Studies on Data Management in Manufacturing Line Monitoring and Control

In recent years, as general purpose computers like personal computers spread into control systems in manufacturing fields, the control systems are expected to be integrated with business systems such as quality management systems, cost management systems and ERP(Enterprise Resource Planning) systems. This integration makes it possible to deliver data rapidly between control systems and business systems, compared with traditional methods like manual inputs and batch based transferring, and is considered to optimize production activities by dissolving information delay and improving data accuracy. To achieve this integration, frameworks for cooperation and data exchange among applications like COM (Component Object Model) by Microsoft and XML (eXtensible Markup Language) are not enough. In control systems, sensor data are sequenced by the time stamp order for each facility or each equipment, and these data are meaningless as themselves in business systems. That is, a new contrivance is required to reform the sensor data in control systems to useful data in business systems. For the purpose of effective management of product quality and cost, it is required to provide useful information which is tightly connected with object flow, such as "where this object was passed and how was its status" and "what object is (was) existing in which processes and how was its status." To realize this requirement, the following characteristics of data in manufacturing systems should be handled. The first one is the temporal property. Although temporal database languages like TSQL and TQuel have the ability to deal with temporal aspects, an object itself has a temporal aspect in manufacturing systems. For example, the value field in a sensor object should have a valid value when accessed by an application. The second one is the change of identity and form of objects. Workflow models and some geographical data models handle the change of the object identity. These models, however, cannot handle the change of object fonn. A manufactured object may exist in one manufacturing process as a point or in some manufacturing processes as a line. The last one is multi-dimensionality of manufacturing result data. Since the concepts of views in object-oriented databases mainly focus on restructuring class hierarchies, they cannot represent the aspects of data in manufacturing systems focusing on one of the time, a manufacturing process and a manufactured object. In this thesis, the data management models for manufacturing systems are presented. First of all, a temporal object model based on temporal validity is proposed. The model defines the concept of temporal validity to assure the correctness of temporal objects. Secondly, a multidimensional data utilization model for manufacturing management is proposed. The model has the ability of reforming sensor values to multidimensional aspects of the entire plant, each of manufacturing processes and histories of manufactured products. Thirdly, the Time/Place/Object model to represent the organization and the behavior of a manufacturing line is presented. This model enables the integration of the monitoring function and the control function by the ECA mechanism. These models are incorporated with database middlewares and applied to real plants. The implementation issues are also introduced.