Research on Self-iterative Method of Vibration Signal Classification Model Based on Incremental Learning of Critical Equipment in Nuclear Power Plant

In order to promote the application of fault diagnosis or prediction models in the field of nuclear power engineering, and realize the long-term operation and continuous iteration of the edge end algorithm model with equipment, this paper proposes a self-iterative method that applies the incremental learning idea to traditional machine learning or deep learning. Firstly, the convolutional neural network established a initial model for intelligent identification of vibration signals, and then the incremental learning method based on fine tuning is used to enable the model to continuously receive training samples, Overcoming the "disastrous forgetting" problem in the learning process and realizing the model self-iteration; Finally, the model self-iteration process is trained, tested and verified by using the labeled vibration signals in the operation of a nuclear power plant. The results show that the proposed incremental learning process can realize the model self-iteration and continuously improve the accuracy while saving computing resources.