Research On Soft-Measuring Of Aluminum Strip's Grain Size Based On Mnn Fusion

During the process of aluminum electromagnetic casting, the grain size is a important index for measuring the aluminum strip's quality. In order to control the aluminum electromagnetic casting system optimally, the key is to measure the grain size of aluminium strip online. However, until now there is no any instrument can be used to measure the value of grain size directly, only rely on some offline method such as microstructure analysis, whose big lag will largely affect the system's optimal control. With maturity of soft sensor technology in solving parameter measurement issues, this measurement technique is introduced to achieving online measurement of grain size.Based on the process mechanism of aluminum electromagnetic casting, the working principle of electromagnetic casting system is analysed in detail, and the influencing factors of grain size is deeply studied which is contributed to selecting secondary variables for soft sensor model.According to the complex mechanism and high nonlinear characteristic of electromagnetic casting process, multiple neural networks (MNN) is adopted to establish soft measurement model for aluminum strip's grain size. After normalization and principal component analysis (PCA), sample data is divided into four sample subspaces through K-means clustering algorithm, which are used to create different sub-neural networks, and finally obtained the predictive value by weighted fusing the sub-network output.Sub-network uses BP or RBF neural network. Different combinations models of MNN grain size soft-sensor are simulated by MATLAB, and the results show that comparing with single neural network, MNN model can reduce the predictive risk and improve forecasting accuracy based on same modeling time; for the soft measurement problem of aluminum strip's grain size, the model combining two RBF and two BP networks is more robust and accurate.