The present invention generally relates to failure prediction of an infrastructure (110). A failure likelihood for one or more components of an infrastructure (110) is determined. History data (210) representing prior failures of the components of the infrastructure (110) is applied (230-250) to a Bayesian nonparametric statistical model using a beta process. Then the failure likelihood of one or more components of the infrastructure from the Bayesian nonparametric statistical model is estimated (270). Aspects of the invention include computer-implemented methods (200, 300), software and computer systems (100).