) The analytical problems encountered in solving the optimum capacity expansion problem are, as we note, difficult ones. For this reason, simplifying assumptions were made yielding an approximate capacity expansion. For practical application, however, the capacity probability distributions defmed earlier allow the simulation of complex capacity expansion problems. Instead of solving a numerical problem to détermine the optimum size of an expansion, we may instead be able to compare the returns and risks of two capacity expansion plans. 5. CONCLUSION This paper has provided an approach to determining an optimum capacity expansion under uncertainty. We have considered, as in Smith [12], a single plant expansion but with a deteriorating capacity and a probabilistic future demand. Thus, the paper can be viewed as an extension of past studies on capacity expansion. Capacity expansions require a large commitment of resources by a firm in the expectation of future returns. The costs of such an expansion, the risks of expansion (or of not expanding) are very important in determining the size of capacity expansion. To reach a judicious décision concerning capacity expansion, it is necessary to quantify this risk and include it as a part of managerial décision criteria. This paper has reduced the mathematical problem of determining an optimum capacity expansion to an analytically tractable form. When analytical solutions cannot be obtained, the capacity probability distributions can be used to compare through simulation alternative expansion plans. REFERENCES 1. F. EGGENBERGER and G. POLYA, Uber die Statistik Verketteter Vorgange, Zeitschrift fur Angewandte Mathematik und Mechanik, Vol. 1, 1923, pp. 179-289. 2. D. ERLENKOTTER, Preinvestment Planning for Capacity Expansion: A Multi-Location Dynamic Model, Agency for International Development, New Delhi, India, 1970. vol. 13, n° 1, février 197S 66 C.S. TAPIERO 3. D. ERLENKOTTER, Sequencing Expansion Projects, Opérations Research, Vol. 21, Mardi-April, 1973. 4. D. ERLENKOTTER, Dynamic Muiti-Location Models for Capacity Expansion, Proceedings of the 1973 Joint Automatic Control Conference, The Ohio State University, Columbus, Ohio, June 20-22, 1973. 5. A. HALD, The Mixed Binomial Distribution and the Posterior Distribution, J. Roy. Statist. Soc, Series B, Vol. 30, 1968, pp. 359-367. 6. N. L. JOHNSON and S. KOTZ, Discrete Distributions, New York, Houghton Mifflin Company, 1969. 7. A. S. MANNE (Ed.), Investments for Capacity Expansion — Size, Location and Time Phasing, Cambridge, Mass. M.I.T. Press, 1967. 8. A. S. MANNE, Capacity Expansion and Probabilistic Growth, Econometrica, Vol. 29, 1961, pp. 632-649. 9. A. S. MANNE and A. F. VEINOTT, Jr., Optimal Plant Size with Arbitrary Increasing Time Paths of Demand, in [7], 1967, pp. 178-192. 10. G. POLYA, Sur Quelques Points de la Théorie des Probabilités, Ann. Inst. H.-Poincaré, Vol. 1, 1930, pp. 117-161. 11. J. K. SENGUPTA and K. A. Fox, Optimization Techniques in Quantitative Economie Models, Chicago, North-Holland Publishing Company, 1969. 12. V. L. SMITH, Investment and Production, Harvard, Harvard University Press, 1961. 13. C. S. TAPÏERO, Optimal Capacity Expansion with Storable Output, Cahiers du Centre d'Études de Recherche Opérationnelle, 1972, pp. 159-168. R.A.LR.O. Recherche opérationnelle/Opérations Research