神经网络建模在热膨胀螺栓形变测量中的应用

Neural net based Modeling Used in the Protraction Measurement for Thermal Expansion Die bolt

基于神经网络建立热膨胀螺栓形变的非线性数学模型。神经网络的辨识采用变尺度二阶快速学习算法 ,利用二阶插值法来优化搜索学习速率。新方法具有很快的收敛速度和良好的收敛精度 ,克服了 BP算法在神经网络的权值训练中收敛速度过慢的缺点。热膨胀螺栓的受热形变测量结果表明 。

Based upon neural networks, a nonlinear mathematical model of the protraction for the thermal expansion die bolt is developed. A kind of variable metric fast second order learning algorithm was proposed for the identification of neural network. The second order interpolating method is used in the optimization of learning rate. The new algorithm has fast convergence rate and good precision. Therefore, it overcomes the drawback of BP algorithm which converge too slowly during the weights training of neural network. The method was applied into the protraction measurement for the thermal expansion die bolt . The research result shows that the method is suitable for the modeling and identification of nonlinear system.