博文
名正才能言顺
精选
||
引 子
2001年春，我收到John Wiley & Sons出版社寄来的Kusiak教授新著《Computational Intelligence in Design and Manufacture（设计和制造中的计算智能）》，书是赠送的，但要求我写一篇书评，书评将在一份设计与制造自动化方面的专业杂志上发表。因为书的题目所隐含的内容与我的研究领域很吻合，所以我当即回E-mail答应写一篇书评。
当我开始读这本书时，不解、不安、吃惊也随之而来。不解的是书中为何不见有关计算智能的内容，不安的是这篇书评如何写为好？吃惊的是作为国际制造与设计方面还算知名的专家，作者竟敢如此大胆写这种内容与其书名几乎不相关的专著，岂不知一篇坏著作的影响抵百篇好著作的影响？最终我还是客观地写了一篇书评，寄给了负责杂志书评的编辑。我知道该编辑（也是一位教授）同作者有着十分好的私人关系，所以随书评附了一封短信，大意是我必须根据实际情况写这篇书评，但如果他觉得刊登有困难，就不必发表了。直到2002年底，书评还是没有发表，询问也没有回音，2003年初，我写信给那位编辑，通知说我正式撤回书评，仍然是没有回音。
此事本已过去，可去年我偶然同自己的一位学生谈起国内有些书东拼西凑、粗制滥造，有些人什么话都敢说，什么书都敢出。没有想到他很不以为然地说，这没有什么，他大三时就给人“攒”了几本书，过程很简单：他从“二级”或“三级”承包商接书名和书号，再按章节甚至页数，分包给其他同学，这样不用一个月的时间，一本书就“攒”成了。至于通编、修改、校样等工作，根本就是没影的事，其实就连书的作者最后是谁也不知道。我像听“天方夜谭”般的听完之后，立即想起曾写的这篇书评，决定把它投到《自动化学报》，衷心希望在与自己最相关的研究领域里不会再有这种现象出现。
《科学时报》读书周刊希望转载这篇书评的中文版，对此十分感谢。我总觉得书是有些“神圣”的，所以希望作者与读者都能认真对待。
书 评
“当某人讲真话时，是什么使他的话为真？” 这是奎因在其经典名著《逻辑的哲学》（哈佛出版社，第二修正版，1986年）的开篇问题。我一直觉得真话不过就是话隐含的意思与话所指的事实相符而已，这是公认的常识，没有什么可说的。作为一个工程师，我常想奎因关于这个问题的长篇大论是哲学家小题大做的一个绝好的例子，直到我读了Kusiak教授的专著《设计和制造中的计算智能》（Computational Intelligence in Design and Manufacturing, John Wiley & Sons出版社，纽约，2000年）。
作为设计和制造方面的国际著名和多产专家，我毫不怀疑Kusiak教授在书中说的都是真话和事实，但我也必须说真话：其书名与其书的内容不符。我相信每个读者都会从书名认为本书主要涉及计算智能在设计与制造中的应用，但读过此书后，我也相信多数读者会觉得此书内容反映出的事实与其书名所隐含的意思相去甚远。对我来说，本书更恰当的名称应当是《设计和制造中的计算方法》，即Computational Methods，而不是Computational Intelligence。
更使我非常不解的是，作者并没有试图在本书中定义什么是计算智能，甚至就连计算智能在这本书中的具体内容是什么也没有说明。在第一章的30页中，只有一页涉及计算智能（见1.6节），但却没有说清计算智能到底在本书中是指什么。短短的一页，只告诉你计算智能有多么美妙，举了“机器人及视觉”和“模拟”作为计算智能的两个例子，提了“基于知识的系统”，“计划、检验和诊断系统”以及“试图解释复杂、不完备或冲突数据的多义辨析”作为在制造方面有重大影响的计算智能的几个产品。从事智能系统的研究差不多也有快20年的历史，我当然同意这些都是计算智能可以发挥作用的领域，但我也确信它们并不是当今计算智能的焦点方向。事实上，它们属于传统的人工智能、专家系统、机器人与自动化，或更加具体地说，属于《智能制造系统》：一个十分明确并且作者也应非常熟悉的领域，因为十多年前他曾以此为书名由同家出版公司发行过一本专著（John Wiley & Sons, 1990）。
我对计算智能的理解是：采用各类软计算方法，主要是神经元网络、模糊逻辑和遗传进化算法等去模拟人类决策智能的一个新兴领域。当然，人们可以认为当前还没有一个精确的、广泛接受的什么是计算智能的定义，人人都有权利提出自己的看法。但一般人，特别是从事本领域的专家学者，应尊重由当今教育者和研究者就其核心内容所达成的共识。 我让自己的学生去我们图书馆查阅了所有书名含有“计算智能”的书，共找到33本（但按主题字“计算智能”可以找到80多本），包括本书和《制造中的计算智能手册》（Computational Intelligence in Manufacturing Handbook, Wang和Kusiak编辑，CRC Press, 2001）。除了本书之外的所有的书，差不多都与我对计算智能的理解一致。举例来说，在其专著《计算智能引论》（Computational Intelligence: An Introduce, CRC Press, 1998），Pedrycz教授指出“计算智能”的概念是由Bezdek首次在其论文“关于神经元网络、模式识别和智能之间的关系” （On the Relationship between Neural Networks, Pattern Recognition and Intelligence, International Journal of Approximate Reasoning, 6, pp.85-107, 1992）中引入，同时声称：“计算智能的主要组成部分包含神经元网络、模糊集合技术和进化计算”。