h1

TY  - THES
AU  - Müller, Tobias
TI  - Modellbildung mittels symbolischer Regression zur Messunsicherheitsbestimmung komplexer Messprozesse
VL  - 7/2023
PB  - RWTH Aachen University
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2023-01146
SN  - 978-3-98555-146-0
T2  - Ergebnisse aus der Produktionstechnik
SP  - 1 Online-Ressource : Illustrationen, Diagramme
PY  - 2023
N1  - Druckausgabe: 2023. - Auch veröffentlicht auf dem Publikationsserver der RWTH Aachen University. - Weitere Reihe: Fertigungsmesstechnik & Qualitätsmanagement. - Weitere Reihe: Edition Wissenschaft Apprimus
N1  - Dissertation, RWTH Aachen University, 2022
AB  - The knowledge and the determination of the measurement uncertainty of a measurement process represents an elementary component of production metrology. Only with a quantified measurement uncertainty it is possible to determine the suitability of measurement systems and to indicate the risk of wrong decisions (e.g. in the proof of conformity of products). Rough estimates of the measurement uncertainty on the safe side often lead to unnecessary restrictions of specification limits and thus to rising production costs. At the same time, the complexity of measurement systems and thus also the determination of the measurement uncertainty is increasing due to higher quality requirements and a growing number of product variants. In determining the measurement uncertainty, the modelling of complex measurement processes in particular poses challenges. Therefore, the aim is to investigate the creation of a valid model of the measurement for the determination of the measurement uncertainty of complex measurement processes. In the research work, the influencing variables of a measurement process are first investigated as a basis for the modelling. A relevance evaluation with feature selection algorithms ensures that the selected influencing variables have a significant contribution to the functional relationship. The integration of the relevance assessment is important, because the integration of irrelevant influence quantities leads to an additional effort in the model building as well as in the later quantification of the uncertainty contributions for the determination of the combined measurement uncertainty. For the modelling itself, the symbolic regression is considered in more detail. Objects of investigation are the integration of prior knowledge, the reduction of the bloating effect as well as the optimization of the hyperparameters of the symbolic regression. In order to evaluate the validity of the model, a further procedure is necessary. The challenge here is that experiments and the model building are very time-consuming for complex measurement processes. Accordingly, it can be advantageous to obtain detailed information about the validity of the model in parts of the investigated value range. In the research work, a regression-based procedure is developed, which compares the model results with real experimental results. By comparing confidence intervals, a statement can then be made as to whether the model is valid, partially valid (= valid for a certain sub-range of the value range) or not valid. The result of the research work is a procedure for white box modelling using symbolic regression, which is especially designed for the challenge of determining the measurement uncertainty of complex measurement processes. Using a relevance evaluation with a combination of two wrapper methods, irrelevant influencing variables can be filtered out and the effort of model building and measurement uncertainty determination can be reduced. Furthermore, the validity of the model can be determined specifically for the measurement range by comparing model and real data and comparing the confidence intervals.
LB  - PUB:(DE-HGF)11 ; PUB:(DE-HGF)3
DO  - DOI:10.18154/RWTH-2023-01146
UR  - https://publications.rwth-aachen.de/record/889588
ER  -