您当前位置: 首页  >  学术讲座

讲准字118号:基于卷积神经网络的故障诊断方法及其应用研究

发布时间:2018-05-07|浏览次数:

题目:基于卷积神经网络的故障诊断方法及其应用研究

主讲:郑西显

时间:2018年5月28日 10:00

地点:电气楼415

主办:电气学院


主讲简介:郑西显,中国台湾清华大学教授。研究专长:过程控制。Honors and Awards: (1) Award of Industrial Collaboration 2007, National Tsing-Hua University (2) Coordinator TSMC/NTHU Collaboration Project, 2007. (3) Teacher of the year Award, Ministry of Education, Taiwan, 1992. (4) Member of Trustee, Advanced Control System, 2004-2007. PUBLICATIONS:(1)Kai Sun, Chen-Ting Tseng, David Shan-Hill Wong, Shyan-Shu Shieh, Shi-Shang Jang*, Jia-Lin Kang, Wei-Dong Hsieh, “Model Predictive Control for Improving Waste Heat Recovery in Coke Dry Quenching Processes”, Energy, 80, (2015), 275-283. (2)S. -P. Lee, D. S. H. Wong, C. -I. Sunc, W. -H. Chenc and S.-S. Jang, “Integrated Statistical Process Control and Engineering Process Control for a Manufacturing Process with Multiple Tools and Multiple Products”, Journal of Industrial and Production Engineering,32, 3, (2015), 174-185. (3)Chun-Cheng Chang, Shyan-Shu Shieh, Shi-Shang Jang*, Chan-Wei Wu, Ying Tsou, “Energy Conservation Improvement and ON-OFF Switch Times Reduction for an Existing VFD-fan-based Cooling Tower”, Applied Energy, 154(2015), 491-499. (4)Albazzaz, Hamza; Kang, Jia-Lin; Chehadeh, Dduha; Bahzad, Dawood; Wong, David; Jang, Shi-Shang*,” Robust Predictions of Catalyst Deactivation of Atmospheric Residual Desulfurization”, Energy & Fuels, 29,7089-7100, (2015). (5)Jia-Lin Kang, David Shan-Hill Wong, Shi-Shang Jang*, Chung-Sung Tan, “A Comparison between Packed Beds and Rotating Packed Beds for CO2 Capture Using Monoethanolamine and Dilute Aqueous Ammonia Solutions”, International Journal of Green House Gas Control, 46, 228-239, 2016. (6)Kai Sun, Shao-hsuan Huang, David Shan-Hill Wong, Shi-Shang Jang,” Design and Application of a Variable Selection Method for Multi-layer Perceptron Neural Network with LASSO” IEEE Transactions on Neural Networks and Learning Systems, 28, 6, 1386 – 1396, (2017). (7)Jialin Liua David Shan-Hill Wong, Shi-Shang Jang, Yu-Ting Shen, “Energy-saving design for regeneration process in large-scale CO2 capture using aqueous ammonia”, Journal of the Taiwan Institute of Chemical Engineers, 73, 12-19, (2017)

 

主讲内容:故障诊断一直是一个非常活跃的研究领域,特别是在化工行业。数据驱动故障诊断方法在制程监控为目前理想解决方法。数据驱动的方法通常包括统计分析和人工智能。多变量统计分析,如主成分分析、独立成分分析和偏最小二乘法方法、当制程含有大量高度相关的变量时,它们被广泛用于时简化与降维度来进行诊断异常操作。制程故障可以藉由观察制程是否保持在“统计控制状态”来检测。当制程超出预期的操作范围时,则表示有异常变化发生。然而,总有一些失误不容易被发现,因为监控模型是以正常数据的数据分布为基础进行建模,其数据中包含的噪声可能与异常讯号类似而无法发现异常。人工智能技术已经成功被应用于故障诊断在不确定性和高度复杂性的系统中。神经网络(NN)已经在化学工程应用中都得到验证。故障诊断可以通过模式识别故障。然而监督式监控缺点在于无法正确分辨不曾看过的分类项目,而新故障类型会被强制分给类似分类进而降低分辨正确率。故此,本研究提出可自我扩充知识的监督式监控方法来改善无法分类新分类的问题,并采用卷积类神经网络来实现可自我扩充知识的算法,并使用经典的田纳西伊士曼制程(TEP)评估算法的性能。使用移动窗口的方式来实现在线监测与早期预警的能力。自我扩充知识的算法配合神经网络取得故障特征并藉由T2管制图辨别新错误,再将新分类与旧有分类以转换学习方法重新训练分类层。藉由逐步增加故障分类学习方式习得所有新分类,以达成化工制程上可自我扩充知识学习新故障之目的。


欢迎广大师生参加!