THE BIAS TEMPERATURE DEPENDENCE ESTIMATION AND COMPENSATION FOR AN ACCELEROMETER BY USE OF THE NEURO-FUZZY TECHNIQUES

In this paper, we describe a new method for improved performance of inertial sensors, with applications in strap-down inertial systems. A new empirical model is proposed for the bias temperature dependence compensation of accelerometers using their input and output data. Experimental testing of the accelerometer is first realized, as data for 2 inputs and 1 output are collected. Based on this data, an empirical model is built using a neuro-fuzzy network, which learns the process behavior and uses a Fuzzy Inference System (FIS) for model realization. The improvement in the reproduction quality of the experimental surface by the neuro-fuzzy model is achieved through the FIS training using a Sugeno learning algorithm with two inputs and one output. Generation and training of the FIS are performed with Matlab functions, the training of which is realized on a high number of epochs, for example, on a number of 105 training epochs. It is noticed that the proposed algorithm leads to a 35.5 times reduction in the ...