Browsing by Subject "Kalman filter"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    A method for measuring time-resolved, path-integrated temperature in a reciprocating internal combustion engine cylinder using ultrasonic thermometry
    (2016-09) Weigelt, Chad
    A limitation currently facing internal combustion engines research is the lack of a direct method of measuring the temperature of the gases inside the cylinder. The rate at which a combustion cycle evolves is too rapid for conventional, direct measurement systems such as thermocouples. Advanced optical systems like laser-induced fluorescence (LIF) require heavy engine modifications to be effective and cannot withstand typical engine operating conditions. The direct measurement of the gas temperature in an engine cylinder, at realistic operating conditions is needed to better understand the combustion cycle and its effects on cycle efficiency. The study presented here uses a Kalman filtering technique to measure the time of flight of an ultrasonic signal to calculate the temperature of the gas in an engine cylinder. Results show that the method is capable of measuring temperatures within an accuracy of 10% and within sample deviations below 0.25% with sufficient data.
  • Thumbnail Image
    Item
    Mobile Robot Localization Under Processing And Communication Constraints
    (2016-03) Nerurkar, Esha
    Mobile robot localization is one of the most fundamental problems in robotics. For robots assisting humans in tasks such as surveillance, search and rescue, and space exploration, accurate localization, that is, precisely estimating the robot's pose (position and orientation), is a prerequisite for autonomous operation. The system resources (processing and communication) for localization, however, are often limited, and their availability varies widely depending upon the application and the operating environment. Therefore, the objective of this work is to develop resource-aware estimators for robot localization, which optimally utilize all available resources in order to maximize estimation accuracy. In the first part of this thesis, we address the problem of robot localization under processing constraints, focusing on the key applications of single-robot Simultaneous Localization and Mapping (SLAM) and multi-robot Cooperative Localization (CL). For SLAM, we propose two resource-aware approaches, the approximate Minimum Mean Squared Error (MMSE) estimator-based Power-SLAM algorithm and the approximate batch Maximum A Posterior (MAP) estimator-based Constrained Keyframe-based Localization and Mapping (C-KLAM). When approximations are inevitable due to processing constraints, both approaches aim to minimize the information loss while generating consistent estimates. For CL, we exploit the sparse structure of the batch MAP estimator to develop a resource-aware, fully-distributed multi-robot localization algorithm, that harnesses the processing, storage, and communication resources of the entire team, to obtain substantial speed-up. The second part of this thesis focuses on CL under communication constraints, in particular, asynchronous communication and bandwidth constraints. Due to limited communication range or the presence of obstacles, robots communicate asynchronously, that is, they can only interact with different sub-teams over time and exchange information intermittently. For this scenario, we develop a family of resource-aware information exchange rules for the robots, in order to ensure optimal and consistent localization performance. Lastly, this thesis investigates the problem of decentralized estimation under stringent communication bandwidth constraints. Here, robots can communicate only a severely quantized version (few or only one bit), of their real-valued sensor measurements, to the team. Existing estimation frameworks, however, are designed to process either real-valued or quantized measurements. To overcome this drawback, we propose a paradigm shift in estimation methodology by focusing on the design and performance evaluation of the first-ever, resource-aware, hybrid estimators. The proposed hybrid estimators are able to process both locally-available real-valued information, along with the quantized information received from the team, in order to maximize localization accuracy. Finally, we note that mobile robot applications are no longer limited to specialized and expensive robots. Commonly-available hand-held devices such as cell phones, PDAs, and even cars, are equipped with processing, sensing, and networking capabilities. Therefore, when coupled with the proposed innovative, scalable, and resource-aware algorithms, these ubiquitous mobile devices can lead to a proliferation of novel location-based services.
  • Thumbnail Image
    Item
    Online Identification of Abdominal Tissues During Grasping Using an Instrumented Laparoscopic Grasper
    (2013-08) Sie, Astrini
    Modern surgical tools provide no advanced features like automated error avoidance or diagnostic information regarding the tissues they interact with. This work motivates and presents the design of a “smart” laparoscopic surgical grasper that can identify the tissue it is grasping while the grasp is occurring. This allows automated prevention of certain errors like crush injury. A nonlinear dynamical model of tissue mechanics is adopted along with an extended Kalman filter to demonstrate the feasibility of this design in simulation and in situin vivo on porcine models. Results indicate that while the approach is sensitive to initial conditions, tissue can be identified during the first 0.3s of a grasp.
  • Thumbnail Image
    Item
    Sparsity-Promoting Estimator Design for Acceleration Sensor Placement in Civil Structures
    (2019-05) Gustafson, Kali
    The scale of civil systems makes it impossible to measure all degrees of freedom. Therefore, a limited number of measurements are leveraged to obtain a full set of state estimates (e.g. displacement and velocity responses). Spatially sparse feedback, which limits the information and the number of sensors, requires the selection of essential measurements. The exact placement problem considers all possible combinations of sensors, which presents computational challenges for large systems. When applied to benchmark structures, the Kalman filter alternating direction method of multipliers (kfadmm) algorithm systematically balances measurement sparsity and estimator error covariance in acceleration sensor selection. Compared to the exact and sequential sensor placement methods, the kfadmm approach selects similar sensors with slightly higher estimation error and fewer combinations considered. In kfadmm, the best number of sensors for a given application can be determined after looking at the increase in the error as sensors are removed from the system.
  • Thumbnail Image
    Item
    Synthetic air data estimation: a case study of model-aided estimation
    (2014-09) Lie, F. Adhika Pradipta
    A method for estimating airspeed, angle of attack, and sideslip without using conventional, pitot-static airdata system is presented. The method relies on measurements from GPS, an inertial measurement unit (IMU) and a low-fidelity model of the aircraft's dynamics which are fused using two, cascaded Extended Kalman Filters. In the cascaded architecture, the first filter uses information from the IMU and GPS to estimate the aircraft's absolute velocity and attitude. These estimates are used as the measurement updates for the second filter where they are fused with the aircraft dynamics model to generate estimates of airspeed, angle of attack and sideslip. Methods for dealing with the time and inter-state correlation in the measurements coming from the first filter are discussed. Simulation and flight test results of the method are presented. Simulation results using high fidelity nonlinear model show that airspeed, angle of attack, and sideslip angle estimation errors are less than 0.5 m/s, 0.1 deg, and 0.2 deg RMS, respectively.Factors that affect the accuracy including the implication and impact of using a low fidelity aircraft model are discussed. It is shown using flight tests that a single linearized aircraft model can be used in lieu of a high-fidelity, non-linear model to provide reasonably accurate estimates of airspeed (less than 2 m/s error), angle of attack (less than 3 deg error), and sideslip angle (less than 5 deg error). This performance is shown to be relatively insensitive to off-trim attitudes but very sensitive to off-trim velocity.