This paper presents a combined approach for sensor fault identification looking for changes within one channel on one hand and for changes between the different channels on the other hand. The first method is based on the identification of autoregressive (AR) models from the reference time signals for each sensor channel separately. The reference models are then used for the prediction of the future sensors signals. The statistical properties of the residuals between this prediction and the true measurement allow a statement about changes of the sensor signals. The second method is based on the concept of mutual information between two signals X and Y from two different sensors. Mutual information or transinformation measures the information about the channel X that is shared by Y. This requires a certain redundancy of information represented in the different sensor signals. It can be seen that the mutual information changes as soon as a sensor fault occurs because the sensor fault information is not present in the other sensor signals.