This work provides an overall framework and three tools that together substantially reduce the time and effort required to construct a model that is fit for purpose.
Computer Aided Process Engineering (CAPE) provides invaluable assistance in the design, analysis and improvement of Process Systems. Models are an integral part of the framework in which CAPE tools operate, however their development is an expensive task due to the specialist knowledge and technology· that is required. This restricts the development of models to those projects that are rich in time or money, or to those projects that will be foreseeably troubled without CAPE assistance.
The work presented in this thesis has examined those factors which contribute to the barriers in conceptual model development, and found three areas which are particularly problematic. These are (a) the reuse of existing model information, (b) the development of models and (c) the characterisation of models. This thesis develops tools to improve on these problems, and describes a framework in which these tools can work together for still greater productivity.
The first tool, called Model Families, builds on concepts originating in object oriented programming to provide an information-rich system for model storage. A Model Family is a hierarchical collection of models which provide many different representations of the same system. These are then collated in a manner which focusses on the similarities and differences between each model in the collection. This draws on object-oriented programming techniques to create the concept of a model member. Each member belongs to a family of alternatives together, this family comprises all the variations on a model that have been implemented. Activating a particular member sees it take dominance within its family of alternatives in this manner, the currently active model member is used in any modelling activities. Other members that are not used in a particular instance remain present in the conceptual model, but are effectively ignored by any model compilation or analysis tools. Modelling elements that appear in one location may be copied and/or modified for use elsewhere by extending existing elements. The user may either take a copy of the existing element as it already exists, or add more information to it.
The reuse encouraged in this system through inheritance and extension makes it possible to modify all instances of a concept within a model by changing it in its primary location. This speeds up model refinement and ensures consistency within the model. Secondly, the use of families and members makes it possible to store a large number of models of the same system in the one computer document, centralising system-describing information and encouraging information reuse.
The second tool is a number of formalised model manipulation procedures collectively called Model Transformations. This work involved research on a number of case studies, revealing that there are six typical methods for changing the information in a model. These methods have been refined and developed into semi-automated algorithms which have augmented the SCHEMA modelling framework. Addition introduces entirely new concepts into a model, and defines the relationships between the new information and the existing information. Its opposite transformation, Neglection, removes concepts from a model and redefines the remaining concepts so that the model is still defined and conceptually complete without the neglected information. The Merging transformation simplifies the conceptual record of an object, reducing its effect while not removing it completely. This effectively promotes all the other concepts in a model. Demerging is the reverse of this, allowing the user to replace a simple concept in the model with a more complicated one. It was found that the automation of these four model transformation algorithms has a theoretical upper limit, caused by the need for system-describing information which in this system is only available from the users conception of the physical system. The final two algorithms involve conceptual arrangement of information within a model rather than changing the meaning of the model. Aggregation conceptually groups a number of model parts together, and is especially useful for producing multi-scale models. Disaggregation is the reverse operation, removing a set of objects from such a group.
A number of benefits have resulted from the development and implementation of these transformations. The automation of these six commonly used modelling procedures has been shown to speed up the modelling development process by up to 75%. The scope for modelling errors decreases because the implications of each model modification are considered by the algorithms. The implications that are problematic are then either automatically resolved, or flagged to the user for attention. Finally, these transformations used the concepts of model families to automatically produce a document of a models development with little, or no additional effort from the user.
The third tool, Model Metrics introduces tools which are used to analyse models individually and with respect to each other. The existence of models in a number of forms is explained, and the metrics that are available within each form is discussed. Fourteen basic metrics that quantify and describe the basic form of a model are developed and described in this work. From these, a further five metrics have been developed which extract more complex and meaningful information from a model. These nineteen conceptual metrics provide the user with tools to guide them in model construction and refinement by illustrating the connection between the conceptual information in their model and the performance of their model. Nine empirical metrics are also collated and presented to complete a substantial collection of modelling metrics useful for high level modelling work.
The development of a set of modelling metrics makes it possible to quantify the concept of modelling goals. When a modeller is building a model for some purpose, there are a number of qualities that they seek to impart to their model, such as the amount of detail that certain areas of the model will contain, and the speed with which the model will be solved. This set of desirable characteristics form part of the modelling goal. The use of modelling metrics allow the user to measure these attributes of their model and compare them against some criterion derived from an "ideal” model developed for the same purpose. The same concept can also be used to assist in the selection of one model for a particular goal when choosing from a pool of models, by determining which model(s) best fit the modelling goal.
The three tools developed in this work are demonstrated to substantially reduce the workload on the modeller, and through this encourage the development of models. These tools are not restricted to Process Engineering, but are applicable to all areas which involve the conceptual modelling of a system.