Abstract
High-performance induction motor drives are penalised at both terminals: on the supply side by low-order current harmonics injected by the diode front end, and on the machine side by the parameter dependence and the weighting-factor tuning burden of fi-nite-control-set model predictive control. This paper studies a terminal-to-terminal archi-tecture that addresses both. An eighteen-pulse phase-shifting autotransformer rectifier can-cels every characteristic harmonic below the seventeenth, and its residual dc-link ripple is measured and fed forward into the machine-side observer. The machine-side controller replaces the motor model by a first-order ultra-local model whose lumped disturbance is reconstructed by a fractional-order extended state observer, while the ultra-local input gain is identified on line by a gradient law; the resulting predictive current law is single-objec-tive and therefore needs no weighting factor. Convergence is established with a Lyapunov argument and Matignon’s criterion. Simulation of a 4-kW drive shows the eighteen-pulse front end reducing the idealised supply-current distortion from 30.8% to 9.8%, and a stator-current distortion of 1.34% for a controller that uses no machine parameter at all, against 1.53% for a well-tuned model-based benchmark. Three deliberately negative findings are reported: the input-gain sensitivity is strongly asymmetric, so that the gain should be biased low rather than centred; the fractional order brings no measurable benefit over the integer observer under the bandwidth parameterisation used here; and the ripple feed-forward is largely redundant once the eighteen-pulse rectifier is present.