Research And Application Of Key Technologies For Fault Diagnosis Of Steel Rolling Furnace Based On Deep Learning

Building a smart steel factory is an important direction for steel industry to respond to "Made in China 2025".Hot rolling is the last process in the steelmaking-continuous casting-hot rolling process of steel production.The health status of its production line equipment is the important factor affecting the sustainable production of the enterprise.Realizing intelligent diagnosis of equipment fault status is the key to building a smart steel factory.However,the reliability of current fault diagnosis algorithms is not high,mainly reflected in:(1)The fault feature extraction method relies on expert knowledge in related fields,and has poor generality and generalization.(2)The feature extraction ability is poor,and it is impossible to extract deep fault features in complex sensor signals.(3)For equipment with multi-sensor monitoring in the production line,the lack of multi-sensor data fusion analysis methods is easy to ignore the correlation between multi-source data,which affects the determination of fault status.In view of these key issues,the research contents of this paper are as follows:Firstly,the overall architecture of the fault diagnosis system of the steel rolling furnace is constructed.In combination with the importance of equipment fault diagnosis to realize the fusion of cyber and physical in the steel hot rolling production line,the research on the fault diagnosis algorithm based on the multi-sensor monitoring furnace as the analysis object is determined.The mechanical structure of the heating furnace and the characteristics of multi-sensor monitoring are analyzed,and a fault diagnosis model is proposed that not only needs to extract the fault features in the sensor signals,but also needs to extract the correlation features between the multi-sensor signals.Based on this,the fault diagnosis process and system architecture of the heating furnace are designed.Secondly,the fault feature extraction algorithm based on multi-scale convolutional neural network(Multi-scale CNN)is studied.Based on the convolutional neural network,multi-scale convolution kernels are used to extract the features of the sensor's original signals to increase the diversity of features,and to suppress over-fitting of the model by adding BN layers and Dropout.The multi-sensor signal of the heating furnace is used as the input of the model,and the feature extraction capability of the model is verified through experiments to achieve the deep extraction of fault features in the multi-sensor signal and eliminate the dependence on expert knowledge.Finally,the fault feature fusion method based on ensemble deep learning is studied and a fault diagnosis model is constructed.A fault feature correlation analysis method based on multi-layer bidirectional long short-term memory network(Bi-LSTMs)is proposed to realize the extracting of the correlation characteristics between multiple sensors.The Stacking ensemble learning strategy is used to fuse the Multi-scale CNN and Bi-LSTMs,so as to realize the fusion of fault features and correlation features,and build a furnace fault diagnosis model based on ensemble deep learning.Then,the application process and application effect of the fault diagnosis model are analyzed.