燃气轮机非线性故障诊断中梯度计算的新方法

Gradient Calculation in Nonlinear Component Fault Diagnosis of Single-shaft Gas Turbine

针对PG9171E型单轴发电用燃气轮机,建立了一套非线性故障诊断方法。该诊断方法主要由两部分组成:燃机仿真模型和故障诊断目标函数。在对目标函数的优化计算过程中采用了梯度法,梯度在数值上等于目标函数对各自变量的偏微分值的负数。为了减少计算量,传统方法常用差分值代替偏微分值,但实际计算过程中发现,用差分代替偏微分会给计算结果带来不可忽略的误差,而且该误差无法通过增加计算次数消除。文中采用三点拟合的方法近似地求取目标函数的偏微分值,实际应用显示该方法并不会明显增加计算量,但却较大地提高了非线性故障诊断方法的精度。

A nonlinear method for gas path analysis was presented for PG9171E gas turbine. It used nonlinear techniques to determine the values of health parameters, with the help of a suitably formulated engine simulation model. Took parameter deviations as independent parameters, the differences between model outputs and measurements as object, an object function was built. Gradient method was used to process the object function. For the purpose of reducing computational workload, differences have been usually used in gradient method to substitute partial derivatives. Nonetheless, the application showed that this substitution would induce a big error. Thereby, three point calculating method was proposed to calculate partial differential values. The result showed that this new method could significantly increase the accuracy of nonlinear diagnostics, and the computational workload was not increased.