Improving Optical Fiber Current Sensor Accuracy using Artificial Neural Networks to Compensate Temperature and Minor Non-Ideal Effects

This article presents a practical signal processing methodology, based on Artificial Neural Networks - ANN, to process the measurement signals of typical Fiber Optic Current Sensors - FOCS, achieving higher accuracy from temperature and non-linearity compensation. The proposed idea resolve FOCS primary problems, mainly when it is difficult to determine all errors sources present in the physical phenomenon or the measurement equation becomes too nonlinear to be applied in a wide measurement range. The great benefit of ANN is to get a transfer function for the measurement system taking in account all unknowns, even those from unwanted and unknowing effects, providing a compensated output after the ANN training session. Then, the ANN training is treated like a black box, based on experimental data, where the transfer function of the measurement system, its unknowns and non-idealities are processed and compensated at once, given a fast and robust alternative to the FOCS theoretical method. A real FOCS system was built and the signals acquired from the photo-detectors are processed by the Faraday's Laws formulas and the ANN method, giving measurement results for both signal processing strategies. The coil temperature measurements are also included in the ANN signal processing. To compare these results, a current measuring instrument standard is used together with a metrological calibration procedure. Preliminary results from a variable temperature experiment shows the higher accuracy, better them 0.2% of maximum error, of the ANN methodology, resulting in a quick and robust method to hands with FOCS difficulties on of non-idealities compensation.