Measuring system for d- and q- axes flux in synchronous machines and modeling using adaptive network fuzzy inference system

Abstract

This thesis deals with the development of a new measuring system to determine experimentally the magnetic flux characteristics of Projected pole synchronous machines in direct axis (d-axis) and quadrature axis (q-axis). New methods for the estimation of flux in quadrature axis (q-axis) using regression analysis and Adaptive Network Fuzzy Inference System are also proposed. Magnetic characteristics during saturation of synchronous machine are the main consideration for improving the design and modeling of the machines. Regression techniques such as Multivariate linear regression and Multivariate polynomial regression are used for the estimation of q- axis flux. Along with the experimental test results measured using Germanium diode flux sensors, available q-axis magnetic flux data calculated from the measured d-axis flux by Ahmed El- Serafi and Kar (2003), for various Salient pole synchronous machines namely, a) Microalternator, b) Machine No. 1 and c) Machine No. 2 form the basis of the modeling schemes reported in this thesis. Further, the statistical test of agreement between the modeled and reference values is performed by determining the root mean square error (RMSE). It is found that the ANFIS model 3 which uses gbell membership function yields a negligibly minimal error in comparison with the other models and the predicted values by this model are in close agreement with the measured and calculated values. On the other hand, MVLR, MVPR and two other ANFIS models estimate q-axis flux comparatively with little higher value of error. RMSE of ANFIS model 3 is proved to be very low and closer to zero, indicates better fit of the model. Thus, this research work concludes with the contributions of experimental investigations using Germanium diode flux sensors. The related experimental results on d-q axes flux in projected pole synchronous machine, MVLR and MVPR mathematical models, ANFIS based soft computing modeling schemes are found to be in good agreement with the published results by the researchers.