为解决卷烟滤棒生产过程中需要人工对滤棒指标进行监测并调整设备参数，导致成本高昂且效率低下的问题，提出一种多尺度特征融合注意力机制的滤棒质检算法。该算法利用不同大小的一维卷积获取滤棒多尺度的特征，降低了滤棒局部特征遗漏的可能性。为进一步增强特征表示，本文引入注意力机制使算法模型聚焦于更重要的特征。实验结果表明，与一维卷积神经网络（1D convolutional neural network, 1DCNN）、反向传播（back propagation, BP）神经网络、循环神经网络等5种方法相比较，本文提出的算法模型在测试集上的性能更加突出，尤其是与1DCNN、BP神经网络和分布式梯度增强库这3种方法相比，模型的准确率分别提高了3.27%、4.28%和5.70%；精确率分别提高了3.12%、4.68%和5.10%；召回率分别提高了3.28%、4.57%和2.97%； F1-Score分别提高了3.20%、4.13%和4.33%。本文提出的算法不仅可以提高滤棒的生产效率，还可以降低人工成本，具有良好的工程实用价值。
To address the high cost and low efficiency caused by manual monitoring and adjustment of equipment parameters in the production of cigarette filter rods, a filter rod quality inspection algorithm with multi-scale feature fusing attention mechanism is proposed. The algorithm uses one-dimensional convolutions of various sizes to obtain the multi-scale features of the filter rod, which greatly reduces the possibility of missed detection of local features. To further enhance the feature representation, this paper introduces an attention mechanism, enabling the model to focus on regions with richer feature information. Experimental results show that compared with five existing methods such as 1D convolutional neural network (1DCNN), back propagation (BP) neural network and recurrent neural network (RNN), the proposed algorithm shows superior performance on the test set, especially compared with the three methods of 1DCNN, BP neural network and extreme gradient boosting. Specifically, the accuracy of the model increases by 3.27%, 4.28% and 5.70%; the precision improves by 3.12%, 4.68% and 5.10%; the recall rate rises by 3.28%, 4.57% and 2.97%; and the F1-Score enhances by 3.20%, 4.13% and 4.33%, respectively. In general, the algorithm proposed in this paper not only increases the production efficiency of filter rods, but also reduces labor costs, demonstrating practical engineering value.
[1] 常纪恒, 邢军, 于川芳, 等. 滤棒成形工艺技术与质量稳定性研究[C]//中国烟草学会2004年学术年会论文集, 2004: 244-255.
[2] 曹毅, 金绍平, 王斌, 等. KDF4成型机组滤棒吸阻影响因素及控制方法研究[J]. 中国新技术新产品, 2021(23): 42-45. Cao Y, Jin S P, Wang B, et al. Research on the influencing factors and control methods of filter rod suction resistance of KDF4 forming unit [J]. China New Technology and New Products, 2021(23): 42-45. (in Chinese)
[3] 尚跃, 刘与铭, 胡世龙, 等. 滤棒储存环境条件与滤棒关键指标相关性研究[J]. 轻工科技, 2021, 37(10): 79-82. Shang Y, Liu Y M, Hu S L, et al. Research on the correlation between filter rod storage environmental conditions and filter rod key indicators [J]. Light Industry Science and Technology, 2021, 37(10): 79-82. (in Chinese)
[4] 常纪恒, 常建勇, 盛培秀, 等. 丝束卷曲特性与滤棒物理特性的关系[J]. 烟草科技, 2011, 44(8): 9-12. Chang J H, Chang J Y, Sheng P X, et al. Relationships between tow crimp and physical characteristics of filter rod [J]. Tobacco Science & Technology, 2011, 44(8): 9-12. (in Chinese)
[5] 常纪恒, 阮晓明, 赵荣, 等. 滤棒物性参数之间的相关关系[J]. 烟草科技, 2003, 36(10): 9-12. Chang J H, Ruan X M, Zhao R, et al. Relationships among physical parameters of filter rod [J]. Tobacco Science & Technology, 2003, 36(10): 9-12. (in Chinese)
[6] Zhao L, Ying W, Liu L W, et al. Study on resistance characteristic of groove filter rod [C]//Advanced Science and Industry Research Center: Science and Engineering Research Center, 2017: 6.
