US Army Materiel Command AMCP 706-197 — AMCP 706-197 Reliability Prediction Handbook — US Army Engineering Design Guide PDF

Downloaded from https://www.everyspec.com Dynamic programming suffers from a One ought to be concerned with the re- major drawback--dimensionality. Problems gion around the optimum point. If it is very with two or three state variables may be fiat, then it makes little difference where, in solved with increasing difficulty, and solution the flat region, one chooses a solution. There with more than three state variables is very are usually many important variables, mostly difficult. This is because the functions f(h), qualitative, that are left out of the formal where h is the state vector of dimension k, analysis. These may well determine where in must be tabulat.d over a k-dimensional grid.

the fat region one chooses the solution. If each dimension has 10 subdivisions, this If there are a great many independent requires the storage of VPb numbers, which variables, it is difficult to visualize the generally exceeds the fast memory space of "space" in which the problem is to be solved. most computers for k > 4. Any increase in k The ramifications of assumptions and solu- is then quite difficult and can be accomplish- tions are difficult to grasp. Therefore, most ed only by trading memory spnee for compu- big problems ought to be reduced to a series tation time. of little ones whose meaning can be compre- 12-5LUUSJAAKLA MTHODhended.

If necessary, one can go back after the first trial solutions and modify the way Lus and Jaakola developed a very simple the little problems were formulated. method for optimization by direct-search and Perhaps the biggest difficulty of all with interval-reduction, Refs. 35 and 36. It is ex- optimizing a very large problem is that when tremely simple to program, evaluates no deiv- it is finished, people tend to be extremely atives, does not invert any matrices and can pleased and impressed. They tend to believe handle inequality constraints. Equality con- that they now know the answer to some real- - ~ straints are presumed to have been eliminated world problem. But they don't.

What they do by usual methods. know is the answer to a mathematical prob- For integer problem~s, e.g., parallel redun- lem which contains gross approximations (to dancy, Luus has extended the method, again be tractable) and which was solved with in a very simple way (both programming and guessed-at data. Since "no one" can under- conceptually), see Ref. 36. Especially for the stand the whole problem at once, there is a novice, but even for the high-powered theo- tendency to grasp the computerized solution rists, this method has a great deal of appeal like a drowning man grasping at straws. and utility. Ref. 36 is reproduced as Ap- Obviously, some very complicated prob- pendix A.

lems have been solved by optimization tech- niques. These tend to be problems where 12-6 APPLICATIONS plant process operation is quite well known, cations of complicated reliability optimiza- be t h m e onept them plety but where the magnitude of the calculation is It is difficult to find good nontrivial appli- just too much. The models themselves tend to tion in the literature. Generally, in the litera- be r their scoe ture, the analyst has to make too many unre- alistic assumptions, or picks a problem no one Some journal articles which apply optimi- in practice is really going to care about. For zation techniques are Refs. 22-33; Ref.

33 is a example, cost and weight are usually major relaHvely new approach. Anyone who wishes real constraints; but there is not a continuum to apnly optimization techniques to compli- of equipments available with reliability tabu- cated reliability engineering problems ought lated as functions of cost and weight. Solving to find professional assistance from people for optimum parallel redundancy in the pres- who are skilled in using the available comput- ence of constraints is another favorite prob- er programs. To begin from scratch is usually . lem.

But rarely are there more than a few to waste inordinate amounts of time and redundant units; so the calculations cGuld money, except that the Luus-Jaakola method easily be carried out for all feasible combinq- (par. 12-5) can be used by almost anyone-- tions. conceptually and practically it's so simple. 12-19