Experimental Simplex-Genetic Algorithm for Self-Commissioning of Electric Drives EPE Journal ⋅ Vol. 17 ⋅ n o 3 ⋅ October 2007 31 Introduction The electric motor drives are the key component of automation over the last twenty years, and according to recent studies [1], [2], the electric drives market is one of the most dynamic sectors. Recently there has been an increased interest in smarter electric drives, i.e., more user friendly and capable of self-commissioning. Self-commissioning should avoid the typical situation, where the final user is forced to buy drive and motor from the same manu- facturer to obtain good performance without any extra-tuning. Self-commissioning consists of an automatic procedure for the tuning of controllers based on a previous parameter identification, when a motor is initially connected to the drive [3]. In this way the final user can utilize the drive with a third-party motor without worrying about tuning. The standard techniques for self-commissioning consist of specific sequences of tests to measure electrical and mechanical para- meters of the motor [4, 5]. These identified parameters are then used for a model-based control design. By using such a technique, the control system can be tuned in a very short time. Although good enough for several applications, only a sub-optimal tuning is achieved because the issues concerning the multiple inputs, system nonlinearities and uncertainties of the model-based design remain unaddressed. In order to overcome this obstacle, self-commissioning can online optimize control system parameters, prior to the actual use of the system, which then uses these optimized controllers. This is also known in the research area of automatic control as auto-tuning or hardware-in-the-loop optimization [6]. This on-line optimization offers the advantages of the model-free design and makes self- commissioning very reliable because the controllers are experi- mentally evaluated. This paper deals with the auto-tuning of electric drives based on a Simplex-Genetic Algorithm. The tuning of the control system implemented in an electric drive is a multiobjective problem that involves a large number of parameters in the presence of noise. Recently GAs have been successfully applied to this kind of opti- mization and have been proved to be more robust than classical techniques [7] also in the power electronics and electric drives [8]. More details and recent trends in the applications of evolutionary techniques can be found in [9]. Unfortunately the on-line evolu- tion of the control system can need many tests, and the controlled process can be critically stressed by poorly performing solutions. For this reason the successful applications of on-line GAs are really limited in number. In this work, original suggestions and innovative solutions are proposed to fully perform an on-line optimization without any risk for the hardware. Moreover, a new real-time fitness implementation can halt the carrying out of an experiment, if a highly unsatisfactory solution is recognized, and a new hybrid architecture integrates GA and simplex method in order to speed up the convergence. Vector-controlled PMSM drive In Fig. 1a the block diagram of a vector-controlled PMSM drive is shown. This control scheme is based on the dynamic equations of the motor in the rotor flux reference frame (d,q) [10]: (1) where v sd and v sq are the d- and q-axis stator voltages; i sd and i sq are the d- and q-axis stator currents; R s , L sd and L sq are the stator resistance, d- and q-axis inductances; Ψ is the flux linkage of the permanent magnet (PM); and ω r is the electrical rotor speed. Both the d- and q-axis voltage references, v * sd and v * sq , are provided partly by current controllers, partly by voltage compensators. The d- and q-axis current controllers are two PI controllers whose transfer functions are K isd (1 + τ isd s)/s and K isq (1 + τ isq s)/s respec- tively. The d- and q-axis voltage compensators, whose feedfor- ward actions are – ω r i sq K 1 and ω r (i sd K 2 + K 3 ) respectively, improve the performance of the current control loops by reducing the influence of the cross coupling terms, i.e. the last terms in (1). The reference of the q- axis current is provided by the speed con- v Ri L i t Li v Ri L i sd s sd sd sd r sq sq sq s sq sq d d d = + - = + ω sq r sd sd dt Li + + ( ) ω Ψ Experimental Simplex-Genetic Algorithm for Self-Commissioning of Electric Drives Giuseppe L. Cascella, Nadia Salvatore, Luigi Salvatore, Dip. di Elettrotecnica ed Elettronica - Politecnico di Bari, Bari, Italy Mark Sumner; School of Electrical and Electronic Engineering, University of Nottingham Keywords: Variable speed drives, Permanent magnet motors, Self-Commissioning, Hybrid Genetic Algorithms Abstract This paper deals with the self-commissioning of electric drives. To improve the performance of the available industrial drives, an on-line auto-tuning based on a hybrid genetic algorithm is proposed. This strategy integrates the simplex method, local searcher, in a genetic framework, global searcher, in order to speed up the convergence. Moreover it is very reliable because experimentally tests each possible solution and consequently the final result is not affected by the accu- racy of the motor model. Finally, the proposed on-line hybrid optimization can be embedded as a fully-automated tool without anyextra-hardware on industrial drives. Extensive experimental results prove the effectiveness of the proposed approach not only in comparison with conventional commissioning, but also when compared with further accurate hand- calibration.