The master thesis took place in Stockholm at a company that both designs and manufactures motion solutions, electric motors in particular. There, a motor temperature protection algorithm was developed in the hope of eliminating the need for a physical sensor. The site in Stockholm specializes in system engineering based on customer needs but also in testing the motor software. The software helps in managing a problem present in all electric motors, namely heat. Motors are often limited by the heat generated inside of them. The motor insulating lacquer around the copper windings will melt and the motor will get destroyed if it operates at high torque for an extended period of time. Some motors utilize a fuse that burns off and opens the circuit so to prevent the motor from short-circuiting when the exposed windings connect. The synchronous electric motor used in the thesis has a microcontroller incorporated. The controller reads the temperature from a sensor in the motor and can limit the torque output if the temperature gets too high. This is good but the sensor reading is not accurate and adds to the production cost in an industry with tight margins. The thesis goal is therefore to develop an algorithm that can estimate the motor temperature instead of measuring it.

The estimation algorithm is a Kalman filter that uses information about the microcontroller temperature in addition to knowing the torque and speed of the motor. The thesis investigates whether this all together is enough information to estimate the motor temperature with reasonable accuracy. The algorithm was developed and evaluated using many tests. The tests were performed at different ambient temperatures and with different speed-torque profiles. The algorithm was coded and evaluated in Matlab. The process of creating the algorithm involved experimentation and tuning of multiple parameters. It had to be computationally light so as to not overload the limited processing power of the controller.  

The final version of the algorithm performed reasonably well. The error was small but not zero. The response was not fast but still fast enough to be practical. The free dynamics were not good in the long run but were corrected for by the Kalman filter gain. The project was deemed successful and inspired hope for future work.