Reduced-order model with radial basis function network for leak detection

Abstract

An inverse transient analysis technique for detecting leaks in water pipe systems through proper orthogonal decomposition (POD) with a radial basis function network (RBFN) is proposed. To verify its novelty and credibility, the performance of this technique was compared with a conventional technique which uses a metaheuristic algorithm in artificial cases with various leak conditions. The inherent shortcomings of heuristic techniques requiring a substantial computational cost were shown to have been resolved. This is because POD acquires a basis by using singular value decomposition and handles data in a reduced-order space which is composed of that basis. Several conclusions were derived. First, the reliability to detect leaks was confirmed. Next, the RBFN learned the relationship between the POD coefficients and leak coefficients through map learning supervised by snapshots with a reliable resolution. Finally, even if another leak occurred, it could be assessed using the presented technique without any data updates.