TY - BOOK

T1 - Data-Driven Fault Detection and Identification in Rubber Injection Molding

AU - Hutterer, Thomas

N1 - embargoed until 20-11-2023

PY - 2021

Y1 - 2021

N2 - In injection molding, the properties of rubber compounds, and the manufacturing conditions themselves are subject to fluctuations, which impact the quality of the finished rubber parts negatively. To detect such fluctuations and to keep part quality constant, process monitoring systems are used in a wide range of industrial applications. For this purpose, a data-driven statistical monitoring system for rubber injection molding needs to be developed. By employing multivariate statistics, faulty parts can be detected and, once a fault database is present, fault types can be identified. To develop the system, first a dynamic compression testing methodology is presented, which is able to determine the relevant dynamic quality parameters of rubber parts in a way that is fast enough for on-line implementation and 100 % quality control. Second, the temperature of the rubber, the most important factor influencing the final part quality, is determined at every stage of the rubber injection molding process. By employing ultrasound and thermography, the rubber temperature resulting from the processing conditions can be modeled. On this base, the data-driven process monitoring system is built. This process monitoring system detects fluctuations causing faults by using a Principal Component Analysis (PCA) based approach. As a result, all available process signals can be evaluated simultaneously, even linear dependent ones. Additionally, by Fisher Discriminant Analysis (FDA), the type of fault can be automatically identified. Compared to other methods of process monitoring available for rubber injection molding, the presented data-driven system eliminates the need for any preliminary material tests, modelling or data selection. It is therefore much more straightforward and cost-efficient in implementing in smart manufacturing facilities.

AB - In injection molding, the properties of rubber compounds, and the manufacturing conditions themselves are subject to fluctuations, which impact the quality of the finished rubber parts negatively. To detect such fluctuations and to keep part quality constant, process monitoring systems are used in a wide range of industrial applications. For this purpose, a data-driven statistical monitoring system for rubber injection molding needs to be developed. By employing multivariate statistics, faulty parts can be detected and, once a fault database is present, fault types can be identified. To develop the system, first a dynamic compression testing methodology is presented, which is able to determine the relevant dynamic quality parameters of rubber parts in a way that is fast enough for on-line implementation and 100 % quality control. Second, the temperature of the rubber, the most important factor influencing the final part quality, is determined at every stage of the rubber injection molding process. By employing ultrasound and thermography, the rubber temperature resulting from the processing conditions can be modeled. On this base, the data-driven process monitoring system is built. This process monitoring system detects fluctuations causing faults by using a Principal Component Analysis (PCA) based approach. As a result, all available process signals can be evaluated simultaneously, even linear dependent ones. Additionally, by Fisher Discriminant Analysis (FDA), the type of fault can be automatically identified. Compared to other methods of process monitoring available for rubber injection molding, the presented data-driven system eliminates the need for any preliminary material tests, modelling or data selection. It is therefore much more straightforward and cost-efficient in implementing in smart manufacturing facilities.

KW - Spritzgießen

KW - Elastomere

KW - Prozessmonitoring

KW - maschinelles Lernen

KW - datenbasierte Modellierung

KW - Prozessmodelle

KW - Qualität

KW - Vernetzungsreaktion

KW - Injection molding

KW - Rubber

KW - Process monitoring

KW - Machine Learning

KW - Data-driven modelling

KW - Process modelling

KW - Quality

KW - Curing

U2 - 10.34901/mul.pub.2024.037

DO - 10.34901/mul.pub.2024.037

M3 - Doctoral Thesis

ER -