基于K-近邻隶属度模糊支持向量机的再造抽油杆损伤等级磁记忆定量识别

Quantitative MMM identification of damage levels based on KNN FSVM for remanufactured sucker rod

针对磁记忆技术在再造抽油杆损伤等级定量评价中小样本和分散性的难题,从再造抽油杆疲劳损伤实验出发,通过获取损伤演化过程的磁记忆特征规律,提取五维磁记忆特征向量,首次引入基于K-近邻隶属度的模糊支持向量机多分类算法,结合参数组合寻优方法,建立再造抽油杆损伤等级的多分类磁记忆定量识别模型。结果表明:利用K-近邻隶属度将分散性和模糊性加以量化,结合支持向量机的小样本优势,构造的K-近邻隶属度模糊支持向量机多分类磁记忆模型,可以进行再造抽油杆损伤等级的磁记忆定量识别,并进一步进行参数组合寻优,避免了盲目选择固定参数导致模型精度过低,具有较好的抗噪性和鲁棒性,为再造抽油杆损伤等级定量评价提供了一种新的方法。

Data dispersion and small-and medium-sized samples incur the difficulty in quantitative identification of damage levels for remanufactured sucker rods by applying metal magnetic memory(MMM)technology.To solve the bottleneck,based on the fatigue and damage experiment of remanufactured sucker rods,the MMM characteristic rule of damage evolution is obtained to further extract five dimensional MMM parameters.A new multi-classification algorithm of the fuzzy support vector machine based on the membership degree obtained using k-nearest neighbor method are first put forward in this study.On this basis,the multiple classification MMM model is established for quantitative identification of damage levels for remanufactured sucker rods using parameter combination optimization method.The result shows that this membership degree can be used to quantify the data dispersion and fuzziness,and then a multi-classification MMM model based on the fuzzy support vector machine with the membership degree can be established by utilizing the advantage in solving small sample.This model can perform quantitative identification of the damage level for remanufactured sucker rods;parameter combination optimization is further implemented to avoid the over-low precision caused by blind selection of fixed parameter.It also has better robustness and anti-noise property,providing a new method to quantitatively identify the damage level for remanufactured sucker rods.