Bewertung und Optimierung der Vollständigkeit von Betriebsdaten im Kontext der Fehlerprädiktion

• Assessing and optimizing completeness of operational data in the context of defect prediction

Schlegel, Peter; Schmitt, Robert H. (Thesis advisor); Kampker, Achim (Thesis advisor)

1. Auflage. - Aachen : Apprimus Verlag (2022)
Book, Dissertation / PhD Thesis

In: Ergebnisse aus der Produktionstechnik 36/2022
Page(s)/Article-Nr.: 1 Online-Ressource : Illustrationen, Diagramme

Dissertation, RWTH Aachen University, 2022

Abstract

The performance of data-based defect prediction models strongly depends on the quality of the underlying data and thus on its suitability for the individual application. As one of the key dimensions of data quality, the context-specific completeness of operational data in particular is often insufficient in practice. Gaps in the database thus massively inhibit the creation of value from data.The objective of this work is therefore to create transparency by assessing the completeness of information captured in operational data in the context of defect prediction. Furthermore, the objective includes the utilization of this transparency for the optimization of completeness in order to improve suitability of data for defect prediction. Existing approaches are based on very generic assessment schemes without the analysis context of defect prediction and show a clear deficit regarding the systematic optimization of completeness. Given this background, the development of an assessment methodology consisting of four essential and sequential modules is carried out within the scope of this work.In the first module, the process and information modeling, relevant processes of the production chain are modelled and the information captured by systems is documented and classified. By visualizing this information, an overview of the actual state of the information captured along the production chain is created. The second module addresses the identification and weighting of potential defect influencing variables in order to determine the target state of information capture in the context of defect prediction. The focus of this is the formulation of causal relationships in the form of quantitative theses based on explicit and implicit knowledge. The following comparison of the actual and target state takes place in the third module in the form of a quantitative completeness assessment. The focus of this is the development of a context-specific assessment metric. Through a qualitative visualization of the resulting key indicators in the process model, transparency is created with regard to gaps in the capture of information. In the fourth and last module the derivation of concrete recommendations for the optimization of completeness in the sense of a model-based decision support is specified. Based on the developed metric, alternative measures are identified and prioritized in order to maximize the completeness of the information captured in the operational data.Lastly, the developed methodology is applied and validated on a case study from industrial practice and the fundamental solution hypothesis of this work is verified. By comparing the classification performance of predictive models with the corresponding completeness indicators, an improvement of the context-related data suitability can be shown by applying the methodology in the case study.