基于AE重构误差和SVDD的滚筒叶丝干燥异常工况检测  

Anomaly detection for cut tobacco drum dryer based on AE reconstruction error and SVDD

作　　者：安佳敏 张雷 李善莲[1] 张二强 邹严颉 刘朝贤[1] AN Jiamin;ZHANG Lei;LI Shanlian;ZHANG Erqiang;ZOU Yanxie;LIU Chaoxian(Zhengzhou Tobacco Research Institute of CNTC,Zhengzhou 450001,China;Zhengzhou University of Light Industry,Zhengzhou 450002,China;Technology Center,China Tobacco Shanxi Industrial Co.,Ltd.,Xian 721013,China)

机构地区：[1]中国烟草总公司郑州烟草研究院,郑州450001 [2]郑州轻工业大学,郑州450002 [3]陕西中烟工业有限责任公司技术中心,陕西省宝鸡市721013

出　　处：《中国烟草学报》2023年第6期31-37,共7页Acta Tabacaria Sinica

基　　金：国家局烟草科研大数据重大专项项目“卷烟制丝加工大数据关键技术研究与应用”(110202101083(SJ-07));河南省科技攻关计划项目(212102210155);数据驱动的复烤过程状态监测与智能诊断技术研究(202022AWCX04)。

摘　　要：【目的】为保证烘丝过程安全稳定运行,研究滚筒叶丝干燥过程异常工况检测具有重大价值。【方法】本文提出基于自动编码器(Auto encoder,AE)和支持向量数据描述(Support vector data description,SVDD)的AE-SVDD算法。首先,使用深度学习自动编码器提取数据深层特征,构建重构误差,利用重构误差训练SVDD分类模型得到超球体半径阈值,建立检测率指标。通过工业实际生产案例进行模型验证,并应用PCA、SVDD算法分别建立异常检测模型作对比实验。【结果】基于AE-SVDD的算法模型检测率可提高约63%,并能预测4~8min后即将发生的质量异常,明显优于其他算法模型。【结论】与传统方法相比,AE-SVDD异常工况检测方法不仅显著提高了检测率,而且具有良好的异常工况预警作用,有助于及时发现、控制滚筒叶丝干燥过程潜在异常工况,降低质量异常的产生几率。[Background]In order to ensure the safe and stable operation of the drying process,it is of great significance to carry out anomaly detection of cut tobacco drum dryer during drying process.[Methods]The proposed AE-SVDD algorithm is based on auto encoder(AE)and support vector data description(SVDD).Firstly,auto encoder was used to extract the deep feature of data;meanwhile the reconstruction error was constructed.The SVDD model was trained using the reconstruction error so as to obtain the hypersphere radius threshold,and the detection rate was established.The feasibility of model was verified based on the actual industrial production data,and the PCA and SVDD algorithms were used to establish anomaly detection models for comparative experiments.[Results]The detection rate of the algorithm model based on AE-SVDD was improved by about 63%,and it well predicted the quality anomaly that occurred in 4-8 minutes,which is obviously superior to other algorithm models.[Conclusion]Compared with the traditional methods,AE-SVDD detection method for abnormal working conditions not only significantly improves the detection rate,but also has a good warning function for abnormal working conditions,which is helpful to timely find and control the potential abnormal working conditions in the drying process of drum dryer,and reduce the probability of quality abnormality.

关 键 词：滚筒烘丝机 自动编码器 支持向量数据描述 重构误差 异常检测 

分 类 号：TS452[农业科学—烟草工业]