FPGA based artificial neural network motor control of PM assisted synchronous reluctance motor in washers

Abstract

Electric motors play a significant role in energy consumption, accounting for approximately 70% of the total energy produced, and it becomes evident that electric motors have a substantial impact on energy-saving efforts. Consequently, there is a growing interest in permanent magnet (PM) motors, which have the potential for higher efficiencies. PM motors rely on rare earth elements as the primary source of magnetism, and the increasing demand for these elements has led to their scarcity and subsequent price hikes. In light of this situation, synchronous reluctance motors (SynRM) have gained importance as an alternative motor type. The efficiency of the system is not only determined by the motor topology but also by the control method employed. However, it should be noted that all these methods require knowledge of the motor parameters. Maximum Torque per Ampere (MTPA) is a control strategy used in electric motor control to optimize torque production while minimizing the current drawn by the motor. Motor parameters like inductance, and flux linkage are typically measured or estimated to implement an MTPA control,. The motor parameters used for the MTPA do not remain constant and change over time. For this reason, the values should be updated constantly to ensure operating at the optimum point. Artificial neural networks are an alternative that can be used for this purpose. In this thesis, a PM assisted synchronous reluctance motor (PMaSynRM) is controlled for the application of a washer. MTPA algorithm is used in the motor control algorithm. The motor parameters used by this algorithm are estimated, and the operating point is constantly updated. The mentioned parameter estimation is provided by using artificial neural networks. Since artificial neural networks require high processing power in terms of their structures, this requirement is met by using a field programable gate array (FPGA). The dataset is created analytically to train the artificial neural networks used. It is important to determine the structure of the network, the number of neurons, and layers. At this point, many studies are carried out to find the most suitable combination. Improvement in motor efficiency is observed in the gathered results as expected. A comparison is made with the data of the system whose parameters are controlled by a fixed parameter MTPA to demonstrate this improvement. The motor parameters used for the fixed parameter MTPA application are taken for the rated speed and torque values of the motor. In the results, it is observed that both control systems operate at the same efficiency point for the fully loaded condition. However, it is observed that the efficiency level of the adaptive system is higher under the lower speeds and loads. This causes by the difference between the required parameters, and the high difference increases the deviation from the MTPA operating point.