Innovative solutions for converters and motor drives oriented to smart cities and communities

Smart cities and communities are conjugated by European Union in different areas, including energy efficiency, low carbon technologies and mobility which are deeply merged with electric motors. Electric machines are ubiquitous in industry for a wide range of applications, consuming between 43% and 46% of all electricity that is generated in the world. Although some machines are used for high-performance applications, such as robots and machine tools, the majority are used in industrial processes for pumps, compressors, fans, conveyors, and other slower-dynamic applications. It is estimated that 92% - 95% of the life cycle costs of electric motors are associated with the energy they consume, leading to typical payback periods of < 2 years for the installation of an adjustable-speed drive. It is rather surprising to learn that, despite overwhelming evidence of the attainable savings, only 10% - 15% of all industrial motors presently use electronic adjustable speed drives. On the motor side, Synchronous Reluctance (SynR) motors are gaining lots of attention from industrial researchers and academics, due to their inherent characteristics like the high efficiency, the low cost and the low environmental footprint. Their characteristics fully meet the requirements imposed by smart cities and communities and the aforementioned low-dynamics applications, so they could be the heart of the revamping of those plants. There is wide agreement that the potential for future growth in the sales of industrial drives and SynR motors is still very substantial. SynR motors are prone to magnetic saturation, making the classic model with lumped parameters unsuitable. The main part of this thesis concerns the development of a new magnetic model for anisotropic motors, especially for SynR motors. It is based on a special kind of neural network (NN), called Radial Bases Function (RBF) NN, which is particularly advisable for an online updating due to its local property. A complete training procedure is proposed in which some considerations are done to define several NN parameters and to convert the nonlinear training problem into a linear one. Two different training algorithms are presented, the former one is fast but computationally cumbersome then suitable for an offline training while the latter one is lighter then proper for an online training. In order to complete the online parameters identification, a scheme based on a DC current injection is developed to estimate the stator resistance. An exhaustive analysis is carried out to disclose that the proposed method is independent from other motor parameters which is a strength asset in a saturable motor. An accurate stator resistance value improves in turn of the magnetic model. The second part of this dissertation deals with how to exploit an accurate magnetic model to enhance the motor control. In order to improve the efficiency of the motor, exploiting the RBF NN model and the online training algorithm, the Maximum Torque per Ampere (MTPA) curve is found. Starting from a blank NN, it is continuously online trained and a proper algorithm understands where the MTPA curve is respect to the current working point. Afterwards, the drive moves itself towards the actual MTPA. Finally, three different current control schemes tailored for anisotropic motors are presented, all based on the available NN-based magnetic model. The first one is a gain-scheduling PI control where the control gains are accordingly tuned to the working point to keep constant the control bandwidth. The second one is based on a classical PI regulator with a FF action to compensate for all the nonlinearity of magnetic maps. The third one is a constrained direct Model Predictive Control (MPC) where a long prediction horizon is achieved. In order to accomplish a long prediction horizon, the Sphere Decoding Algorithm is properly modified to make it suitable for a nonlinear system. The whole thesis was fully validated through an intensive simulation and experimental stage, except the long--horizon MPC which was tested only by simulation.