Title: Detection of biases and Faults in Navigation Sensor Measurements
Author(s): Sasha Draganov
Published in: Proceedings of IEEE/ION PLANS 2016
April 11 - 14, 2016
Hyatt Regency Hotel
Savannah, GA
Pages: 1008 - 1014
Cite this article: Draganov, Sasha, "Detection of biases and Faults in Navigation Sensor Measurements," Proceedings of IEEE/ION PLANS 2016, Savannah, GA, April 2016, pp. 1008-1014.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: A modern navigation system integrates data from different sensors that have different error statistics, including biases. For example, inertial measurement units (IMUs) are known to have biases and other errors, and it is common to allocate states in the filter for estimating such errors. However, models for such biases are typically based on some assumptions about error statistics, such as a random walk for the bias magnitude. When these assumptions are incorrect, a new approach is required. Measurements from a barometric altitude sensor present an example: a bias may change if a user steps in or out of a building, or if a forced-air heating system is turned on/off while the user is in a building. In cases like this, a bias may exhibit a jump. In another example, one may need to watch for faults in the data that may occur due to equipment malfunction. If the magnitude of a measurement error is large enough, a measurement fault can be detected in a single epoch by using measurement redundancy. An example of such algorithm is the Receiver Autonomous Integrity Monitor (RAIM), which is widely used in GPS measurement processing. However, biases in the measurements can be small enough to make their detection in a single epoch difficult. In this paper, we present an algorithm for detecting hidden biases in the measurements. Unlike the RAIM algorithm, we use redundancy not just across measurements in a single epoch, but also over time. By leveraging time history, we are able to detect small, otherwise hidden biases and faults in sensor measurements. This approach requires that the navigation system employs an IMU or some other time update mechanism, which “stitches together” the user state estimates over time. The algorithm runs several independent processing threads, comparing user states with different sensors (or channels) excluded. The key challenge in this approach is to minimize the computational load, so that processing requirements scale less than the number of threads. We achieve this goal by observing that most computations in the filter are common across all threads and can be recycled. As the result of the computation, we arrive at bias/fault estimates for each sensor (or a measurement channel). Obviously, the desirable value for a measurement error is close to zero, and the decision on the existence of a bias or a fault in a sensor must be made based on two pieces of data: (1) a bias/fault estimate and (2) an uncertainty of this estimate. The latter requires a separate computation; again, the key for its use in real applications is to minimize the processing requirements. We derive an analytical recursive formulation, which updates the uncertainty for fault estimate sequentially, using minimum resources. With the fault estimate and its uncertainty in hand, a decision on fault existence can be made using a chi-square test or a similar approach. These algorithms have been implemented and tested on simulated and on real data. We present results that illustrate detection of small, hidden faults in the data.