INDUSTRIAL AI & FLUID DYNAMICS

Towards parameter identification in pipeline hydraulics: integrating data-driven discovery and knowledge embedding

An effective parameter identification method is critical in hydraulic transient simulation for pipeline condition assessment. Existing studies neglect the hydraulic spatiotemporal dynamic characteristics and multi-frequency updating characteristics of simulation parameters, resulting in unsatisfactory interpretability and simulation accuracy. In this study, a knowledge-discovery and embedded intelligent framework is proposed to identify optimal friction and capture the multi-frequency variation of friction for accurate hydraulic simulation of liquid pipelines. Particularly, the proposed framework identifies optimal friction by transforming conventional evaluation criteria in optimization theory-based methods based on quantified representations of hydraulic spatiotemporal dynamics. By leveraging underlying physical principles of hydraulic transients, an enhanced neural network is proposed by reforming forward and backward propagation for an efficient surrogate of parameter identification. Subsequently, the proposed framework achieves a multi-frequency parameter refreshment under both pseudo-steady and transient conditions. In this way, a synchronous and flexible online simulation is achieved by integrating knowledge-discovery identification with knowledge-embedded modeling. By comparing to representative squared-error-based method, the efficacy and accuracy of the proposed framework are demonstrated experimentally and numerically on real-world cases. The results suggest a promising application of the proposed framework for industry pipeline simulation and process optimization.