军事知识服务平台
基于LDA的水电厂房结构振动局部特征强化与损伤定位方法
Method for local feature enhancement and damage localization in hydropower factory building structural vibration based on LDA
摘要 随着抽水蓄能电站等水利工程的规模扩大及振动环境日趋复杂,水电站厂房结构的抗振安全性问题日益突出.提出一种基于线性判别分析(LDA)的厂房结构振动局部特征强化与损伤定位方法,构建非监督学习框架,首先利用随机减量技术(RDT)消除外部激励影响,随后引入 LDA 实现响应数据的降维投影,强化振动局部特征,最后结合平方马氏距离进行损伤量化与定位.通过对河口村水库大电站厂房进行数值模拟,验证了该方法的有效性.应用结果表明,RDT 算法能够有效提取原始响应中的自由振动响应,结合 LDA 与平方马氏距离能够准确量化和定位厂房单一梁柱损伤或多处梁柱损伤,所提出方法仅依赖未损伤状态的响应数据,即可完成模型训练.
Abstract:
With the expansion of hydropower projects such as pumped storage power stations and increasingly complex vibration environ-ments,the seismic safety of hydropower station powerhouse structures has become increasingly prominent.A method for enhancing local vi-bration characteristics and locating damage in powerhouse structures based on Linear Discriminant Analysis(LDA)was proposed.An unsu-pervised learning framework was constructed.First,the Random Decrement Technique(RDT)was employed to eliminate external excitation influences,followed by the introduction of LDA to achieve dimensionality reduction and projection of response data,thereby enhancing local vibration characteristics.Finally,the squared Mahalanobis distance was combined for damage quantification and localization.Through numeri-cal simulations of the Heikou Village Hydropower Station Factory Building,the effectiveness of the proposed method was verified.Application results demonstrate that:The RDT algorithm can effectively extract free vibration responses from original responses.By combining LDA and the squared Mahalanobis distance,it accurately quantifies and locates single or multiple beam-column damages in the factory building.The proposed method relies solely on response data from undamaged states to complete model training.
刊 名 人民黄河 ISTICPKU
年,卷(期): 2026, 48(10)
分类号: TV731
机标分类号: TV731 TU311.3 TP391.41
在线出版日期: 2026年10月19日
基金项目: 2024YFC3214901:国家重点研发计划825024516:河南省财政厅资金资助项目B240201069:中央高校基本科研业务费专项,