Masterarbeit MSTR-2020-36

Bibliograph.
Daten
Sadhasivan, Hema: Knowledge Representation and Automated Reasoning for Decision Support on Fault Correction in Manufacturing.
Universität Stuttgart, Fakultät Informatik, Elektrotechnik und Informationstechnik, Masterarbeit Nr. 36 (2020).
139 Seiten, englisch.
Kurzfassung

Data acquisition is immeasurably increasing with the accelerated growth of communication and information technology in the manufacturing domain. Semi and unstructured data formats constitute a significant part of the acquired data. This work concentrates on combining heterogeneous data sources, namely Failure Mode and Effects Analysis (FMEA) document, digital maintenance logbook, and the Eight Disciplines (8D) report for decision support on fault action in manufacturing. The FMEA consists of information regarding the failure details speculated by the Subject Matter Experts (SME). This is usually performed during the early stages like design or planning phase of a product development life cycle. The digital logbook consists of failures observed by the employees of the organization, as it occurs during the manufacturing phase. Finally, product-related complaints raised by customers are captured in the 8D report to prevent the recurrence of the errors. All three sources consists of heuristic data that are predominantly in natural language form. Therefore, it is necessary to alleviate the structural and semantic heterogeneity prevailing in these sources to facilitate an effective knowledge-based decision support system for fault diagnosis and fault correction in manufacturing. The reconciliation of these heterogeneities is attained by delineating the semantic relationship between the data fields of the sources. An ontology model based on Unified Modeling Language (UML) is used to describe the semantic meaning between the data fields in the unstructured data sources. This ontology model is then instantiated into a graph-based knowledge model for failure instances using the Bayesian Network (BN), that are proficient in handling uncertainties in data. Before the instantiation of the knowledge model based on BNs, the quality of these text sources must be improved to perceive the underlying failure details as intended. The main challenge in knowledge modeling with BNs is to identify the network structure schema and determine the prior and conditional probability distributions for the nodes in the identified structure. The former is achieved through ontology modeling. The latter part was more demanding as one of the multiple sources is purely qualitative and has to be quantified for this purpose. The findings of this approach provide substantial results that are verified qualitatively. Thus, this thesis is a proof of concept and shows that BN may be used to model and represent knowledge contained in unstructured text sources and to utilize it for automated reasoning under uncertainties to support decision-making processes for fault correction in manufacturing.

Abteilung(en)Universität Stuttgart, Institut für Parallele und Verteilte Systeme, Anwendersoftware
BetreuerMitschang, Prof. Bernhard; Reimann. Dr. Peter
Eingabedatum17. Dezember 2020
   Publ. Informatik