دانلود کتاب یادگیری ماشینی مبتنی بر دامنه برای تولید هوشمند

Over the past two decades, technology breakthroughs in manufacturing and computing have paved the way for the Fourth Industrial Revolution. Personalized products such as 3D-printed medical implants are more accessible to the public. The fast beats of Artificial Intelligence (AI), Machine Learning, and the digital transformation of manufacturing can be felt in our daily life. The transition from the century-old mass production paradigm to the personalized manufacturing paradigm has increasingly become a reality for value differentiation. It might be one of the best times for smart manufacturing research. Product quality control remains essential for both the economy of scale in mass production and the economy of scope in personalized manufacturing. Statistics has provided a scientific basis to infer the quality of the whole product population in mass production where the cost of learning from product samples diminishes with the scale. In personalized manufacturing, quality control faces the fundamental question of how to ensure quality of every product variety with one or a few units; more specifically, how to infer the quality of the whole product variety population based on a sample or a subset of product varieties? Does that mean that we have to break away from the existing statistical quality control framework? This book aims to present domain-informed machine/statistical learning as a viable strategy to address the quality control challenge in personalized manufacturing. By integrating domain knowledge, machine learning models can remain robust in the presence of imperfect or limited data, providing interpretable prediction and generalizable insights. However, machine learning for manufacturing is often motivated by real-world specific problems with many engineering details. Substantial training or experience is usually needed to define and formulate domain-meaningful learning problems for individual cases, collect the right data economically, identify contextualized information representation, devise interpretable models, generalize the insights, and deploy validated models for applications. The buildup of experi ence through learning isolated cases and connecting dots could take a long period of time. Using additive manufacturing and nanomanufacturing as exemplary domains, this book walks through the development of methodology from simple cases to general forms for contextualized information representation, domain-informed dimension reduction, fabrication-aware machine learning and transfer learning, and process-informed optimal compensation. Through case studies, the book illustrates how manufacturing domain knowledge in non-analytical forms is integrated into various aspects of machine learning for smart manufacturing quality. The book is aimed at manufacturing and quality engineering researchers and engineers, AI and machine learning professionals with interest in smart manufacturing and digital twins, and graduate students specialized in advanced manufacturing, quality engineering, or machine learning for engineering applications. It is suited for a second-year graduate-level course on Machine Learning for Smart Manufacturing, requiring prerequisites in multivariate statistics or graduate-level machine learning knowledge. The knowledge requirements for the manufacturing domain are minimal. Additional learning and teaching materials, including data, codes, slides, and other resources, can be found at http://huanglab.usc.edu/book. The author sincerely thanks his mentor, Prof. Jianjun Shi, for his wise suggestions, unfailing support, and continuous encouragement. The author is also thankful to all his coauthors Hui Wang, Xi Zhang, Lijuan Xu, Li Wang, Yuanqin Duanmu, Yuanxiang Wang, Cesar Ruiz, Nathan Decker, Weizhi Lin, Shekhar Bhansali, Tirthankar Dasgupta, Yong Chen, Chongwu Zhou, Arman Sabbaghi, Malancha Gupta, Noah Malmstadt, Raquel de S.B. Ferreira, and Nabil Anwer, whose work contributed to the book content. Lastly, the author extends deep gratitude to the US National Science Foundation for its decade-long support of research in nanomanufacturing and additive manufacturing, as well as to the US Office of Naval Research for its support of cyber-enabled additive manufacturing research from 2011 to 2014. Lastly, there are numerous open research challenges in domain-informed machine learning for smart manufacturing. This book represents author’s initial exploration of the development of quality control theories and methods for personalized manufacturing.