基于遗传神经网络的旋转机械故障预测方法研究

许多大型旋转机械运行工况恶劣,非平稳、非线性特征明显,以及各种突发性、偶然性因素的影响,给基于振动信号处理的状态预测和状态维护分析带来困难。神经网络以其强大的处理非线性系统的能力在故障预测中得到广泛的应用,但由于其在追求高精度训练目标时易陷入局部极值,且收敛速度慢甚至发散。针对这个问题,提出了采用遗传算法对神经网络连接权值和阈值进行优化,这样不仅发挥了神经网络广泛的映射特性也使遗传算法的全局搜索优势尽显无疑。通过组合这两种算法,在提升网络学习的准确度方面,优点尤其突出,最终提高对旋转机械故障预测和寿命估计的性能,这在某环境模拟试验系统动力风机的轴承磨损故障预测中得到了验证。

Research on Mechanical Fault Prediction Based on Improved Neural Network[JZ)][HS)]

The representative characteristics of large-scale rotating machine in operation are non-stationary and nonlinear, and also influenced by sudden and accidental factors, thus the difficulty in condition monitoring and fault prediction based on vibration signal analysis is great. Artificial neural networks, which perform a nonlinear mapping between inputs and outputs, are widely used in fault prediction, but easy to fall into local optimal solution and converge with slow speed or even diverge. In this paper, aimed at this problem, the dynamic prediction model is studied,in which back propagation(BP) algorithm coupled with genetic algorithm(GA) will be used to train and optimize the networks. BP of ANN has been recognized as a powerful mapping approach to model extremely complex nonlinear process while GA for global search ability was used in various diverse optimization systems. Owing to complementary advantages of both merged, the accuracy of the GA-BP networks is improved significantly. The final goal is to improve the performance of GA-BP network in predicting faulty and estimating residual life for rotating machinery. Ultimately, verification of the optimization was showed at the bearing wear data from the power fan of a environmental simulation test system.