Contributions to improve the method of sensorless control of PMSM

Electrical machines that become indispensible for human life are the greatest energy consumer in electrical networks and advances in its theory, control, manufacturing and cost are reached day by day. Electric machines are mainly divided into three as direct current, induction and synchronous machines. In induction machines, the interaction between the stator and the rotor results in the voltage induction by rotating field. In synchronous and direct current machines, the interaction between the armature field and the excitation field is the subject. In synchronous machines, the windings on the rotor namely round rotor and salient pole are fed with the direct voltage to create the excitation field. In direct current machines, the excitation windings located in the stationary part are fed and the rotor is connected via brush-collector arrangement. For both machines, excitation windings and excitation losses exist and brush assemblies with friction are used. As a result of increased attention for importance of efficiency in electrical machines, magnet motor technology has come to forefront in order to eliminate excitation winding losses. In recent years, as a result of developments in materials science, stronger and more special magnets could be produced and the production of greater power electric motors is available. It is possible to use magnets in the motors which include excitation windings such as synchronous and direct current machines. This situation also removes the brush-collector structure. Magnetically excited machines are called magnet machines and machines with electronic commutation are called brushless machines. However, since the position of the rotor is to be known in order to control the motor, different control methods are required. The permanent magnet synchronous motor is a motor that was born related to fastly developing technology. The stator structures of permanent magnet synchronous motors or brushless direct current machines have a conventional three-phase winding distribution, while the rotors have magnets. If the induced voltage wave is trapezoidal it is called brushless direct current machine, if the wave is sinusoidal it is called permanent magnet synchronous motor. In the motors, rotors including magnet, the air gap length can be constant or variable, the magnets can be placed to different positions. The rotor structure, position and size of the magnets affect the characteristics of the motor directly. Increasing the efficiency of the electric machines, lowering the cost, etc., are generally considered when the machine is designed, but it must also be taken account after the machine has been manufactured. In addition to the production cost, operating costs should also be considered. Operating a machine away from highest efficiency, reduces the efficiency of the machine. Therefore, the cost per power is increased and this does not satisfy the user. Inefficient operation increases operating costs. Control methods are directly influential on time and energy cost savings. In the control strategy of the magnet machines, it is necessary to know the position of the rotor in order to ensure the interaction between magnets and the magnetic field generated by the other windings. The rotor position can be determined in two ways namely sensored and sensorless. The sensored control is based on the information provided by the sensing elements placed in the motor. These sensors may be hall sensors, optical encoders or optical decoders. From sensors, position information and current information can be obtained. These sensors must be added to the structure while the machine is in the design phase so that the rotor can give position information. Attaching these elements to the structure increaseses the cost as well as the maintenance needs of the machine and in case of corruption, it will endanger the operational security. Moreover, due to the supply voltage required by these elements, the machine structure becomes more complex. The information from the output in the sensored control is the actual signal and an error mark is obtained by comparing it with the reference input. In the next feedback step, the controller runs to minimize the error mark. The running time for minimizing varies according to the controller parameters. Since sensored control has disadvantages such as increasing cost, complicating machine structure, researches are focused on the methods of controlling the machine without using a sensor. The sensorless control can be divided into scalar and vector control. The scalar control is based on keeping the ratio of input voltage to frequency constant so as to keep the flux in the machine constant. The voltage and frequency are increased within a certain ratio up to the nominal voltage, which is called the constant moment operation. It can not be increased over the rated voltage, but the frequency can continue to be increased and this operational section is called the constant power operation. Speed estimation is performed by this method, but the transient state performance of the motor is not high. The high performance requirement leads way to vector control. Vector control is used if estimation of field angle or rotor position is also desired. With vector control, the machine's three-phase voltages and currents are transformed to dq axis reference frame. Thus, torque and speed can be controlled by two different current components, as in DC machines. Not only the amplitude but also the angle between the current and the flux are taken into consideration, so the moment control becomes more effective and the transient behavior is improved. Since the position of the field angle is difficult to measure, many estimation algorithms have been developed. Some of these algorithms are open-loop, others are closed-loop. An example model-based adaptive control can be given to the open loop estimation algorithm while the Kalman filter and the Luenberger observer can be given as closed loop method examples. The performance of the Luenberger observer lies behind the Kalman filter in terms of steady state response. Since its founding in the 1950's, the Kalman filter has been widely used in aircraft, space, machinery, and electronics. This method, which has a statistical basis, can be used in estimating the future value from past values and estimating the quantities that can not be sensed from measurable quantities. The Kalman filter comes before other observer-based algorithms because of its advantages such as good transient response, disturbance rejection effects, etc. The mathematics of the Kalman filter is based on the state space model of the system to be used. The filter gives realistic results since it considers the system and measurement noises. System and measurement noises are white noises with an average value of zero added to the state space model. The values of the Q and R matrices, which are the covariance matrices of these noises, affect the performance of the system. Higher values of Q indicate that system noises are high; which leads to uncertainty in the model parameters. The higher the R value, the higher the measurement noises. In many applications, the values of these matrices are found through trial and error method. If Q is dominant to R, the gain of the filter is high, transient response is good. If R is dominant to Q, the filter gain is low. In cases where these values are too high or too low, problems of instability in the filter arises, so that the components of these matrices must be optimized. Optimization can be called the best way to use the facilities available, or it can be called a minimizing or maximizing problem. If the problem of optimization is a mathematical problem, classical algebraic methods may not be always sufficient. Therefore, artificial intelligence algorithms are used. The genetic algorithm is a natural-inspired algorithm based on the theory of evolution. This algorithm finds the maximum or minimum of the problem concerned by following the steps which are similar to the evolution theory. When the optimization process is applied to the electric machine, the fitness function to be optimized is selected. This selection is based on the error signal based on the difference between the actual output and the reference input. These error functions based on the error signal are minimized to make the output signal as close as possible to the reference input signal. In this study, a Kalman filter structure was applied on a permanent magnet synchronous motor. In simulations, firstly Kalman filter's performance is evaluated with random Q and R matrices. The performances of the filter in different speed reference inputs are examined. Then, R and Q matrix values which are found by trial and error method in the application were optimized by using genetic algorithm. Integral square error was used as the fitness function. Simulations with new optimized R and Q matrices were repeated and performance was improved. Since the result in the simulations are successful, practical case was initiated. In the application, hall sensored permanent magnet synchronous motor with the parameters of 69 W, 3000 rpm, 24 V and TMS28027F processor of C2000 series of Texas Instruments as digital signal processor were used. The driving was performed by the DRV8301 card, and the Kalman filter was written in the C language in the Code Composer Studio development environment. GUI Composer was used as an extension of the development environment so that the desired sizes can be observed on the graphic user interface. The reference inputs were given to the system with the code written and the actual speed output was read through the Hall sensors and the filter performance was tested. The R and Q matrix components obtained by the optimization process were used in the code and the filter performance was also evaluated in the real system. It was shown that the simulations results substantially match up with the experimental results and the filter performance was improved by optimizing the R and Q matrices.