基于RBF-NN近似模型的地铁钢轨探伤车转向架悬挂参数多目标优化#br#

以某地铁钢轨探伤车为研究对象，利用多体动力学软件UM建立车辆-轨道耦合动力学模型，分析探伤车在直线、曲线两种工况下的运行平稳性、安全性和曲线通过能力，为后续悬挂参数多目标优化提供原始数据支撑。基于MATLAB编程建立UM-ISight 联合仿真平台，在ISight 中以转向架一、二系悬挂刚度、阻尼等为目标参数，通过最优拉丁超立方法进行参数组合样本设计，并在此基础上构建符合精度要求的径向基函数神经网络(RBF-NN)近似模型，最后通过多目标优化算法NSGA-Ⅱ对近似模型进行寻优计算，得到最优的转向架悬挂参数组合。结果表明，所建近似模型具有较高的拟合精度，优化后车辆平稳性指标都得到明显改善，最多可降低35.39 %，且车辆的脱轨系数、轮轨横向力、轮轴横向力、轮重减载率等曲线通过性能指标也有不同程度好转。

Detection and Identification of Acoustic Fault of Underwater Navigational Object Based on KNN Classification Algorithm

In this paper, a method of detection and classification of acoustic fault of underwater navigational objects is proposed. Firstly, a benchmark acoustic database of the collected acoustic signals of underwater navigational object in normal working condition is constructed. Scan analysis of these acoustic signals is made, and the benchmark threshold value of the signals is determined. The signals deviating from the threshold are extracted as the acoustic faults. Secondly, the database of characteristic classification of the acoustic faults is established based on the analysis of the historical data of the acoustic faults of the underwater navigational objects. Then, based on the acoustic fault database, the faults modes are identified and classified using K-Nearest Neighbor （KNN） algorithm. Finally, the feasibility of this method is verified by a tank experiment.