Data analysis methods for process understanding and improvement in injection moulding production

Research output: Book/ReportPh.D. thesis

343 Downloads (Pure)


The fourth industrial revolution, sometimes known as Industry 4.0 is bringing a new focus on digitalization and data collection in different manufacturing industries. These changes are also observed in the area of plastic products manufacturing usually performed using an injection moulding process. In the plastic injection moulding industry, there is a continuous focus on improving product quality. The hope is that the newly collected data, mostly from sensors, will bring new possibilities in the area of product quality. The goal of this Ph.D. project is to make use of the available data for process understanding as well as product quality improvement. The Ph.D. project has been carried out in close collaboration with an industrial partner coming from the plastic injection moulding industry. Thus, all the contributions presented in the Ph.D. thesis are based on real problems. The Ph.D. thesis begins with an introduction chapter where the context of the Ph.D. project is addressed along with the contributions. A short introduction to injection moulding of plastics is presentedin a separate chapter. The chapter emphasizes the complexity of plastic products manufacturing using injection moulding and also presents the most common plastic defects. Contributions wise, the Ph.D. thesis is mostly focusing on predictive manufacturing, in other words how to use available process data for product quality prediction. All the contributions are presented as separate chapters in the Ph.D. thesis. In order tounderstand the complexityof product qualityassessment, one chapteris dedicated fully to an exploratory dataanalys is of a metrology dataset with intolerance products. Some of the analysis results are further used for creating a prediction environment. Due to the complexity of the industrial setting, a simulation study was preferred. The injection moulding process runs at high frequencies while the rate of acquiring quality measures is slow. This creates an abundance of process data with very few quality measures (responses). To build a model between process variables and quality measures, the focus has been on the incorporation of all available data. For this, a semi-supervised approach, where all available data is used, has been checked against benchmark latent methods which were built only on data with responses. If the data without responses contains outliers, then a semi-supervised model can perform badly in terms of prediction. In order to solve this problem, an iterative strategy using a combined statistic has been proposed. The problem of product quality prediction can also be seen as a classification problem with two classes where one of the classes contains good products, while the other faulty products. This results into a imbalanced problem since the number of faulty products is very low. In order to tackle this problem, a cost-sensitive learning strategy has been proposed and applied on data coming from the electronics manufacturing industry. It is important to mention that the strategy has not been applied on injection moulding data. The product quality can also be affected if the moulds are worn-out. Different types of data like mould characteristics, pre-chosen running settings, production and maintenance data have been used to predict mould worn-out with the help of clustering and survival analysis. Moreover, a dashboard containing an early warning system has also been implemented. Lastly, the final contribution is given in the form of lessons learned from the current Ph.D. project with a focus set on Big Data matters.
Original languageEnglish
Place of PublicationKgs. Lyngby
PublisherTechnical University of Denmark
Number of pages170
Publication statusPublished - 2019


Dive into the research topics of 'Data analysis methods for process understanding and improvement in injection moulding production'. Together they form a unique fingerprint.

Cite this