A Master of Science Thesis in Mechatronics Submitted by Tariq Salman AbuHashim Entitled, "Improving INS/GPS Integration for Mobile Robotics Applications" April 2008. Available are both Soft and Hard Copies of the Thesis.
As unmanned systems become more and more important, reliability and integrity issues become definite, specially when being implemented with low-cost (or sometimes are referred to as commercial-of-the shelf or COTS) sensors while being designed to operate in remote, hazardous and harsh environments. As a result, fault (and failure) detection and identification (FDI) is a must, and it is a crucial requirement in designing unmanned vehicles. In this thesis, integrity is defined as the ability of the system to provide reliable navigation information, to monitor the health of the aids, to detect abnormalities in their behavior, and to survive once a failure in one of its components (whether they are sensors, actuators, mathematical models, and computations) occurs. On the other hand, reliability is component dependent. A navigation system is reliable as its most unreliable component. Therefore, integrity implies reliability while reliability not necessarily implies integrity. This thesis, mainly, discusses the issue of implementing a low-cost inertial navigation system, aided with satellite navigation system. In doing so, a fault detection and identification scheme must be involved and the performance of all the system components must be verified. The FDI system should take into account types of failures commonly occur, guarantee that all faults will be detected, assist design specifications and respond as fast as possible to faults. On the other hand, it should take into account the complexity of the implementation and its robustness in the presence of mismodelling. Innovation-based techniques, in particular the 2 SCT, offer tradeoffs between complexity and performance and detect a large set of failures. However, they are sensitive to filter tuning and have no fault identification ability. On the other hand, the model-based approaches, in particular the multiple model adaptive estimation (MMAE), have an outstanding decision making ability and are insensitive to filter tuning. However, they require a priori knowledge on the system and failure model and are computationally expensive. The integration of both techniques can enhance the FDI performance of both systems. In this thesis a sequential FDI algorithm is proposed. This algorithm employs an innovation-based technique for fault detection and a model-based technique for identification. The performance of the 2-MMAE sequential algorithm is simulated and tested on actual IMU and GPS data. Results showed that the sequential algorithm has a comparable identification ability as the MMAE algorithm with a substantial reduction in computational requirements, since the filters bank was only allowed to operate on segments of time where faults were detected. On the other hand, unlike the MMAE algorithm where the performance of the filter was affected during no-fault conditions, the sequential scheme guaranteed the consistency of the estimator in all of its modes of operation and didn't affect its performance during normal no-fault modes of operation.