Prediction of Membrane Failure in a Water Purification Plant Using Nonhomogeneous Poisson Process Models

Abstract

The prediction of membrane failure in full-scale water purification plants is an important but difficult task. Although previous studies employed accelerated laboratory-scale tests of membrane failure, it is not possible to reproduce the complex operational conditions of full-scale plants. Therefore, we aimed to develop prediction models of membrane failure using actual membrane failure data. Because membrane filtration systems are repairable systems, nonhomogeneous Poisson process (NHPP) models, i.e., power law and log-linear models, were employed; the model parameters were estimated using the membrane failure data from a full-scale plant operated for 13 years. Both models were able to predict cumulative failures for forthcoming years; nonetheless, the power law model showed higher stability and narrower confidence intervals than the log-linear model. By integrating two membrane replacement criteria, namely deterioration of filtrate water quality and reduction of membrane permeability, it was possible to predict the time to replace all the membranes on a water purification plant. Finally, the NHPP models coupled with a nonparametric bootstrap method provided a method to select membrane modules for earlier replacement than others. Although the criteria for membrane replacement may vary among membrane filtration plants, the NHPP models presented in this study could be applied to any other plant with membrane failure data.