Mid-Infrared Monitoring and Image Processing in a Rotary Combustor ^†

The impact of the declining domestic workforce in Japan is becoming increasingly serious, making it difficult to secure workers for the operation and maintenance of waste treatment facilities. To cope with this situation, there is an urgent need to develop technologies that can support and automate the operation of waste incineration facilities by utilizing techniques such as IoT and AI.

We at KOBELCO Group have been developing combustion control techniques for rotary combustors (rotary stoker-type) to make their operation automatic, instead of manual by workers. As an effort to contribute to the automatic operation of these combustors, we attempted to observe the inside of the rotary combustor using a mid-infrared camera with a passing-through-flame bandpass filter, as shown in Figure 1; it is difficult to see inside the combustor using visible light due to the influence of flames and gasses.

As a result of mid-infrared monitoring, it was confirmed that the refuse layer, which is fed from the hopper during the process of combustion, could be clearly observed. To evaluate the condition of the rotary combustor, image segmentation was applied to the mid-infrared images using a deep learning model (Mask2Former [1]) by fine-tuning it to our data set. As shown in Figure 2, we confirmed that the model could identify the refuse layer and rotary tube (combustor structure) accurately in the mid-infrared image. In this presentation, a technique for monitoring the combustor’s condition using the image segmentation results of mid-infrared images is reported.