An intelligent computer aided defect analysis (ICADA) system, based on
artificial intelligence techniques, has been developed to identify de
sign, process or material parameters which could be responsible for th
e occurrence of defective castings in a manufacturing campaign. The da
ta on defective castings for a particular time frame, which is an inpu
t to the ICADA system, has been analysed. It was observed that a large
proportion, i.e. 50-80% of all the defective castings produced in a f
oundry, have two, three or four types of defects occurring above a thr
eshold proportion, say 10%. Also, a large number of defect types are e
ither not found at all or found in a very small proportion, with a thr
eshold value below 2%. An important feature of the ICADA system is the
recognition of this pattern in the analysis. Thirty casting defect ty
pes and a large number of causes numbering between 50 and 70 for each,
as identified in the AFS analysis of casting defects-the standard ref
erence source for a casting process-constituted the foundation for bui
lding the knowledge base. Scientific rationale underlying the formatio
n of a defect during the casting process was identified and 38 metacau
ses were coded. Process, material and design parameters which contribu
te to the metacauses were systematically examined and 112 were identif
ied as rootcauses. The interconnections between defects, metacauses an
d rootcauses were represented as a three tier structured graph and the
handling of uncertainty in the occurrence of events such as defects,
metacauses and rootcauses was achieved by Bayesian analysis. The hill
climbing search technique, associated with forward reasoning, was empl
oyed to recognize one or several root causes.