Optimization of hydraulic machinery by exploiting previous successful designs

Abstract

A design-optimization method for hydraulic machinery is proposed. Optimal designs are obtained using the appropriate CFD evaluation software driven by an evolutionary algorithm which is also assisted by artificial neural networks used as surrogate evaluation models or metamodels. As shown in a previous IAHR paper by the same authors, such an optimization method substantially reduces the CPU cost, since the metamodels can discard numerous non-promising candidate solutions generated during the evolution, at almost negligible CPU cost, without evaluating them by means of the costly CFD tool. The present paper extends the optimization method of the previous paper by making it capable to accommodate and exploit pieces of useful information archived during previous relevant successful designs. So, instead of parameterizing the geometry of the hydraulic machine components, which inevitably leads to many design variables, enough to slow down the design procedure, in the proposed method all new designs are expressed as weighted combinations of the archived ones. The archived designs act as the design space bases. The role of the optimization algorithms is to find the set (or sets, for more than one objectives, where the Pareto front of non-dominated solutions is sought) of weight values, corresponding to the hydraulic machine configuration(s) with optimal performance. Since the number of weights is much less that the number of design variables of the conventional shape parameterization, the design space dimension reduces and the CPU cost of the metamodel-assisted evolutionary algorithm is much lower. The design of a Francis runner is used to demonstrate the capabilities of the proposed method.