基于声学特征提取和支持向量机的风机叶片缺陷识别

Wind Turbine Blade Defect Recognition Based on Acoustic Feature Extraction and Support Vector Machine

为了诊断风机叶片工作中出现的腐蚀、裂纹、穿孔甚至断裂等故障,我们提出了一种基于声音特征提取的风机叶片健康监测和预警方法。在运行过程中收集风扇叶片的声音信号并对声音信号进行预处理,然后采用六分之一倍频程划分频带,将各频带能量作为特征数据,训练支持向量机二分类模型,采用的风机声音特征样本共4762个,其中有缺陷样本3341个,随机抽取百分之七十的数据作为训练集,剩下的则为测试集。对训练后的模型进行测试,测试集总正确率达到95.91%。实验证明该方法得到的风力发电机叶片健康状况分类结果可靠性高,可以达到无接触实时监测目的。

For diagnosis of the wind turbine blade in the work of such as corrosion, crack, perforation and fracture failure, we put forward a kind of acoustics-based feature extraction method of wind turbine blade health monitoring and early warning. We first collect the voice of the fan blade run time signal, and the acoustic signal preprocessing, and then use one-sixth octave band, divided into each frequency band energy as the feature data, training support vector machine (SVM) classification model, USES the fan noise features a total of 4762 samples, including defect sample of 3341, according to the 7:3. The training set and the test set were randomly divided in proportion to each other. After the training model was tested, the total correct rate of the test set reached 95.91%. The experimental results show that the classification results of wind turbine blade health condition obtained by this method are reliable and can achieve the purpose of non-contact real-time monitoring.