Kusiak共同编辑的《制造中的计算智能手册》一书也进一步地支持了这一断言：此手册共有19章，其中17章的章名中含有神经元网络、模糊集合或者遗传算法，剩下的两章是关于神经元网络应用方面的论述。尤其是，手册的编辑说明出版手册的原因是：“尽管有大量专门的神经元网络、模糊逻辑和遗传编程的出版物，但其中很少讨论设计与制造中计算智能。” 不知编辑是否注意到这里最后的几个字恰好就是本书的题目。
回到本书的讨论。在其全部的17章中，只有一章专门用于论述神经元网络（第11章，当然人们也可争论说最后关于数据挖掘的一章也可作为计算智能），全书没有真正讨论模糊逻辑，没有提及遗传算法或进化计算，更使我失望的是，这本书其实根本就没有涉及有意义的计算智能应用。我讲此话的事实依据是：在全书的535页之中，仅有5页，差不多就是百分之一，是关于能同计算智能挂上钩的神经元网络在部件识别中的应用。这本书主要是关于设计、运作和过程中运筹学的数学公式和表述。只有某些章节在其最后部分涉及计算实例，而且往往是在没有开始之前就停止了 —— 这是从有关材料的表达方式中所感受到的。其实，在前言的一开始，作者就声称：“本书的目的是表述在企业中应用计算方法的最近进展。” 这也是我感到本书应正名为《设计与制造中的计算方法》的原因之一。就其范围和表述而言，我认为在新的名称下，它将是一本扎实的著作。
本书共含17章535页。第一章引入现代制造的基本功能领域和相关技术，而第二章讨论人工智能和基于知识系统的一些基本概念。第三章至第十章包含设计和制造中的主要课题，从特征、产品建模、过程规划、装置减少、设备选择、生产规划和调度到Kanban系统和群组技术。这些材料十分充实但不涉及计算智能。第十一章给出了神经元网络学习算法的一些基本概念，尤其是BP和ART网络，并提供了它们在部件识别中的两个应用例子（共5页，本章共35页），很难说这是对神经元网络的一个很好的介绍。尽管有许多缺点，Lippmann十多年前的18页“科普” 文章“神经元网络计算介绍”（An Introduction to Computing with Neural Nets, IEEE Transactions on Acoustics, Speech and Signal Processing, Vol. 14, No. 12, 1987，本书本章的参考文献之一）在范围、表述、组织和深度方面都要好得多。第十二到十四章涉及设施和仓库布置和库存空间分配。这里引用了计算机化的相对设施分配（CRAFT）之类的启发式算法，但没有当代计算智能的痕迹。第十五章和十六章讨论灵巧设计和供应商中的“软课题”，从设计规划、产品区别到供应商客户关系和建立供应商的全面模型，但同样也没有找到计算智能的应用，尽管在理论上这些都是最该使用计算智能的地方。最后，第十七章以数据挖掘中的一些基本概念、方法、算法和步骤，以及数据农场（Data Farming）的简短讨论结束本书。
总体而言，这是适合于研究人员和研究生的一本关于设计和制造中的计算方法以及可能的计算智能方法应用的较好资源书，但将本书说成是设计和制造中的计算智能，可能的确是过分了，我不能在如此的标题下向读者推荐本书。
Fei-Yue Wang
The Key Lab of Complex Systems and Intelligence Science, Institute of Automation
Chinese Academy of Sciences, Beijing, China, 100080
(Email: feiyue@mail.ia.ac.cn)
Abstract This is an invited brief review on Andrew Kusiak’s new book Computational Intelligence in Design and Manufacturing, John Wiley & Sons, Inc., New York, NY, 2000, which reflects my opinion on what constitutes the main components of computational intelligence and my belief that the title of a book must be true statement of its contents.
Keywords Computational intelligence, computational methods, design, manufacturing.
