Fault Prediction Method Based on Migration Convolutional Neural Network

Abstract

The invention belongs to the field of neural network fault prediction and discloses a fault prediction method based on a migration convolutional neural network. The method includes the following steps: (a) number the fault type, collect the time domain signal of the object to be predicted and obtain the initial fault type number, convert the time domain signal into an RGB image; (b) convert the FC of the deep residual network model Initialize the layer and add a classifier to obtain an improved network model; (c) input the RGB image into the network model to train the FC layer and classifier, and continuously update the weight value of the FC layer. When the obtained fault type number is similar to the initial fault type number The corresponding weight value is the required new weight value, and the migration of the network model is completed; (d) Input the RGB image of the object to be predicted into the migration convolutional neural network model, and output the number of the predicted fault type. Through the present invention, the adopted migration convolutional neural network model has simple structure, fast prediction speed and accurate prediction result.

Description

technical field [0001] The invention belongs to the field of neural network fault prediction, and more specifically relates to a fault prediction method based on a migration convolutional neural network. Background technique [0002] In recent years, many researchers have studied fault prediction. As a typical fault prediction method, data-driven fault prediction can use historical data to establish fault modes without any explicit models or signal symptoms, which is very suitable for complex systems. , with the rapid development of intelligent manufacturing, the data generated by machinery and equipment has been well improved and collected. Mechanical big data has brought new opportunities for the manufacturing industry to achieve a trouble-free process. Data-driven failure prediction is becoming more and more popular. The attention of researchers and engineers is critical to finding more robust data-driven failure prediction methods. [0003] Learning from a large amount ...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the above defects or improvement needs of the prior art, the present invention provides a fault prediction method based on the migration convolutional neural network. The method first converts the time-domain signal into an RGB image, and then migrates the convolutional neural network to obtain The migrated convolutional neural network, and then use the migrated convolutional neural network to predict faults, thus solving the technical problem of low fault prediction accuracy

Images