Carolin Wagner

Qualitative Assessment of Machine Learning Techniques in the Context of Fault Diagnostics

Nowadays, in the light of high data availability and computational power, Machine Learning (ML) techniques are widely applied to the area of fault diagnostics in the context of Condition-based Maintenance (CBM). Those techniques are able to learn intelligently from data to build suitable classification models, which enable the labeling of unknown data based on observed patterns. Even though plenty of research papers deal with this topic, the question remains open, which technique should be chosen for a specific problem. In order to select appropriate methods for a given problem, the problem characteristics have to be assessed against the strengths and weaknesses of relevant ML techniques. This paper presents a qualitative assessment of well-known ML techniques based on criteria obtained from literature. It is completed by a case study to identify the most suitable techniques to perform fault diagnostics in in-vitro diagnostic instruments with regard to the presented qualitative assessment.

Ontology-driven IoT code generation for FIWARE

The Internet of Things concept is increasingly being used in projects from different areas. Often, projects encompass applications from different domains that need to share information for a common purpose. These types of systems can become complex because they involve different domains and it can be difficult to share information without misinterpreting certain information. Ontologies can be a way to describe information to be shared between different domains. The present work aims to propose a tool that facilitates the creation of Internet of Things systems using ontologies and through this model generate code automatically to pre-automate the integration of devices. A tool has been developed that facilitates the extraction of information from the ontology and it has been validated in the case study.

Component Analysis Based Approach to Support the Design of Meta-Heuristics for MLCLSP Providing Guidelines

Todays supply chain are highly complex and globally set-up underlying a constant change with increasing speed. This has to be reflected by the planning processes and algorithms being utilized in the different stages of a supply chain. In the context of production planning, meta-heuristics are usually applied due to their ability to handle high complex problems. As a consequence, these algorithms require adaptation to the new scenario or even new solution approaches/strategies have to be devised. However, designing a meta-heuristic of good performance for a problem is a hard task, since it requires deep knowledge on the problem, as well as on the meta-heuristic side. Therefore, the existence of supporting guidelines for meta-heuristics might ease and speed-up the adaptation or design of these algorithms to better cope with the problem. In this paper, meta-heuristics are deconstructed into its components and an approach for component-based analysis is proposed to gain knowledge about their performance and how they perform the search. Based on the results of this analysis, guidelines can be devised. The proposed approach is applied for analyzing components of a good performing Genetic Algorithm (GA) for multi-level capacitated lot sizing problem (MLCLSP) and initial guidelines for the construction of GA in the domain of MLCLSP are generated.