Predictive torque and flux control of an induction machine drive using fuzzy multi-criteria decision making VIKAS KUMAR * , PRERNA GAUR and A P MITTAL Division of Instrumentation and Control Engineering, Netaji Subhas Institute of Technology, New Delhi, India e-mail: sainivika@gmail.com MS received 9 February 2015; revised 28 August 2016; accepted 10 November 2016 Abstract. Among the numerous direct torque control techniques, the finite-state predictive torque control (FS- PTC) has emerged as a powerful alternative as it offers the fast dynamic response and the flexibility to optimize multiple objectives simultaneously. However, the implementation of FS-PTC for multiple objectives opti- mization requires the optimization of a single objective function, which is constructed using weighting factors as a linear combination of individual objective functions. Traditionally, the weighting factors are determined through a non-trivial process, which is a complex and time-consuming task. In an effort to avoid the time- consuming task of weighting factor selection, this paper aims at replacing the weighting factor calculation with a systematic fuzzy multiple-criteria decision making in which the individual objective functions may have equal or varying degrees of importance. As a result the weighting factor calculation can be completely avoided. The simulation and experimental tests are conducted on a 2.2 kW induction motor drive to validate the proposed approach. The result outcomes are compared with the conventional predictive torque control (PTC) using weighting factors on the same experimental platform. Keywords. Finite-state model predictive control; fuzzy decision making; multi-objective optimization; predictive torque control. 1. Introduction For high dynamic servo control, direct torque control (DTC) is an indispensable control method, due to its simple implementation, low computational time and reduced parameter sensitivity [1, 2]. However the conventional DTC is implemented using a torque and flux hysteresis controller, which causes large torque ripples as compared with the field oriented control (FOC) [3]. In order to reduce the torque ripples, several modifications to the conventional DTC scheme like space-vector-modulation-based DTC (SVM- DTC), sliding-mode-based SVM-DTC, fuzzy-logic-based variable duty cycle DTC and predictive torque control (PTC) [4–6] have been proposed. In the recent years, due to the advent of high-speed digital signal processors (DSPs) the PTC has emerged as a powerful alternative to the control of power converters and electrical drives due to its fast dynamic response, flexibility to include multiple objectives and easy inclusion of system constraints and plant non-linearity [7–9]. PTC has been implemented using a discrete model of the converter that is evaluated for each possible actuation and a cost function based upon the desired control objective is optimized in every sampling period linearity [19]. However, for applications where more than one control objectives are desired to be met, a single cost function or aggregate objective function (AOF) that represents the desired beha- viour is created using weighting factors as a linear combi- nation of individual objective functions. The AOF is evaluated for every possible switching state and the switching state that minimizes the AOF is applied in the next sampling instant [8, 19]. The determination of these weighting factors for a given objective function is a complex and time-consuming task due to the different natures of individual objective functions. Traditionally the weighting factors are determined ana- lytically and no systematic design guidelines have been reported. Some weighting factor determination guidelines can be found in Cortes et al [10] as proposed by them, in which the branch and bound algorithm is effectively used for multi-objective optimization. However the branch and bound algorithm is an offline procedure and is tough to perform. The computational complexity of the algorithm increases as the number of objective functions increases. Rojas et al [11] used a ranking-based method for multi-objective optimiza- tion. Using the ranking-based method, the branch and bound algorithm is completely avoided, but the ranking-based method requires sorting of the individual objective functions or the error values, which require significant computational burden. The computational burden depends upon the sorting algorithm used and it can be demanding if more than two *For correspondence 343 Sa ¯dhana ¯ Vol. 42, No. 3, March 2017, pp. 343–352 Ó Indian Academy of Sciences DOI 10.1007/s12046-017-0606-z