【作者】 陈兆勋；

【导师】 曾明哲；

【作者基本信息】 上海交通大学 ， 机械工程， 2009， 博士

【摘要】 产品配置是一种特殊的产品设计行为,该行为基于给定构件集与构件间的约束关系,根据客户需求,从构件集中选择出一组构件形成满足客户需求的产品。产品配置问题源于20世纪80年代初的人工智能研究,经过20多年的理论研究,产品配置在表达形式与求解方法上形成了扎实的理论基础。在90年代初,产品配置被学者们引入大规模定制研究中,成为大规模定制个性化订单实现的核心技术。虽然产品配置具有成熟的理论研究基础,但企业的产品配置实践却一直困难重重。进入21世纪后,为了缓解产品配置实施过程中遇到的困难,学者们深入企业实践对产品配置管理问题进行了提炼与分析,并逐渐将产品配置研究的重心由传统的配置表达与求解研究转向针对具体管理问题的配置管理研究。依托于前人的研究基础,本文基于产品配置系统视角,针对性地选取了产品配置实施过程中遇到的若干显著的配置管理问题作为研究目标,进行了如下的研究工作:在产品配置实践中,企业虽然提供了定制化服务,却一直忽视了配置活动的个性化特征,使客户在配置过程中承担着显著的感知困扰风险。为了缓解配置活动个性化的缺失,本文提出企业需要通过个性化的配置规则为客户提供经过提炼的配置选项空间。但个性化配置规则的实施对配置规则的获取方法提出了很高的要求。为了实现个性化配置规则,本文提出了基于LCNN网络与RULEX算法的双型个性化配置规则构造方法。该方法通过对历史数据的有效训练与解释,实现了个性化配置规则的智能获取、表达与演绎。在传统产品配置研究中,学者们往往忽略了动态配置环境对产品配置活动的影响。在大规模定制环境下,产品配置环境(客户群体需求与产品实现技术)存在较为显著的动态性,因此如何让支持配置活动的配置知识保持良好的环境适应性困扰着企业的产品配置实施。为了缓解此问题,本文提出了基于LCNN网络与RULEX算法的双型配置规则联合实施与更新方法。该方法通过将配置规则的获取、表达、演绎与进化整合入由LCNN网络与RULEX算法联合形成的智能方法中,减少了规则形态转化过程耗用的企业资源与时间,提高了配置规则的实施与更新效率。在当前产品配置研究中,学者们对配置活动的认识逐渐由传统的单向配置转向更接近实际的双向的、交互式的配置活动。在双向交互配置研究中,如何保证由客户需求映射而来的配置方案能够满足所有配置约束(又称为一致性矫正)是交互配置研究的核心问题。但由于客户需求的不确定性与柔性,现有基于CSP求解方法的产品配置一致性矫正方法存在计算复杂度问题,并且忽视了客户偏好对一致性矫正的影响。为了缓解上述问题,本文提出了基于Content-addresable Memory(CAM)的产品配置一致性矫正技术。通过将基于CSP描述的一致性矫正问题转化为基于CAM机制的记忆唤醒问题,本文用Hopfield神经网络以合理的计算复杂度模拟了一致性矫正过程,同时通过将客户偏好引入CAM模型,实现了客户偏好对一致性矫正过程的引导。在产品配置活动中,客户需要的不是一个孤立的产品配置方案,配置方案所体现的成本信息将有助于企业对产品满足客户需求的程度进行综合的评价,并为后续的配置优化问题的构建与求解提供准确的定量信息。在配置实践中,多数企业采用了以产量为基础的传统成本计算逻辑。但在大规模定制环境下,传统成本计算逻辑忽视对间接成本要素的分析,往往导致成本估算的失真。为了缓解此问题,本文将作业成本法引入产品配置研究中,通过建立面向作业成本的产品模型、作业模型与作业组织体系,实现了对多样化作业成本信息的有效管理,并形成了基于作业成本法的产品配置成本估算方法。本文通过将多种人工智能技术与作业成本法引入产品配置研究中,一定程度上缓解了产品配置实施中遇到的若干难题。本文的研究成果虽然不能涵盖所有产品配置难题,却为产品配置的研究与应用提供了若干具有创新性的方法与实施框架,有助于缩短产品配置理论研究与企业实践之间的差距。更多还原

【Abstract】 Product configuration is a special design process of selecting and arranging combinations of predefined parts to satisfy given specifications. Although product configuration has been recognized as a key enabler for individualized order realization in mass customization and related research has established a sound theoretical foundation, the practices of product configuration in industries still face great challenges. To leverage the difficulties in configuration implementation, in the 21th century, several research literatures concentrated on the survey and analysis of management problems of product configuration and the research focus of product configuration gradually moved from technical issues to management issues. Based on the foundation of existing research, this dissertation selects four critical management problems of product configuration as research issues. The main contents include:Although companies are able to deliver customized products, most of them neglect the personalization in configuration activities. The lack of personalization probably results in the perceived risk of custom confusion and hinders the effectiveness of product configuration. To leverage the problem, this research introduces customers’personal characteristics into the modeling of configuration rules. A methodology which combines Local Cluster Neural Network (LCNN) and RULEX algorithm is proposed to efficiently acquire and apply personalized configuration rules by data learning and network explaining.In mass customization, product configuration is probably carried out in dynamic environment. However, existing configuration research mostly neglects the dynamics of configuration implementation. Therefore, the crucial issue about how to efficiently implement and update configuration knowledge to fit for dynamic environment is still unaddressed. To solve this problem, a framework combining LCNN and RULEX is developed to coordinate the implementation and update of configuration rules. In the framework, rule acquisition, representation, application and evolution are incorporated into the same intelligent methodology, by which the transfers between different rule formulations are accelerated.In current research, product configuration is gradually considered as bidirectional design interactions. In interactive configuration, maintaining the configuration assignment satisfies all configuration constraints (termed as consistency restoration, CR) is one of the primary issues which should be addressed. However, because of the uncertainty and flexibility of customer needs, existing CSP-based CR approaches mostly face the challenge from computational complexity. Besides, most approaches neglect the impact of customer preferences on CR. To address the problems, this research proposes a responsive CR technique based on content-addressable memory (CAM). The CSP-based CR problem is compiled into a CAM recalling process. Then Hopfield network is adopted to automatically correct the inconsistent assignment. Meanwhile, to introduce customer preference into the CAM model, specific orientation mechanisms are developed.Product costing is a process of estimating the cost of a product at design stage. The cost information helps companies to comprehensively evaluate how the configuration satisfies customers. However, traditional costing methods suffer the drawbacks of poor accuracy, low agility and undesirable degree of detail. To leverage the drawbacks, this research introduces activity-based costing (ABC) into product configuration. By developing generic and ABC- based product model, activity model and activity organization architecture, companies are able to provide detailed process-oriented cost information to support product configuration.更多还原