基于神经网络的动态轧制中摩擦因数仿真估算

Simulating Estimate Based on Neural Network for Friction Factor in Dynamic Rolling Process

基于多层前馈人工神经网络原理,研究了冷轧机组动态轧制过程中轧辊与轧件间摩擦因数的估算新方法.将轧制过程中摩擦因数视为轧制温度、润滑剂黏度、轧制速度等多参量共同作用的多变量非线性函数,采用神经网络非线性映射功能,构建了摩擦因数与影响参数之间的复杂函数关系的数学模型,提出了轧制过程中动态摩擦因数的神经网络插值逼近算法.同时编制程序进行多参数共同作用下的摩擦因数实例仿真分析,结果表明研究方法的可行性.

Based on feed forward multi-layer artificial neural network, a new way to estimate the friction factor between roller and workpiece was studied. The factor was taken as a multi-variable non-linear function which was affected jointly by such independent parameters as rolling temperature, lubricant viscosity and rolling speed. The nonlinear mapping of artificial neural network was used to construct a complicated mathematical model in terms of dynamic friction factor and influencing parameters. An interpolation-approximation method based on ANN was proposed and programmed to estimate the working friction factor in rolling process, with a simulation done as an example. The results showed that the new method provides a means available to control and stabilize effectively the rolling process without slippage and prevent it from unbalanced load.