This paper presents a physics informed artificial intelligence approach for reducing the dynamic error of a sensor, which receives vibration signals. Therefore, a two-stage process is proposed, which employs in the first step a Deep Neural Network (DNN) as an autoencoder which is pre-trained to the sensor’s physical forward dynamics. In the second step a deep learning algorithm iteratively reconstructs the vibration signal by adapting the previously defined underlying deep neural network weights. A case study is carried out by the analysis of a disturbed piezoelectric acceleration sensor signal. The validation results demonstrate that the proposed approach significantly reduces dynamic error and outperforms the most common deep learning approach.

dynamic error compensation,deep learning,vibration signal,inverse problem