摘要：The failures data of tractors operating in field conditions were collected by failure tracking tests. A mathematical model is established with non-homogenous Poisson process to get the tractor failure intensity function. Bootstrap method is presented for constructing confidence regions for the failure intensity of a repairable system. Combined with fault data in tracking experiment, the early period failure intensity curves are drawn, revealing the failure regular in the initial stages of tractors. A confidence interval can be estimated by standard likelihood asymptotic theory in the parametric estimation, however in the non-parametric case it can be obtained by using the bootstrap. Comprising the MSE of cumulative intensity and cumulative number of failures between the two estimations, it is seen that the curve which was estimated by non-parameter estimation method is more realistic to the actual experiment process. Results provide the reference to improve tractor reliability, maintainability and maintenance strategy.
