Machine Learning for Data-Driven Quality Control

The image is a layout of various technological processes and applications related to laser technology and machine learning. At the top left, a close-up of a battery cell shows a laser beam making contact with the cell, surrounded by metallic components. Below this, a laser structuring process is depicted, with an engraving of "ILT" and a faint laser plume. Centered in the layout are multiple boxes outlining steps in a "Machine Learning Pipeline," including Data Integration, Data Preparation, Modeling, and Deployment, represented with icons of people. To the bottom left, there’s an image of a real metal component next to a colorful digital twin, illustrating data-driven quality control. Finally, the bottom right features a precision laser nozzle emitting a focused beam, labeled "Additive Turbine Blade Manufacturing using Laser-Metal-Deposition." The overall color scheme includes dark backgrounds with bright highlights and text.

Machine Learning for Data-Driven Quality Control

Summary

Data-driven modeling with machine learning algorithms allows for the extraction of knowledge and identification of features in complex data sets.


Topic Fields	Data Analytics
Published	2021
Involved Institutes	IPT
Project Type	ICNAP Research/Transfer Project
Responsibles	Lars Leyendecker

Description

Data-driven modelling utilizing machine learning (ML) algorithms facilitates the extraction of knowledge and the identification of process characteristics from high- dimensional data. In this project, we investigated on four different use-cases, how ML can enhance process development and quality control in manufacturing. For this, data- driven process models were developed for (1) laser beam welding in battery cell contacting, (2) ultra-short-pulse laser structuring of metallic surfaces, (3) 3D-printing of turbine blades using laser metal deposition, and (4) recombinant protein expression. According to each process, supervised or unsupervised ML models were trained and analyzed based on process parameters, process monitoring data and quality measures.

Our results demonstrate the enormous potential of ML-based data processing for quality, efficiency, and sustainability enhancement. The developed data models allow not only to predict and to control the final product qualities, but also to identify the sequence of influence of process parameters on the outcome. Our research further shows that in real- world applications, data quality is paramount to the performance of the models. Data quality proved to be more important than its mere quantity. The future design or retrofitting of processes should therefore aim for achieving a highest possible data quality. Analyzing and processing these high-dimensional and large datasets places utmost demands on computational infrastructure and resources. Furthermore, for feeding gained insights back into process development, but also for relying on the models during operation, it is decisive that the model functionalities and their predictions must be transparent and comprehensible for both the process experts and operators. The optimization of data quality, model performances and also the explainability of ML will be the focus of our future work in order to achieve a benefit-maximizing, area-wide use of trustworthy ML models in a variety of production domains.

Get in touch