Metodología dirigida por modelos para las pruebas de un sistema distribuido multiagente de fabricación

Resumen

Market pressures have pushed manufacturing firms to reduce costs while improving their products by specialising in their main value-adding activities and collaborating with other specialist firms for the others. These distributed manufacturing systems bring new challenges, as it is difficult to integrate their disparate information systems and organise them in a coherent manner, prompting researchers to provide a number of abstractions, architectures and specifications that attempt to tackle this complexity. Among them, holonic manufacturing systems have achieved considerable attention: these view enterprises as networks of holons, entities which are at the same time formed by and part of several other holons. So far, holons have been implemented as intelligent self-aware agents for manufacturing control, but their steep learning curve and difficulties in integrating with legacy systems has hindered their industrial adoption. Additionally, their emergent behaviour may not be desirable in contexts with tasks that need specific guarantees in their behaviour, such as business-to-business or business-to-customer relations and high-level plant management operations. This Thesis proposes a more flexible view that allows a holon to have varying degrees of self-awareness and emergent behaviour, and defends that business holons may be better implemented as services or software components that can be reused through standards-based technologies from anywhere in the organisation. These services are usually organised into coherent catalogues, known as Service Oriented Architectures. A successful SOA initiative may provide large benefits to the organisation, but it is not a trivial task. For this reason, many methodologies have been proposed in the literature, but none of them cover the need to test the services that are finally developed. Considering that the goal of a SOA is to increase reuse of the available software throughout an organisation, this is an important oversight: high-quality services are crucial to a successful SOA. Therefore, the main goal of this Thesis is defining an extended methodology that assists users in testing the services that implement their business holons. After considering the available options, the model-driven SODM methodology was selected as a starting point and was then largely rewritten using the open source Epsilon framework, allowing users to model their partial knowledge about the expected performance of the services. This partial knowledge is used by several novel performance requirement inference algorithms that extract the specific performance requirements of each service. While the throughput inference algorithm is based on a simple traversal of the graph, the time limit inference algorithm went through several revisions before reaching the desired level of functionality and performance. After a first formulation based on linear programming, it was replaced with a simple ad hoc algorithm and then with a much faster and more advanced incremental algorithm. The incremental algorithm produces equivalent results to the others while taking much less time, even in large models. In order to extract more value from the models, this Thesis also provides a general approach for generating test artefacts for multiple technologies out of the models annotated by the previous algorithms. To evaluate the viability of this approach, it has been implemented for two different applications: repurposing unit tests written in the Java programming language as performance tests, and generating performance testing projects in the The Grinder framework for any Web Service described using the language-agnostic Web Services Description Language standard. Lastly, the full methodology is successfully applied on a case study based on a rectified tile manufacturing firm of a Spanish manufacturing enterprise group. This case study starts with a high-level description of the business and concludes with the partial implementation of one of the holons involved and the generation and execution of performance test artefacts for one of its Web Services. With its support for designing performance requirements and testing them, SODM+T can therefore help users obtain a higher degree of confidence in the implementation of the business holons observed in their enterprises.