1984)), this classification scheme does not specify how 
to proceed in designing a network. 
3.2 Design-by-Simulation Strategy 
The design-by-simulation strategy (see Figure 3) pro- 
vides a more concrete conceptualisation of how human 
experts solve the network design problem. The following 
characteristics are worthy of note: 
Accuracy 
criteria 
dE 
/ 
  
  
  
  
  
  
     
    
  
  
  
  
  
  
Select observation 
scheme 
Y 
Initial design: wl 
approximate imaging Y 
geometry and ZOD [FOD] | SOD | 
Y Y 
  
  
  
  
Design evaluation: compute variance- 
covariance matrix (Q,,) of object points 
  
  
  
No 
  
  
  
  
Does Qyx 
simply need 
calin 
      
Figure 3 À flow-diagram representation of network de- 
sign-by-simulation (after Fraser, 1984). 
The dataflow in this strategy is sequential and ad- 
dresses the three design tasks - ZOD, FOD and 
SOD, in an ordered (as opposed to simultaneous) 
fashion. In knowledge-engineering terms, this da- 
taflow constitutes control knowledge and is pro- 
cedural in form. 
The simulation strategy is heuristic in nature, hav- 
ing been developed out of the experiences of ex- 
perts. The complexity of the task is reduced by 
initially configuring a first approximation to a 
suitable imaging geometry. Should this configura- 
tion fail to meet the criteria, FOD or SOD meas- 
ures are employed to iteratively refine the 
network, or indeed a redesign may be attempted 
(Fraser, 1984). 
Simulation is the most practical method of de- 
signing close-range photogrammetric networks; 
analytical (direct) design methods have yet to be 
proven practical (Fraser, 1987). 
* Successful application of this strategy requires 
448 
expert decision-making at the individual step lev- 
el, as suggested by Fraser (1984), “...factors such 
as previous experience and intuition will play a 
central role in network optimization". Heuristics 
for network diagnosis, in particular with respect 
to the step "criteria satisfied?", are exemplified in 
Section 3.4 below. 
3.3 Generic Problem Solving Processes in Network 
Design 
The design-by-simulation strategy can be also conceptu- 
alised in terms of generic problem-solving processes. In 
Figure 4, the steps of the design-by-simulation strategy 
have been replaced by a design process, an algorithmic 
step involving the computation of network performance 
measures (e.g. by bundle adjustment), the identification 
of network faults through diagnosis, and a prescribing 
process entailing the design of corrections to a network 
to overcome these faults. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
measurement 
criteria 
Design Prescribe 
(initial design) SE (FOD and SOD 
corrections) 
' A 
Algorithm Diagnosis 
(compute performance P-| (evaluate design and 
measures) identify faults) 
À 
satisfactory 
network design 
Figure 4 Conceptualizing network design in terms of ge- 
neric problem-solving processes. 
Design is the development of configurations of 
objects, entities or items based on set of problem 
constraints. Design systems often use synthesis, 
to generate partial solutions, and simulation, to 
verify or test these solutions (Waterman, 1986). 
This latter function entails either diagnosis of de- 
sign faults or critical appraisal of design quality 
(Oxman and Gero, 1987). 
* The computation of network performance meas- 
ures (e.g. the precision and reliability of object 
point determination) is largely based on procedur- 
al knowledge in the form of algorithms. For in- 
stance, the self-calibrating bundle adjustment is 
based on formal mathematical and statistical 
models. Knowledge in procedural form is best im- 
plemented (as is already the case) in program- 
ming languages such as C or Fortran. 
Diagnosis systems infer faults in a systems (e.g. 
photogrammetric network) functioning from ob-