Downloaded from https://www.everyspec.com AMCP 706-197 APPENDIX A* Optimization of System Reliability by a New Nonlinear Integer Programming Procedure6 Abruut-This paper viewmas a iseful pnosdre of solving aoarw subject to thr "onstraints intege prolrammiug problems. It rld, firt, a pseado.solutioa to the problem. a if the variables wmee contuoutL Then it use a te wc1X, .... x.) = 1, m () " in the neoboulicotho ofewpso -solutio to rod the optmum. The 1(X2 effectivenes of the method is show. with a IS.,ariable probleu, which sequreus about I days FORTRAN programmirg effort and 8 Xcon, x r i= 1. 2 ....
n must be positive integers (2) of computer time for its solution on an IBM 370/165 dilital computer. The constraint functions gi need not be linear and the number Rar : of inequality constraints m need not be less than n. A proce- " Purpow: Widen state-ol-thse Mr. Special math needed for exptanatious: None dure involving three steps is proposed. Special math needed for results: None Results useful to: Desig and Wibility engineers progammene. rs SOLUTION TO THE GENERAL PROBLEM I IT TStep 1:
Solution to the Pseudo-Problem ~~INTRODUCTION NCRESIN reiabiityby he itroucton o reundncy Relax the condition of requiring each x, to be integer and I NCREASING reliability by the introduction of redundancy solve the maximization problem as if the variables were con- Rels well onditio Hoefr theiin proble of how to integerizn tinuous. Only an approximate solution is necessary to this the reliability through the selection of redundancy has not yet pseudo-problem. been edequately solved. Tillman and Liittschwager [I I pre- sented an integer programming formula,,on for the solution Step 2: Filling in the slack by ste.7est ascent ofreliability problems.
The method requires transformation of the objective function and introduction of auxiliary variables. Take the values of x, obtained in Step I and convert them Misra 12] discus,,es the overall applicability of integer program- to integers by truncation (toward zero) so that the inequality ing approach to solving reliability proSlems, later Misra 131 in- constraints (1) are satisfied. troduces the use of Lagrange multipliers and the Maximum c a There may now be adequate slack in (1) to allow an increase principle to solve reliability optimization problems.
Sharma in at least one of the x, Therefore, ateptoirentec and Venkateswaran [4] presented a simpler method with no lc he x ttemt to increment each assurance of obtaining the true pmum. Baneree and x 1 by check to see if (1) is satisfied, and increment only the asaane [of useain the arnge omupli er aroa n e x, which gives the greatest contribution to the maximization of Rajamani (51 use the Lagrange multiplier approach to solve [ otneti iln Jc ni ox a eiceetd the reliability problem to yield optimum or near optimum f. Continue this filling c f lack until no can be incremented results. Misra and Sharma [61 classified the methods into two without v.latingat least one of the constraints.
groups, one which includes methods which require simple form. ulation and yield approximate results and the other which Step 3: Systematic exchange of variables includes methods which are complicated but yield an exact in- teger solution to the problem. These authors then provide a Carry out n(n - I) tests whereby one variable is incremented geometric programming formulation for the reliability problem by I and the others ar decremented by I in turn For example, which gives an approximate answer, suppose x is incremented toxn + I. Now decrement x 2 to The purpose of this poper is to present a method which is x 2 - I and check whether inequalities are satisfied. Ifso.
then easy to formulate and which gives an optimum for the rel. calculate the corresponding value offand compare that value to ability optimization problem. Although there is no assurance the maximum fin Step 2. If the most recently calculated fis of obtaining the global optimum, in practical problems the greater, then retain in the memory the fact that x, incremented method will come very close to finding the global optimutn. by I and xzdecremented by I gives a better value. However, before making a change in this variable, continue through the entire cycle up to x,. Then choose the set x, which has given PROBLEM FORMULATION the greatest value forf.
Perform the cycle by incrementingx 2 Maximize a nonlinear functiot of n variables denoted by and continue with x 3 , x 4 , etc. up to x,. In total. there are thus a maximum ofn(n - I) tests to be done. The set giving Jxt. x 2 . ... x,) the largest value off is retained as the optimum. *Copyrighted 1975 by Institute of Electrical and Electronics Engineers, Inc. Reprinted with pdrmision. 12-22