Realization of shock accelerometer sequence-to-sequence calibration based on deep learning

Abstract

The sensitivity parameters or mathematical model of sensor are obtained through traditional shock accelerometer under calibration test, and then the measurement signal is restored through compensation and correction. As the shock signal with complex components is used in a harsh environment, it is very difficult to accurately restore the measured signals. In the meantime, the deconvolution process of signal recovery is an inverse problem in mathematical physics, which has ubiquitously ill-posed problems, thereby bringing great challenges to the accuracy and precision of the solution. Thus, we propose a depth calibration network based LSTM, which can be used to learn the mapping relationship between the calibrated sensor signal and the standard signal. The measured signal can be restored through a data-driven sequence-to-sequence calibration network, which is trained and verified through the exclusive open-source data set of shock signals. The test results proved the superior performance of the network in shock signal calibration.