As we have seen from past incidents, nuclear reactors failing can lead to extremely disastrous consequences, consequences that we are still dealing with decades later, and possibly for decades to come. However it is possible that these failings could be reduced or possibly eliminated in the future, thanks to the use of AI.
Researchers at the Lyles School of Engineering at Purdue University have developed an AI system (via Futurity) that can help to look out for physical cracks in nuclear reactors, which could not only help reduce accidents and prevent them from happening early on, but could also reduce maintenance costs if these cracks are discovered early.
This is done through a “deep learning” framework that will analyze individual video frames for crack detection. From there, a “data fusion scheme” will aggregate the information from each video frame, and can detect cracks in overlapping patches in each video frame. So far it seems that the system has a 98.3% success rate, which according to assistant professor Mohammad R. Jahanshahi is significantly higher than other approaches.
He adds, “Regular inspection of nuclear power plant components is important to guarantee safe operations. However, current practice is time-consuming, tedious, and subjective and involves human technicians reviewing inspection videos to identify cracks on reactors.”