One Dimensional Integrated GPS - Odometer Navigation With Applications To Rail Track Analysis

摘要

Railroads, are using instrumented, sometimes unattended cars to measure the condition of the tracks along their rail beds. The track Geometry data is recorded every ft., or so, using a strobing function triggered by an odometer. On some systems, GPS is also recorded with a line of data. A Kalman Filter is developed which integrates GPS, the implied odometer, and possibly a measure of heading derived from the track geometry "curvature". It models gradual turns consistent with those possible on rail beds. The process limits the mathematical dynamics of motion allowing for robust data rejection, and it estimates the odometer error. This in turn can be used to precisely position the measurements in terms of distance along the track (or from reference point). That is, generally, less than 3 meters in 120 Km. It can also extrapolate during GPS outages. This filter is unconventional in that data is "position vs. position (or record number)", no time. GPS position is modeled as true position plus noise. Observability of odometer error and heading is through the dynamic equations, and the variability on observed position which is driven by the odometer error and heading. The extended Kalman filter developed (having states which include odometer offset (error) and heading) have all states modeled as changing over position. If there is no direct measure of heading, a special look ahead method is used. This minimizes the potential for the filter to go unstable. The measure of heading which may be available is the parameter "curvature" (Deg/ Unit distance). This can be treated similar to the way a rate gyro is used in GPS integration. Also it is possible to use a heading lookup table (tracks don't move!) Evaluation is carried out by comparing positioning outputs at (virtually) the same locations for two runs over the same track. The differences in estimated position at the same physical location are then measures of the errors on both runs. Proprietary software developed by Rome is used to line up Geometry Data on one run with respect to another to within one foot. The Similarity of gage and cross level over the runs are used to line up the data. Once data alignment is accomplished it is straight forward to compute difference of estimates at the same location.