Adaptive k-Nearest-Neighbor Method for Process Monitoring

In this paper, an adaptive process monitoring method based on the k-nearest neighbor rule (k-NN) is proposed to address the issues arising from nonlinearity, insufficient training data, and time-varying behaviors. Instead of recursively updating every measurement for adaptation, a distance-based updating rule is applied to search target prototypes, thus reducing the computational load for online implementation. Furthermore, for fault identification, a subspace greedy search is also introduced to formulate the complete monitoring system. The approach searches for the combination of variables (subspace) which has the greatest contribution to the discriminant between normal data and faulty data (explanatory subspace). The Tennessee Eastman Process (TEP) and data from an industrial pyrolysis reactor are considered to evaluate the performance of the proposed approach as compared with conventional methods.