Hydraulic Servo System Of TRT Unit Based On Neural Network Integrated Learning

The blast furnace top gas recovery turbine unit（TRT）is a device for energy recovery by virtue of the high pressure and high temperature of blast furnace gas,which can provide various countries all over the world with an effective treatment method for energy conservation,emission reduction and air pollution reduction.At present,the existing studies on fault detection of the TRT mostly focus on the faults caused by unit vibration,and rarely on the faults arising from non-vibration factors.Therefore,the present study starts with the analysis of faults caused by non-vibration factors,and adopts technical means in neural network for fault diagnosis of hydraulic servo system of the blast furnace.The relevant work is mainly as follows:1.For the faults caused by non-vibration factors of the TRT,a fault diagnosis system of TRT hydraulic servo system was established,and 11 fault types involved in the system were classified.Additionally,different types of fault causes were analyzed and determined by data processing technology,and their features were extracted.Due to the imbalance of fault data volume,it was impossible to simply classify the fault data with medium types by grouping,that is,they were divided into the same group number as the fault data with multiple types.In this paper,Bagging algorithm was employed for data preprocessing.2.In terms of the fault data structure of TRT hydraulic servo system,the model structure adopted in this paper integrated 1D-CNN module,Bi LSTM module and Dense module to learn the dependence and association of different node data and to eliminate the redundant data.Further,MATLAB toolbox was applied for network training,and the selection process of model parameters was analyzed and provided to determine the optimal layer parameters of each model.3.1D-CNN,Bi LSTM and Dense models were integrated by Bagging for ensemble learning,which was also adopted for fault diagnosis of TRT hydraulic servo system.The output result of strong classifier was finally determined by voting,so as to further determine the fault type.As ensemble learning has achieved a higher accuracy in the classification of samples in different categories,it can be seen that integrating the models with different performance by Bagging can maximize the performance of model groups on the dataset,and based on the idea of complementary advantages,the final model performance will be significantly better than that of a single model.Therefore,the method of ensemble learning can greatly improve the judgment accuracy of the system in case of any fault,and further reduce the input of human and material resources,shorten the maintenance time,and increase the service hours.