AIGC Enhanced Anomaly Detection in Industrial Internet of Things

With the advent of Industry 4.0, the ubiquity of sensors in large scale industrial equipment increased exponentially. This flood of cheap and reliable sensors allowed monitoring of critical equipment with real-time cloud access to the readings. With most industries squarely in the middle of this paradigm, the ability to utilize this sensor data is at the forefront in many industries. The methodology utilized to accomplish predictions typically revolves around semi-supervised learning, where models are only trained on good data. The reason this methodology is needed is due to the lack of anomalous data to train models on. This is not ideal in some cases, as supervised learning can be a powerful tool, particularly when there is a desire to predict Remaining Useful Life (RUL). Recent advances in generative artificial intelligence–such as diffusion learning–potentially enables the ability to increase the training set in such a way that supervised learning can be used in this anomaly detection framework. This work investigates the use of Artificial Intelligence Generated Content (AIGC) for the purpose of providing augmented training samples. We demonstrate that current methods can generate extremely high quality training samples in this context and improve predictions metrics.

Ҳали таржима қилинмаган