Figure 11: Heatmap of a flooding image. The neural net’s main focus is on the red area with decreasing importance towards the blue area. Grey areas are identified as unimportant for the classification problem. Here, the upper part of the flange is taken as most interesting area for the decision.
It can be seen that the neural net focuses on a reasonable section within the image to determine whether it sees an image of the normal operating state or flooding. With these two criteria satisfied, a first step towards an image-based control of an extraction column is made.