Autoencoder-Based Restoration of Multi-Channel Sensor Signal Loss

We propose a method for restoring lost values in multi -channel sensor signals when specific channels or values are missing by using an autoencoder model. For this purpose, an autoencoder model was trained using normal data and then used to predict the values of the missing channels. Evaluation results showed that the restoration approximated the original values and patterns by utilizing information from the non -missing channels. Additionally, the restoration performance varied, depending on the correlations among different channels. The proposed method can enhance the overall validity of a dataset and contribute to the improvement of the data restoration capability in situations of sensor failures or data loss.