[7] 张涛, 杨利平, 宗兆奇, 等. 减少卷烟滤棒吸阻的波动[J]. 科技信息, 2010(29): 465-466. Zhang T, Yang L P, Zong Z Q, et al. Reducing the fluctuation of suction resistance of cigarette filter rods [J]. Science and Technology Information, 2010(29): 465-466. (in Chinese)
[8] 吴树清. 滤棒成型机丝束恒张力控制系统的设计[J]. 烟草科技, 2020, 53(9): 100-103. Wu S Q. Design of constant tension control system for tow of filter rod forming machine [J]. Tobacco Science & Technology, 2020, 53(9): 100-103. (in Chinese)
[9] 王丽丽. 大数据背景下数据挖掘技术的应用[J]. 计算机与网络, 2021, 47(20): 45-47. Wang L L. Application of data mining technology under the background of big data [J]. Computer and Network, 2021, 47(20): 45-47. (in Chinese)
[10] 凌军, 杨乾栩, 张天栋, 等. 基于大数据挖掘的卷烟内在质量自动检测模型构建[J]. 自动化与仪器仪表, 2018(11): 178-182. Ling J, Yang Q X, Zhang T D, et al. Construction of internal quality automatic detection model of cigarette based on large data mining [J]. Automation & Instrumentation, 2018(11): 178-182. (in Chinese)
[11] 李嘉霖. 基于数据挖掘的滤棒生产工艺参数优化研究[D]. 昆明: 云南财经大学, 2020.
[12] 王乐, 邓国栋, 吴钊, 等. 基于线性网络模型的卷烟逐口动态吸阻与动态通风率的关系[J]. 烟草科技, 2022, 55(5): 66-72. Wang L, Deng G D, Wu Z, et al. Relationship between puff-by-puff dynamic draw resistance and ventilation of cigarette based on linear network model [J]. Tobacco Science & Technology, 2022, 55(5): 66-72. (in Chinese)
[13] 卓鸣, 汪鹏, 望开奎. 基于MIV-BP神经网络的成品烟丝质量预测模型构建[J]. 食品与机械, 2021, 37(12): 161-166, 214. Zhuo M, Wang P, Wang K K. Prediction model building for finished tobacco quality based on MIV-BP neural network [J]. Food & Machinery, 2021, 37(12): 161-166, 214. (in Chinese)
[14] Kiranyaz S, Avci O, Abdeljaber O, et al. 1D convolutional neural networks and applications: a survey [J]. Mechanical Systems and Signal Processing, 2021, 151: 107398.
[15] 陈书辉, 章猛, 刘辉, 等. 一种1DCNN与多传感器信息融合的液压系统故障诊断方法[J]. 机械科学与技术, 2023, 42(5): 715-723. Chen S H, Zhang M, Liu H, et al. A fault diagnosis method of hydraulic system based on 1DCNN and multi-sensor information fusion [J]. Mechanical Science and Technology, 2023, 42(5): 715-723. (in Chinese)
[16] 刘伟, 单雪垠, 李双喜, 等. 基于并行1DCNN的滚动轴承故障诊断研究[J]. 机电工程, 2021, 38(12): 1572-1578. Liu W, Shan X Y, Li S X, et al. Fault diagnosis of rolling bearing based on parallel 1DCNN [J]. Journal of Mechanical & Electrical Engineering, 2021, 38(12): 1572-1578. (in Chinese)
[17] 张豪, 刘其洪, 李伟光, 等. 基于1DCNN特征提取和RF分类的滚动轴承故障诊断[J/OL] [2022- 09-05]. 中国测试, https://kns.cnki.net/kcms/detail//51.1714.TB.20230220.1703.004.html. Zhang H, Liu Q H, Li W G, et al. Rolling bearing fault diagnosis based on 1DCNN feature extraction and RF classification [J/OL] [2022-09-05]. Chinese Test, https://kns.cnki.net/kcms/detail//51.1714.TB.20230220.1703.004.html. (in Chinese)
