Neural Network Estimation of Generator Slot Fields for the Calculation of Radial Field Losses and Circulating Currents.

Abstract

In this paper, an approximate method for the fast calculation of magnetic fields inside the stator slots of large hydro generators is presented. Due to its relatively low computational cost, the approach is suitable for every-day-use calculations in the process of tendering and design of hydro generators. The scheme comprises two parts. The first one is a Boundary Element Method (BEM) solution of a slot with current-carrying strands and ideally permeable walls. The second part consists in an Artificial Neural Network (ANN) for the incorporation of secondary effects, such as tooth saturation. The effects of induced eddy currents on the flux density at the strands’ centres are taken into account by an iterative BEM solution with successive corrections of the resulting fields. Since eddy current effects are comparably small for the configurations usually applied, convergence is readily achieved. The training data for the ANN were obtained from a comparison between the linear BEM results and Finite Element (FE) computations which have been carried out on a set of 65 hydro generators at different slot temperatures. By employing the BEM approach, which captures most of the underlying physics, and a corrective neural network, a grey box model is obtained. This has the advantage that the output of the net is “safeguarded” by the BEM solution. In order to get a robust representation of the data, the training of the ANN was carried out applying a Genetic Algorithm (GA).