Models and inference for corrosion pit depth data (Q2488459)

A stochastic model for the process of soil corrosion of pipelines is proposed. It is assumed that the film thickness of metal oxides \(a(t)\) follows the power-law: \(da/dt=C s(t)/a^\rho\), where \(t\) is the time, \(C\) and \(\rho\) are constants, \(s(t)\) is the stress amplitude at time \(t\), \(s(t)=\sum_{i=1}^\infty u_i\delta(t-t_i)\), where \(t_i\) are occurrence times of some counting process and \(u_i\) are i.i.d. exponential r.v.s. Approximations for the pit depth distribution in this model are considered. The generalized extreme value, new power normal and the power Gumbel (PG) distributions were used for fitting the distribution of real data on maximal pit depth in different subdomains of the pipe. The authors conclusion is that the PG has the best performance.