纺织典型装备故障多特征自适应提取方法

Multi-fault feature adaptive extraction method for textile typical equipment

化纤高速卷绕头故障特征提取是卷绕头故障诊断中的关键步骤。为解决化纤卷绕头故障诊断精度不高、可解释性差的难点,提出了数据驱动的化纤卷绕头故障多特征自适应提取方法。通过改进型基因表达式编程(GEP)的故障特征生成方法,设计了一种运算符与变量符随机组合编码、对位匹配与倒序运算解码方法,构建了突变、插串、重组的遗传算子,实现了多个故障特征构建与生成;提出了低冗余、高互补的多特征提取与分析方法,实现了特征间关系的可解释性分析与关键特征优选。实验结果表明:采用所提出的改进型GEP方法与传统特征提取方法、通用GEP方法所生成的故障特征进行对比,在线速度为1000、2000和3000 m/min状态下,卷绕头故障诊断精度分别提升了8.959%、3.87%、3.77%和2.601%、3.2%、2.018%,有效解决了卷绕头故障特征提取的难题;进一步的特征工程实验表明,所提方法对于多特征组合下的卷绕头故障关键特征提取具有较强的适应性能。

Objective Chemical fiber winder is the core equipment in chemical fiber production.The failure of winder will seriously affect the quality and production efficiency of chemical fiber products,so it is necessary to accurately diagnose the fault of chemical fiber winder.Fault feature extraction is the premise of winder fault diagnosis.It is mainly divided into empirical knowledge-based methods and data-driven methods.Aiming at the problem of low accuracy of empirical knowledge features in fault diagnosis of winder and poor interpretability of data-driven features,this paper proposes a data-driven method for adaptive extraction of multiple fault features of chemical fiber winder.Method The proposed adaptive feature extraction method of chemical fiber winder fault based on improved gene expression programming(GEP)with multi-fault feature correlation analysis and subset evaluation method,which includes gene encoding and decoding of feature initialization,feature subset correlation analysis and decision tree evaluation,roulette screening and Elite retention strategies,feature optimization method based on genetic evolution.Among them,the multi-feature correlation analysis method is combined with the experience and knowledge features to select the advantages of strong correlation,low redundancy and high complementarity.When the last iteration is completed,the dominant feature subset output forms the final feature.Results In order to verify the effectiveness of the proposed improved GEP feature extraction method in industrial applications,the measured vibration data of POY-1800 winder in a chemical fiber enterprise in Zhejiang province are used to test the performance of the proposed feature extraction method.The fault characteristics of 14 winders in the production process of one kind of fiber are extracted.The vibration acceleration sensor was used to collect the vibration data during the rotation of the winder,and the feature extraction test was carried out under the instantaneous linear speed of 1000 m/min,2000 m/min and 3000 m/min.The sampling frequency is 51200 Hz,and the collection time of each class is 1 s.There are 4 categories in total,each of which is a binary classification task.The verification was carried out by using the multi-round cycle data set,and the original data was processed in segments according to every 20 points.The number of individuals in the population is set to 100,the number of iterations is 50,and the outermost cycle is 3 rounds to set the optimal individual retention mechanism.In the experiment,the proposed method was compared with the method based on empirical knowledge features and the general GEP method without using multi-feature association analysis.The extracted features are input into the classifier formed by C4.5 decision tree algorithm,and the effect of each method is compared by classification accuracy.To facilitate the observation of the results,the average classification accuracy AVG and the BEST classification accuracy best during GEP are recorded.The experimental results show that compared with the fault features generated by the traditional feature extraction method and the general GEP method,under the line speed of the winder in 1000 m/min,2000 m/min,3000 m/min,the fault diagnosis accuracy of the proposed improved GEP method is increased by 8.959%,3.87%,3.77% respectively 2.601%,3.2%,2.018% respectively,which effectively solves the problem of fault feature extraction of the winder.Conclusion In this study,a data-driven multi-fault feature adaptive extraction method for chemical winder is proposed.Contrast experiment results demonstrate the proposed GEP-based interpretable feature extraction method with experience and knowledge features is effective in improving the accuracy of fault diagnosis;The outcomes of classification accuracy at various speeds illustrate the proposed multi-feature correlation analysis method is validate in augmenting the adaptability of winding scenarios;Subsequent experimental results in feature engineering affirm the proposed enhanced GEP feature extraction method is effective in diagnosing multiple faults of chemical fiber winders.