Research On Resource Modeling And Selection In Cloud Manufacturing

With the rapid development of computer network technology, the globalization and informatization of manufacturing industry increasingly highlights, which improves the manufacturing resource sharing and utilization. Some new manufacturing mode with computer network technology such as networked manufacturing and manufacturing grid has gradually become the main mode in modern manufacturing industry. Cloud Manufacturing is Cloud Computing’s landing and extending in manufacturing and shows the concept of ’distributed resources being integrated for one task’ and ’integrated resources being distributed for service.’It realizes the high degree of sharing and utilization of manufacturing resources by integrating and centralized managing the distributed resources with the many-to-many service mode to improve enterprise production efficiency and provide users with higher satisfaction and more environmentally friendly products and services.Coming down in one continuous line with networked manufacturing and manufacturing grid, Cloud Manufacturing is a large-scale networked distributed manufacturing with three characteristics:(1) Massive, heterogeneous, complexity and coarse granularity on Cloud Manufacturing Resource;(2) High degree of participation and diversity on Cloud Manufacturing Users;(3) Strong self-healing on Cloud Manufacturing Process. According to the three characteristics above, based on the fundings of National Natural Science Foundation of China (Grant No.60970021) and Science and Technology Planning Project of Zhejiang Province (Grant No.2007C21013), this thesis focused on the definition, publication, matching and selection of resource in Cloud Manufacturing, where the major contributions of this thesis were listed as followings:(1) According to the Cloud Resource characteristics of massive, complexity and heterogeneous, a Cloud Manufacturing architecture, a bilayer resource model, and a mechanism of resource package, publication and searching based on RVCS (Resource Via Cloud Service) were proposed. Based on the WSRF architecture, we proposed the Cloud Manufacturing architecture and described the logical structure of Cloud End system and Cloud Manufacturing Platform system in detail. According to the different functionality of the two systems, a Bilayer Cloud Manufacturing Resource Model (BCMRM) was proposed in order to establish an unified data model for complex and heterogeneous resources:build resource basic attributes-oriented basic data model in Cloud End, and resource service attributes-oriented functionality data model in Cloud Manufacturing Platform, both of which have logical mapping relationship with each other. According to functional characteristics and updating frequency, resource dynamic attributes were divided into three levels (resource attributes, service attributes and provider attributes) and two categories (characteristic attributes and state attributes). Then, by introducing the connector and Cloud Broker which played the role of real-time intercepting and attribute selecting, we put forward a RVCS-based mechanism of resource package, publication and searching, in order to distributed storage and rapidly independent updating of massive Cloud Resource data. Moreover, the simulation results showed the superior performance of the method under massive data environment.(2) According to Cloud Resource characteristic of coarse granularity, a calculation method for performance similarity of Cloud Resource based on Multidimensional Extension Theory was proposed. At first, we put forward that the core work of characteristic attributes matching is performance attributes matching of Cloud Service Resource. Then, we described the Cloud Resource performance attributes and users’requirements with Matter Element Model, and divided performance module into four categories by the sum of performance index every module:1-D performance module,2-D performance module,3-D performance module and n-D performance module, in order to transform the problem of Cloud Resource performance matching into calculation problem of extension distance between point and multidimensional cube in multidimensional space, then calculate the similarity of Cloud Resource performance module by establishing correlation function. At last, we got performance similarity of Cloud Resource with module weight which was calculated by user-defined weight, structure weight and example weight. The experiment results showed that the method had better hit ratio of the best resource, and reduced the calculation complexity for performance index weight of coarse-grained Cloud Resource.(3) According to Cloud Manufacturing Users characteristics of high participation and diversity, by introducing Beth trust relationship theory, a QoS evaluation method based on user independent predictive evaluation and other users recommended evaluation was proposed. At first, we put forward that the core work of state attributes matching is QoS evaluation of Cloud Service Resource, and established a QoS evaluation method based on predictive evaluation and recommended evaluation. The predictive evaluation model based on user historical evaluation was optimized by introducing experience factor and modifying factor. The recommenders were preliminarily selected by historical experience, and further selected by evaluation similarity and objectivity. Then, we calculated the weight of predictive evaluation and recommended evaluation by coefficient variation and got the QoS comprehensive evaluation. The simulation results showed that the predictive evaluation has better sensitivity, and the recommender selection algorithm can filter better malicious evaluation and poor evaluation.(4) Aimed at the alternative resources selection problem when a certain Cloud Manufacturing node failed, an extension evaluation model for manufacturing risk based on Reciprocal Judgement Matrix with Triangular Fuzzy Numbers, and Cloud Resource dynamic adjustment strategy were proposed. We selected alternative resource with the consideration from two aspects of manufacturing risk and dynamic attributes matching. For manufacturing risk evaluation, based on extension matter element model, we established an evaluation model for manufacturing risk of Cloud Resource, with the risk, index weight calculated by Reciprocal Judgement Matrix with Triangular Fuzzy Numbers, and gave portfolio risk algorithms for three production relationship; In dynamic attributes matching, by introducing coefficient variation and usage threshold, we solved the incomparable problem of performance similarity and QoS evaluation, as well as the QoS evaluation objectivity problem. The experiment results showed that the method can reflect better manufacturing risk and alternation of resource.