Evaluation of the Potential of Deep Learning for Manufacturing Process Analytics

Elias Hagendorfer

Evaluation of the Potential of Deep Learning for Manufacturing Process Analytics

Titel in Übersetzung: Bewertung des Potentials von Deep Learning zur Prozessanalytik in der Produktion

Elias Hagendorfer

Lehrstuhl für Automation und Messtechnik (530)

Publikation: Thesis / Studienabschlussarbeiten und Habilitationsschriften › Masterarbeit

482 Downloads (Pure)

Abstract

Die vorliegende Arbeit untersucht die Anwendung von Deep Learning zur automatischen Identifikation von Maschinenoperationen aus von Sensoren und Aktoren stammenden multivariaten Zeitreihen. Methoden des Deep Learning sowie der Zeitreihenanalyse werden analysiert um ihre Tauglichkeit festzustellen. Ein neuer Ansatz wird vorgestellt um die Schwächen in aktuellen Ansätzen zu beheben, welche unter anderem unzureichende Signalauswahl, den hohen Bedarf an Trainingsdaten sowie die Vernachlässigung der physikalischen Natur des Systems umfassen. Dieser besteht aus: einer Methodik zur Signalvorverarbeitung rund um Stationaritätstests, Redundanzanalyse und Messung der Entropie; einem Deep Learning Algorithmus, der Segmente von Zeitreihen in verschiedene Operationskategorien klassifiziert; einem Konzept zur Prozessanalytik rund um Länge und Häufigkeit der Operationen. Der Ansatz wurde erfolgreich auf mehrere Datensätze von Schwermaschinen zum Schüttgutumschlag angewandt.

Titel in Übersetzung	Bewertung des Potentials von Deep Learning zur Prozessanalytik in der Produktion
Originalsprache	Englisch
Qualifikation	Dipl.-Ing.
Gradverleihende Hochschule	Montanuniversität
Betreuer/-in / Berater/-in	O'Leary, Paul, Betreuer (intern)
Datum der Bewilligung	23 März 2018
Publikationsstatus	Veröffentlicht - 2018

Bibliographische Notiz

nicht gesperrt

Schlagwörter

Deep Learning
Zeitreihenanlyse
Signalauswahl
Produktionsprozess
Sensordaten

Zugriff auf Dokument

VolltextEndgültige, publizierte Fassung, 5,45 MB

Dieses zitieren

@mastersthesis{c69aacfbf2bd44cea516346781db22c1,

title = "Evaluation of the Potential of Deep Learning for Manufacturing Process Analytics",

abstract = "This thesis investigates the use of deep learning for the automatic identification of machine operations from multivariate time-series data emanating from sensors and actuators. Methods from deep learning and time-series analysis are reviewed with the aim of determining their suitability. A new approach is introduced to alleviate weaknesses in current approaches which include insufficient signal selection, requirement of large amount of training data or neglection of the physical nature of the system. It consists of: a preprocessing methodology based around stationarity tests, redundancy analysis and entropy measures; a deep learning algorithm classifying time series segments into operation categories; a process analytics framework dealing with operation length and frequency. The approach was applied successfully to several datasets from heavy machinery bulk handling systems.",

keywords = "deep learning, time-series analysis, signal selection, manufacturing process, sensor data, Deep Learning, Zeitreihenanlyse, Signalauswahl, Produktionsprozess, Sensordaten",

author = "Elias Hagendorfer",

note = "no embargo",

year = "2018",

language = "English",

school = "Montanuniversitaet Leoben (000)",

}

TY - THES

T1 - Evaluation of the Potential of Deep Learning for Manufacturing Process Analytics

AU - Hagendorfer, Elias

N1 - no embargo

PY - 2018

Y1 - 2018

N2 - This thesis investigates the use of deep learning for the automatic identification of machine operations from multivariate time-series data emanating from sensors and actuators. Methods from deep learning and time-series analysis are reviewed with the aim of determining their suitability. A new approach is introduced to alleviate weaknesses in current approaches which include insufficient signal selection, requirement of large amount of training data or neglection of the physical nature of the system. It consists of: a preprocessing methodology based around stationarity tests, redundancy analysis and entropy measures; a deep learning algorithm classifying time series segments into operation categories; a process analytics framework dealing with operation length and frequency. The approach was applied successfully to several datasets from heavy machinery bulk handling systems.

AB - This thesis investigates the use of deep learning for the automatic identification of machine operations from multivariate time-series data emanating from sensors and actuators. Methods from deep learning and time-series analysis are reviewed with the aim of determining their suitability. A new approach is introduced to alleviate weaknesses in current approaches which include insufficient signal selection, requirement of large amount of training data or neglection of the physical nature of the system. It consists of: a preprocessing methodology based around stationarity tests, redundancy analysis and entropy measures; a deep learning algorithm classifying time series segments into operation categories; a process analytics framework dealing with operation length and frequency. The approach was applied successfully to several datasets from heavy machinery bulk handling systems.

KW - deep learning

KW - time-series analysis

KW - signal selection

KW - manufacturing process

KW - sensor data

KW - Deep Learning

KW - Zeitreihenanlyse

KW - Signalauswahl

KW - Produktionsprozess

KW - Sensordaten

M3 - Master's Thesis

ER -