Signal Processing and Speech Communication Laboratory
hometheses & projects › Requirements Specification of a Systems-Engineering Tool: Example on Effort Estimation using Neural Networks

Requirements Specification of a Systems-Engineering Tool: Example on Effort Estimation using Neural Networks

Status
Finished
Type
  • Master/Diploma Thesis
Announcement date
26 Feb 2019
Student
Christoph Aigner
Mentors
Research Areas

** Abstract **

Product development in the modern automotive industry has to deal with a vast quantity of knowledge. Additionally, shorter cycle times for development come with the economic competition. An attempt to cope with developments, that are more and more complex, is systems engineering. The idea of systems engineering is to organize all entities which are involved with the product development, i.e. engineering teams of different departments and organizations which e.g. cope with project acquisition, requirements engineering, coordination of teams, etc. Two issues which emerge in this context are tackled within this thesis. Thereby, the battery development department of AVL List GmbH serves as the research environment. First, the volatility of gained knowledge from former product developments has to be handled. There may be many tools to document and store process information, but much information might be lost as not being retrievable and usable in a productivity-improving way. A design for an application that enables engineers to properly work with all kinds of information in the forerun of and during product development is being introduced to support model-based system engi- neering. This thesis outlines how existing applications may be combined and which benefits can be expected by doing this. Furthermore, functions that support the development processes are stated. Second, many different data items and large data quantities in various relations to each other make it hard for engineers to draw conclusions. For some of the introduced application functions, possible applications of machine learning are mentioned. Regarding a specific function - the estimation of expected development costs from product requirements - a proof of concept is developed by means of an offer assistance prototype application. The prototype application covers the preparation of sparse datasets as well as the usage of several neural network meth- ods to estimate the costs for product development. The results of the cost estimator revealed problems regarding the documentation of product requirements and offers. Nonetheless, the expected costs could be estimated such that engineers are supported at defining budgets for product developments.