A Deep Neural Network-Based Fault Detection Scheme for Aircraft IMU Sensors

Figure 4

Full architecture of the IMU fault detection network. (a) DNN-FULL architecture. ADS: airspeed, AOA, and sideslip angle measured by air data sensors; ANGLES: Euler angles; AS: angular speeds; ACC: accelerations; Conv2D: convolution layer; Dense: fully connected layer. kernels are also used in Maxpool layers. Flatten layers reshape 2D data to 1D for Dense layers. Concatenate layer merges all the data from previous layers. (b) Detailed configurations of CNN and LSTM modules in acceleration channel, kernel size, and bias are depicted. Configurations of other channels are the same as those of the acceleration channel.