在偏航过程中偏航位置异常一方面会导致偏航位置误差积累,影响偏航对风精准度或导致电缆过度扭缆而影响安全,另一方面频繁位置跳变或频繁短时位置保持均会产生一定的偏航误差、影响偏航控制稳定性,从而导致偏航系统故障发生频率升高和...在偏航过程中偏航位置异常一方面会导致偏航位置误差积累,影响偏航对风精准度或导致电缆过度扭缆而影响安全,另一方面频繁位置跳变或频繁短时位置保持均会产生一定的偏航误差、影响偏航控制稳定性,从而导致偏航系统故障发生频率升高和运维成本增加等问题,因此提出了一种基于数据驱动的故障诊断方法,用于对风电机组偏航位置的异常情况进行预警。首先,针对数据采集与监视控制(Supervisory Control and Data Acquisition, SCADA)系统中的海量数据,采用基于标准化交互增益的Relief-F(Standardized Interaction Gain and Relief-F, SIG-Relief-F)特征算法筛选出用于识别与目标变量(在这种情况下可能是偏航系统故障)具有最强关联性的多个特征参数。这种方法的优势在于能够有效地考虑到特征之间的相关性,最大程度地保留偏航系统故障相关特征与交互特征。其次,建立反向传播神经网络(Back Propagation Neural Network, BPNN)偏航位置预测模型,通过滑动窗口法对残差的分布进行统计,从而确定故障阈值。最后,通过实例验证了所提方法的有效性与准确性,并通过对比多元状态估计技术(Multivariate State Estimation Technique, MSET)和支持向量机(Support Vector Machine, SVM)算法,验证了其具有更优的异常预警性能。研究结果可为实际偏航系统的故障诊断提供参考。展开更多
Fatigue damage is a primary contributor to the failure of composite structures,underscoring the critical importance of monitoring its progression to ensure structural safety.This paper introduces an innovative approac...Fatigue damage is a primary contributor to the failure of composite structures,underscoring the critical importance of monitoring its progression to ensure structural safety.This paper introduces an innovative approach to fatigue damage monitoring in composite structures,leveraging a hybrid methodology that integrates the Whale Optimization Algorithm(WOA)-Backpropagation(BP)neural network with an ultrasonic guided wave feature selection algorithm.Initially,a network of piezoelectric ceramic sensors is employed to transmit and capture ultrasonic-guided waves,thereby establishing a signal space that correlates with the structural condition.Subsequently,the Relief-F algorithm is applied for signal feature extraction,culminating in the formation of a feature matrix.This matrix is then utilized to train the WOA-BP neural network,which optimizes the fatigue damage identification model globally.The proposed model’s efficacy in quantifying fatigue damage is tested against fatigue test datasets,with its performance benchmarked against the traditional BP neural network algorithm.The findings demonstrate that the WOA-BP neural network model not only surpasses the BP model in predictive accuracy but also exhibits enhanced global search capabilities.The effect of different sensor-receiver path signals on the model damage recognition results is also discussed.The results of the discussion found that the path directly through the damaged area is more accurate in modeling damage recognition compared to the path signals away from the damaged area.Consequently,the proposed monitoring method in the fatigue test dataset is adept at accurately tracking and recognizing the progression of fatigue damage.展开更多
文摘在偏航过程中偏航位置异常一方面会导致偏航位置误差积累,影响偏航对风精准度或导致电缆过度扭缆而影响安全,另一方面频繁位置跳变或频繁短时位置保持均会产生一定的偏航误差、影响偏航控制稳定性,从而导致偏航系统故障发生频率升高和运维成本增加等问题,因此提出了一种基于数据驱动的故障诊断方法,用于对风电机组偏航位置的异常情况进行预警。首先,针对数据采集与监视控制(Supervisory Control and Data Acquisition, SCADA)系统中的海量数据,采用基于标准化交互增益的Relief-F(Standardized Interaction Gain and Relief-F, SIG-Relief-F)特征算法筛选出用于识别与目标变量(在这种情况下可能是偏航系统故障)具有最强关联性的多个特征参数。这种方法的优势在于能够有效地考虑到特征之间的相关性,最大程度地保留偏航系统故障相关特征与交互特征。其次,建立反向传播神经网络(Back Propagation Neural Network, BPNN)偏航位置预测模型,通过滑动窗口法对残差的分布进行统计,从而确定故障阈值。最后,通过实例验证了所提方法的有效性与准确性,并通过对比多元状态估计技术(Multivariate State Estimation Technique, MSET)和支持向量机(Support Vector Machine, SVM)算法,验证了其具有更优的异常预警性能。研究结果可为实际偏航系统的故障诊断提供参考。
基金funded by the Key Program of the National Natural Science Foundation of China(U2341235)Youth Fund for Basic Research Program of Jiangnan University(JUSRP123003)+2 种基金Postgraduate Research&Practice Innovation Program of Jiangsu Province(SJCX23_1237)the National Key R&D Program of China(2018YFA0702800)Key Technologies R&D Program of CNBM(2023SJYL01).
文摘Fatigue damage is a primary contributor to the failure of composite structures,underscoring the critical importance of monitoring its progression to ensure structural safety.This paper introduces an innovative approach to fatigue damage monitoring in composite structures,leveraging a hybrid methodology that integrates the Whale Optimization Algorithm(WOA)-Backpropagation(BP)neural network with an ultrasonic guided wave feature selection algorithm.Initially,a network of piezoelectric ceramic sensors is employed to transmit and capture ultrasonic-guided waves,thereby establishing a signal space that correlates with the structural condition.Subsequently,the Relief-F algorithm is applied for signal feature extraction,culminating in the formation of a feature matrix.This matrix is then utilized to train the WOA-BP neural network,which optimizes the fatigue damage identification model globally.The proposed model’s efficacy in quantifying fatigue damage is tested against fatigue test datasets,with its performance benchmarked against the traditional BP neural network algorithm.The findings demonstrate that the WOA-BP neural network model not only surpasses the BP model in predictive accuracy but also exhibits enhanced global search capabilities.The effect of different sensor-receiver path signals on the model damage recognition results is also discussed.The results of the discussion found that the path directly through the damaged area is more accurate in modeling damage recognition compared to the path signals away from the damaged area.Consequently,the proposed monitoring method in the fatigue test dataset is adept at accurately tracking and recognizing the progression of fatigue damage.
