A stochastic model is described and fitted to a set of data on maximum
crack depths in welds on a large system. The initial model is a two-s
tage empirical one, in that it models crack initiation and crack growt
h with time, and concentrates on explaining the data rather than on th
e physical processes which give rise to cracks. The model is fitted to
the data by the method of maximum likelihood, and the goodness-of-fit
is considered. Possible extensions of the basic model are also discus
sed, including a three stage model, in which the delay before which a
defect becomes 'visible' is also considered. The main purpose of the m
odelling is to consider the prediction of crack growth in time and the
consequences of carrying out periodical inspections, with the aim of
minimizing the long term total cost per unit time of maintaining struc
tural integrity. (C) 1996 Elsevier Science Limited.