This means that for economy, for which low results are preferred, the scale decreases from left to right. For the other two, it increases from left to right. We now have to combine the value of each design as a whole.
Value function – multi-attribute We evaluate each of the three competing designs in terms of economy, reliability and robustness by using a version of expression (1) p.49:
V(S)= {c1 x value of economy}+{c2 x value of reliability}+{c3 x value of robustness} (2)
All that remains is to assign values to the three weighting factors c1, c2 and c3 and these evaluations should reflect the relative severity of the corresponding pressures on the designer.
Summary There is a certain amount of caution to be observed if serious use is to be made of the multi-attribute value function. First, a theoretical condition, known as preferential independence must be met. This condition is only satisfied if preferences which hold between any pair of attributes are independent of the values of any other attributes. For instance, if a design with estimated operating costs of £25000 per year and 97 per cent availability is preferred to design with operating costs of £30000 per year and 99 per cent availability, when both have a volume tolerance of 20 per cent, then the condition requires that the same preference should apply if the volume tolerance of both were to change to 15 per cent. It is not too difficult to imagine circumstances where the condition is violated. And maybe our example is a case in point. Perhaps in the eyes of many users, if volume tolerance is reduced as above, then availability becomes more significant and the preference might be reversed.
A second difficulty lies in the choice of weighting factors. We saw in Exercise 1.3 how a casual change in the factors threw up a new best design. A mere reaction to conflicting pressures on the part of the designer may not be an adequate mechanism for valuing the factors. It may be necessary to adopt more formal methods for deriving them. These involve asking hypothetical questions of the user personnel and are very time-consuming, particularly if a large number of criteria are involved, which is usually the case. An excellent description of the techniques is given in Chapman (1980).
A final difficulty which tends to dwarf all others is that many important design criteria are not easily quantifiable. An example is maintainability, a prior measure for which is difficult to conceive, so that any numerically based choice regime is equally difficult to implement.
For these reasons, it is unlikely that a designer will use value functions for the bulk of the selection task. It is much more likely that he or she will use simple trade off techniques based on what is known of user preferences, supported by intuition, as a rough sieve to eliminate the most unlikely designs. More sophisticated value techniques will only be brought to bear once the number of competing designs has been reduced to a reasonably small number.
SOFTWARE ENGINEERING: ANALYSIS AND DESIGN
Charles Easteal
University College London
Gordon Davies
The Open University
MAIDENHEAD.BERKSHIRE.ENGLAND
<< 上一页 [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]