Modelling, simulation and testing of artificial neural network augmented kalman filter for INS/GPS and magnetometer integration

Yıldız, Doğan
The objective of this thesis is to investigate a hybrid Artificial Intelligence/ Kalman Filter (AI/KF) system to determine 3D attitude, velocity and position of a vehicle in challenging GPS environment. In navigation problem, the aim is to determine the position and velocity of the host vehicle from initial conditions. By using Inertial Measurement Unit (IMU), it is possible to calculate position and velocity with an error. In other words, during the integration stage of the IMU measurement, errors will be accumulated throughout the time. In literature to eliminate the divergent characteristic of integral calculation, other sensor measurements are combined with navigation calculation process. The traditional complementary technique to calculate the vehicle position, velocity and attitude is integrated Inertial Navigation System (INS) and Global Positing System (GPS). The integrated INS/GPS shows greater accuracy with respect to standalone INS. To achieve this accuracy it is common to use Kalman Filter as an integration technique. The Kalman Filter approach has been used widely as the standard optimal estimation technique. However because of the acquiring an accurate stochastic model and prior knowledge of the measurement error for the precision of the estimation KF has several shortcomings. Based on these shortcomings Artificial Intelligence (AI) based techniques are motivated. In this thesis, navigation mechanization algorithm and sensors mathematical models were studied. Accelerometer, gyroscope, GPS and magnetometer were selected for the sensor fusion integration. With the help of accelerometer and gyroscope, position and velocity of the host vehicle were realized through integration process of mechanization algorithm. Also GPS and magnetometer measurements were used for position-velocity and heading determination independent from IMU respectively. The design of sensor fusion algorithm is based on Extended Kalman Filter (EKF). The EKF linearized mathematical model relies on the error propagation model of the mechanization equations. As for the Neural Network structure Multilayer Perceptron Neural Network (MLPNN) were used to improve the integration results during GPS outages. After modelling and simulating the results in simulation environment, real test data were used in AI/KF based prediction algorithm. The measurement results were logged in computer to be used in algorithm. The results show how AI/KF based algorithm is more accurate during GPS outages with respect to standalone Kalman Filter algorithm.


Utilization of neural networks for simulation of vehicle induced flow in tunnel systems
Koç, Gencer; Albayrak, Kahraman; Sert, Cüneyt; Department of Mechanical Engineering (2012)
Air velocities induced by underground vehicles in complex metro systems are obtained using artificial neural networks. Complex tunnel shaft-systems with any number of tunnels and shafts and with most of the practically possible geometries encountered in underground structures can be simulated with the proposed method. A single neural network, of type feed-forward back propagation, with a single hidden layer is trained for modelling a single tunnel segment. Train and tunnel parameters that have influence on ...
Numerical Simulation of a Flexible X-Wing Flapping-Wing Micro Air Vehicle
Deng, S.; Perçin, Mustafa; van Oudheusden, B. W.; Bijl, H.; Remes, B.; Xiao, T. (2017-07-01)
Numerical simulations were performed to investigate the flowfield around a flexible flapping-wing micro air vehicle using an in-house-developed computational fluid dynamics solver. To include the dynamics of the flexible wings and its impact on the aerodynamic behavior of the micro air vehicle, the wing-deformation pattern during flapping was experimentally determined by a stereovision measurement. These data were subsequently interpolated to be employed as prescribed flapping kinematics in the numerical fl...
Design and implementation of a 3D range scanner for mobile robots
ASLAN, Gökhan; Konukseven, Erhan İlhan; Koku, Ahmet Buğra (2014-01-06)
In an efficient autonomous navigation and exploration, the robots should sense the environment as exactly as possible in real-time and act correctly on the basis of the acquired 3D data. Laser scanners have been used for the last 30 years for mobile robot navigation. However, they often did not enough speed, accuracy and field of view. In this paper we present the design and implementation of a scanning platform, which can be used for both outdoor and indoor mobile robot navigation and mapping. A 3D scannin...
Flow characterization of full, partial, and inclined ground effect
Koçak, Göktuğ; Yavuz, Mehmet Metin; Nalbantoğlu, Volkan; Department of Mechanical Engineering (2016)
Flight condition of helicopters proximity to the ground, so-called “Ground Effect”, is among one of the most recent research areas since the aerodynamic performance of these vehicles significantly vary due to the flow dynamics associated with the interaction between the rotor and the surface. Thus, a comprehensive understanding of the rotor wake during in and out of ground effect is needed to improve the flight performance. Helicopters operate in ground effect in many different ways including full ground, p...
Mobile robot navigation: Implementing the GVG in the presence of sharp corners
Konukseven, Erhan İlhan (1996-01-01)
A robot can explore an unknown environment by incrementally constructing a roadmap of that environment using line of sight sensor information. Recall that a roadmap is a one-dimensional representation of a robot's environment. This paper addresses one problem that occurs while generating a roadmap: what happens when sonar sensors cannot detect sharp objects because of the specularities? To do this, a new sensor model is combined with an already existing incremental construction procedure for a roadmap. Expe...
Citation Formats
D. Yıldız, “Modelling, simulation and testing of artificial neural network augmented kalman filter for INS/GPS and magnetometer integration,” M.S. - Master of Science, Middle East Technical University, 2016.