“When someone speaks truly, what makes his statement true?” This was the opening question in W. V. Quine’s classical book Philosophy of Logic (Harvard University Press, 2nd edition, 1986). To me, it has always been a common sense that a true statement simply implies the meaning of the statement and the fact of the matter are identical. As an engineer, I often thought Quine’s lengthy discussion to his question is a perfect example of the philosophical extravagance, until I read Kusiak’s book: Computational Intelligence in Design and Manufacturing ( John Wiley & Sons, Inc., New York, NY, 2000).
I have no doubt that Professor Kusiak speaks truly about his book, but I must say that his title is not a true statement about his book. I believe every reader will think the title means the book deals largely with the application of computational intelligence in design and manufacturing. But after read the book, I am confident that most readers would come to the conclusion that the fact of the matter is quite different from the meaning of the book title. To me, a more appropriate title for this book should be Computational Methods in Design and Manufacturing, not Computational Intelligence (CI) there.
Actually, I am quite surprised that the author did not make effort to define what CI is in his book, not even specifically with respect to the context of his materials. Among 30 pages of the first chapter, only 1 page was spent on CI (Section 1.6), but no attempt was made there to clarify what CI is in the book. It just tells you what wonderful things CI can do, takes “robotics and vision systems” and “simulation” as two examples of CI applications, and mentions “knowledge-based systems”, “planning, testing, and diagnostic systems”, and “ambiguity resolves attempting to interpret complex, incomplete, or conflicting data” as the CI products that have a significant impact on manufacturing. I have been working with intelligent systems for almost 20 years now, I do agree those are the areas where CI can find applications, but I am sure they are not the focus of the contemporary CI. As a matter of fact, they belong to the applications of traditional artificial intelligence, expert systems, robotics and automation, or more specifically, Intelligent Manufacturing Systems: a well-defined topic the author should be quite familiar with since he had published a book with that title more than a decade ago (John Wiley & Sons, 1990).
