China Mechanical Engineering ›› 2026, Vol. 37 ›› Issue (5): 1017-1025.DOI: 10.3969/j.issn.1004-132X.2026.05.001

   

A Probabilistic Fatigue Life Prediction Method for Wind Turbine Towers Based on a Physics-informed Neural Network

XIE Bingbing1(), ZHAO Feng2,3, GUO Xinxing2,3, QIAO Li1, CHENG Sichuang2,3, LIU Xiaohui1, ZHANG Tongzhou2,3, HU Weifei2,3()   

  1. 1.CRRC Qihang New Energy Technology Co. ,Ltd. ,Beijing,100192
    2.State Key Laboratory of Fluid Power and Mechatronic Systems,Zhejiang University,Hangzhou,310058
    3.School of Mechanical Engineering,Zhejiang University,Hangzhou,310058
  • Received:2025-09-03 Online:2026-05-25 Published:2026-06-09
  • Contact: HU Weifei

基于物理信息神经网络的风电机组塔筒概率疲劳寿命预测方法

谢冰冰1(), 赵峰2,3, 郭昕兴2,3, 乔莉1, 程思创2,3, 刘晓辉1, 张桐舟2,3, 胡伟飞2,3()   

  1. 1.中车启航新能源技术有限公司, 北京, 100192
    2.浙江大学流体动力基础件与机电系统全国重点实验室, 杭州, 310058
    3.浙江大学机械工程学院, 杭州, 310058
  • 通讯作者: 胡伟飞
  • 作者简介:谢冰冰,男,1991年生,博士、工程师。研究方向为风力发电机整机设计、整机载荷及动力学仿真。E-mail: xiebingbing.xny@crrcgc.cc
    胡伟飞*(通信作者),男,1985年生,研究员、博士研究生导师。研究方向为人工智能、不确定性优化设计、风能及数字孪生。E-mail: weifeihu@zju.edu.cn
  • 基金资助:
    国家自然科学基金(52275275);浙江省“尖兵”“领雁”研发攻关计划(2023C01008);中车集团重大项目(2024CYY023);中车集团重大项目(2025CXA290)

Abstract:

To address the limitations that the traditional fatigue design for wind turbine towers using deterministic S-N curves might not accurately quantify fatigue life dispersion, a probabilistic fatigue life prediction method was proposed based on physics-informed neural networks. By embedding the physical prior knowledge such as fatigue life dispersion, monotonicity and nonlinearity into the neural networks, a probabilistic prediction model capable of accurately quantifying uncertainty was constructed. Compared with traditional methods, the proposed method reduces the normalized root mean square error(NRMSE) by up to 31.58%. A 16 MW wind turbine simulation model was established in accordance with IEC standards, and tower load data were obtained by using Bladed software. Combined with wind-speed distribution, rain flow counting and the Miner rule, the probabilistic fatigue life prediction of the towers was achieved. The results show that the proposed method effectively characterizes the probabilistic features of fatigue damages, and the tower lifetime varies significantly with reliability requirements (shortening from 83.3 years at 50% probability to 18.2 years at 99.9% probability), which provides a reliable basis for probabilistic fatigue design and safety assessment of wind turbine towers.

Key words: wind turbine, tower, probabilistic fatigue life, physics-informed neural network, uncertainty quantification

摘要:

针对风电机组塔筒传统疲劳设计采用确定性S-N曲线难以准确量化寿命分散性的问题,提出一种基于物理信息神经网络的概率疲劳寿命预测方法。通过将疲劳寿命分散性、单调性及非线性等物理先验知识嵌入神经网络,构建了可准确量化不确定性的概率预测模型。相比于传统方法,所提方法预测的归一化均方根误差(NRMSE)最多降低31.58%。基于IEC标准建立了16 MW风电机组仿真模型,通过Bladed软件获取塔筒载荷数据,结合风速分布、雨流计数与Miner准则实现了塔筒概率疲劳寿命预测。研究结果表明:所提方法有效表征了疲劳损伤概率特征,塔筒寿命随可靠度要求显著变化(从50%概率下83.3年缩短至99.9%概率下18.2年),为机组概率疲劳设计与安全评估提供了可靠依据。

关键词: 风电机组, 塔筒, 概率疲劳寿命, 物理信息神经网络, 不确定性量化

CLC Number: