Artificial neural networks for transfer aligment and calibration of inertial navigation systems

Abstract

A new technique is developed for the transfer alignment and calibration of the strapdown inertial navigation systems. The alignment and calibration process is treated as the training of the neural network filter residing between the inertial measurement unit and the navigation computer. The filter takes erroneous inputs from the inertial measurement unit, and after correcting feeds them to the navigation computer. The training of the filter is carried out using the outputs of another inertial measurement system, which is known to be accurate. For this purpose a training methodology is developed which can handle, misalignments, scale factor errors and bias errors of the accelerometers and gyroscopes used in the inertial measurement units. Proper filter architectures are investigated and a suitable one is proposed. The training is done using Hide-and-Seek, which is a continuous simulated annealing algorithm. The algorithm is robust and quite effective in finding the global maximum. The effectiveness of the technique is demonstrated through a case study.