My understanding of CI is an area that uses soft computing methods, mainly neural networks, fuzzy sets, evolutionary algorithms, etc., to mimic human intelligence for decision-making. Of course, one may argue that so far there is no generally accepted, precise definition of what CI is, but one should respect the consensus on its core materials reached by contemporary educators and researchers. I have asked my students to get all books in our library with a title starting with “computational intelligence”. There are 33 of them (more than 80 were found with keyword computational intelligence), including this book and “Computational Intelligence in Manufacturing Handbook” edited by Jun Wang and Kusiak (CRC Press, 2001). Not to my surprise, all the books, except the current one, agree largely with my CI understanding. For example, in his book “Computational Intelligence: An Introduction” (CRC Press, 1998), Pedrycz indicated that the concept of “computational intelligence” was introduced by Bezdek in his paper “On the relationship between neural networks, pattern recognition and intelligence” (International Journal of Approximate Reasoning, 6, pp.85-107, 1992), and stated clearly that “the main components of CI encompass neural networks, fuzzy set technology and evolutionary computation.” This statement is supported further by the fact that 17 out of 19 chapters in “Computational Intelligence in Manufacturing Handbook” have either neural networks, fuzzy sets, or genetic algorithms in their titles and the remaining two are about applications of neural networks. Actually, its editors stated that the reason for the handbook was: “despite the large volume of publications devoted to neural networks, fuzzy logic, and evolutionary programming, few address the applications of computational intelligence in design and manufacturing.” Note that the last few words constitute precisely the title of the current book.
Back to the current book in discussion. Among its 17 chapters, only one is devoted to neural networks (Chapter 11, but one may argue to consider the last chapter on data mining as CI related), the whole book has no real discussion of fuzzy logic, no mention of genetic algorithms or evolutionary programming, and to my disappointment, no meaningful applications of CI methods in design and manufacturing at all (only 5 pages, out of total 535 pages, are on applications of neural networks in part recognition). The book is mainly about the mathematical formulations of design, operations, and processes in the perspective of operations research. Few computational experiences are included at the end of some chapters, and they end before they start --- a feeling from the way the material is presented. Actually, in the very beginning of the preface, the author stated, “the goal of this book is to present recent advances in modeling and applying computational methods to enterprises.” This is why I feel the book should be appropriately titled: Computational Methods in Design and Manufacturing, and in terms of the scope and formulation, I consider the book as a quite solid one under the new title.
The book contains 17 chapters and 535 pages. The first chapter introduces the basic functional areas and related technologies in modern manufacturing, while the second chapter addresses the some basic concepts in AI and knowledge-based systems. Chapters 3 to 10 cover major issues in design and manufacturing, from features, product modeling, process planning, setup reduction, equipment selection, production planning and scheduling, to Kanban systems and group technology. The materials are substantial in their own perspectives but no CI methods are involved. Chapter 11 presents some basic concepts and learning algorithms in neural networks, specifically back-propagation and ART networks, with two examples on their applications in part recognition (totally 5 pages, the whole chapter has 35 pages). It is hard to say this chapter is a good introduction to neural networks. Lippmann’s 18-page paper “An introduction to computing with neural nets,” (IEEE Transactions on Acoustics, Speech and Signal Processing, Vol.14, No.2, 1987, cited as the references of this chapter in the book) is a much better introduction in terms of the scope, presentation, organization, and depth. Chapters 12-14 deal with facility and warehouse layout and inventory space allocation. Heuristic algorithms such as computerized relative allocation of facilities technology (CRAFT) are used here but no sign of contemporary CI methods. Chapters 15 and 16 discuss “soft issues” in design for agility and supplier evaluation, from design rules, product differentiation, to supplier-customer relationship and building a comprehensive model for suppliers. Again no application of CI methods is found. Finally, Chapter 17 concludes the book with some basic concepts, methods, algorithms and procedures in data mining and a brief discussion of data farming.
Overall, this is a reasonable source book for researchers and graduate students for computational methods, as well as for the possible CI applications in design and manufacturing, but saying the book is about computational intelligence in design and manufacturing is farfetched, and I simply can not make that recommendation to readers.
此文发表于《自动化学报》，Vol. 29, No.6, pp.1027-1029, 2003
https://blog.sciencenet.cn/blog-2374-546723.html
上一篇：《科学新领域的探索》读后感
下一篇：从ASOS到嵌入式系统——杂谈电子与信息的机械化进程
