Neural network validation: an illustration from the monitoring of multi-phase flows

One of the key factors limiting the use of neural networks in many industrial applications has been the difficulty of demonstrating that a trained network will continue to generate reliable outputs once it is in routine use. An important potential source of errors arises from novel input data, that is input data which differs significantly from the data used to train the network. In this paper, the authors investigate the relation between the degree of novelty of input data and the corresponding reliability of the outputs from the network. They describe a quantitative procedure for assessing novelty, and demonstrate its performance using an application involving the monitoring of oil flow in multiphase pipelines.