This paper presents neural network regression models for predicting the nonlinear static and linearized dynamic reaction forces of spiral grooved gas journal bearings. The partial differential equations (PDEs) are sampled, based on a full factorial and randomly spaced parameter set. Feed-forward neural network (FNN) architectures are developed for modeling the PDEs and therefore replacing the time-consuming discrete and iterative solution procedure used to this date. A significant speed-up factor of >10³ in computation time is achieved, compared to solving the PDE numerically. Furthermore, the FNN allows for multi-dimensional interpolation, which makes global system optimization easily possible. This is demonstrated by a real-case rotordynamic system optimization. By using the neural network meta-models, a complete rotordynamic system optimization time reduction of factor 300 is achieved.

本文提出了神经网络回归模型，用于预测螺旋槽式气体轴颈轴承的非线性静态和线性动态反作用力。基于完整的阶乘和随机间隔的参数集，对偏微分方程（PDE）进行采样。前馈神经网络（FNN）架构是为PDE建模而开发的，因此替代了迄今为止使用的耗时的离散和迭代求解程序。明显的加速因子> 10 ³与数值求解PDE相比，可以节省计算时间。此外，FNN允许进行多维插值，从而可以轻松实现全局系统优化。实际情况中的转子动力学系统优化证明了这一点。通过使用神经网络元模型，可以将转子动力学系统的优化时间减少300倍。