Adaptive Metaheuristic Scheme for Generalized Multiple Abnormality Detection in a Reservoir Pipeline Valve System

This paper presents mathematical and algorithmic developments related to general abnormality (multiple leakages and multiple partial blockages) detection in a simple pipeline system. Formulations for leakages and blockages were combined and reformulated to address general abnormalities efficiently in the frequency domain. Unsteady friction effects on laminar and turbulent flow conditions were considered during formulation development using 2D frequency-dependent and 1D acceleration-based models, respectively. The developed formula was tested in terms of model parsimony, computational accuracy, and flexibility for superposition in abnormality representation. Based on the proposed formulation, a novel multiple abnormality detection algorithm, called the adaptive metaheuristic scheme (AMS), was developed by integrating a stepwise genetic algorithm. The application of the developed method to a hypothetical pipeline system demonstrated the potential of the AMS for predicting general features of abnormality, even without access to prior information regarding the number and distribution of abnormalities. The developed method demonstrated robustness for the prediction of abnormality distributions and reliability, even in noise-contaminated signals. The adaptive predictability of the AMS can be characterized by not only its robustness for unknown multiple abnormality features but also its self-diagnostic capabilities during the calibration procedure.