Research// Machine Learning

SmartPro – Key to Smart Products!

Within the SmartPro impulse projects BEYOND (2020-2022) and Smart-DATA (expected to start in spring 2022), artificial intelligence is used in the form of Machine Learning (ML) methods. In particular, methods for quality control and process monitoring are investigated and further developed by the SmartPro researchers. Through close exchange with the SmartPro application fields, ML methods are developed specifically for the respective application purposes. Data- and algorithm-based models are built through iterative artificial learning processes that can subsequently recognize relationships within the data, identify patterns or errors, and automatically perform comprehensive analyses.

This establishes a toolbox that provides methods for a wide variety of problems that can be easily adapted for the defined purpose. By using the ML methods in all SmartPro application fields, the partnership is even more intensively connected.

[dsm_image_carousel gallery_ids=”5319,5285,5305,5297,5313,7408,7404″ sizes=”large” slide_to_show=”6″ pause_on_hover=”on” dots=”off” arrow_color=”#FFFFFF” arrow_background_color=”#254B94″ slide_to_show_tablet=”3″ slide_to_show_phone=”” slide_to_show_last_edited=”on|tablet” arrow_size_tablet=”” arrow_size_phone=”28px” arrow_size_last_edited=”on|phone” _builder_version=”4.16″ _module_preset=”default” saved_tabs=”all” locked=”off” global_colors_info=”{}”][/dsm_image_carousel]

Smart-DATA

// Project duration: 01.04.2022 to 31.03.2026

 

Project management

Prof. Dr. Orsolya Csiszár
Mathematik und KI-Anwendungen
Phone: +49 (0) 7361 576-5567
orsolya.csiszar@hs-aalen.de

 

Project partner

Aalen University
Companies
  • Carl Zeiss IMT

  • Carl Zeiss Microscopy GmbH /

  • F. & G. Hachtel GmbH & Co. KG

  • Kessler & Co. GmbH & Co. KG

  • Oskar Frech GmbH & Co. KG

  • PVA TePla Analytical Systems GmbH

  • Wieland-Werke AG

Other research institutions
  • Universität Trier – Informationswissenschaften Wirtschaftsinformatik

Smart-DATA

Domain-optimized machine learning methods for smart production systems

The manufacturing of the next-generation of products requires complex processes and high-quality materials. As a result, the need for innovative procedures for quality control and optimization during production is also constantly increasing. However, classical systems for quality control are often insufficient to meet these increased challenges.

Smart-DATA, therefore, deals with novel ML-based systems and their integration into production and quality control. The decisive factor is that the methods are tailored to the respective research questions or industrial applications and can be easily adapted to other needs. The in the previous project specifically for materials research developed deep learning methods are purposefully advanced for smart, process-accompanying applications for energy-efficient and resource-saving products. Examples include quality assessment in die casting, hybrid components, and magnetic and battery materials.

BEYOND

Domain-optimized highly adaptive deep learning methods for materials research.

The development of novel smart material systems has led to increased requirements in the optimization of process parameters, as well as in the areas of microstructure analysis and quality assessment. These often cannot be handled by classical approaches. The use of self-optimizing, high-precision, and easily adaptable ML methods, more precisely Deep Learning, was the topic of the impulse project BEYOND.

In close cooperation with the four other impulse projects of the SmartPro start-up phase, established Deep Learning methods are specifically adapted to the respective needs and further developed. Particular attention was laid on reducing the required data volumes and training intervals to enable efficient analysis. This allows the application in wide areas – even with limited data volumes.

Another central aspect of the project is the optimization and further development of existing systems based on so-called Convolutional Neural Networks (ConvNets) – artificial neural networks modelled on the human brain – as well as pre-processing methods for preparing the data and combining them for effective use in materials research. The aim is to achieve the broadest possible range of applications, including the combined analysis of imaging and non-imaging data.

BEYOND

// Project duration: 01.01.2020 to 31.08.2022

 

Project management

Prof. Dr. Dagmar Goll (from 08/2021)
Materials Research Institute Aalen
Phone: +49 (0) 7361 576-1601
dagmar.goll@hs-aalen.de

Prof. Dr. Ricardo Büttner (bis 07/2021)
Business Informatics, Aalen University
(since 08/2021 University of Bayreuth)
ricardo.buettner@uni-bayreuth.de

 

Project partner

Aalen University

Follow us!

You can also stay up to date
about the SmartPro network
on LinkedIn.

Click over and have a look!

Explorative projects

Through the many facets of Artificial Intelligence methods, this cross-sectional technology links the application fields with each other – evident by the many explorative projects that were carried out together with the application fields. The main focus was on Machine Learning methods for improving manufacturing processes and quality assessment.

PreMo-LiB // 01.03.2019 to 31.08.2020

Artificial Intelligence and Machine Learning as enablers for improving process quality in battery mass production

High-quality lithium-ion batteries with a long service life are needed in many industrial products, e.g. in electric cars, smartphones or power tools. In the explorative project PreMo-LiB, we in cooperation with the Varta Microbattery GmbH investigated innovative in-line methods enabling to predict the service life of accumulators and to improve their overall quality. To achieve this, we used modern self-learning software algorithms Machine Learning (applied Artificial Intelligence).

Commonly, it is only in later usage that a lithium-ion based battery is revealed whether it meets the customer requirements. Due to the complex physical interactions within a battery, so far, quality prediction during production has been only possible to a very limited extent and not suitable for mass production.

But remarkably, in the project PreMo-LiB, Machine Learning methods could have been developed to predict the quality and service life of (test) batteries during production in a both cost-effective and feasible manner. Most importantly, the established quality assurance process is non-destructive and scalable for the mass production of high-quality lithium-ion batteries.

  • Project management

    Prof. Dr. Ricardo Büttner, Business Informatics

  • Project partner

    VARTA Microbattery GmbH

DiMa // 01.10.2019 to 30.09.2020

Digitization potentials of materials research in SmartPro

The explorative project DiMa on Machine Learning methods was carried out in four subprojects, each of which focused on one of the applications fields of SmartPro or Additive Manufacturing, respectively. Here, the methodological competence of ML experts was combined with expertise in the particular research areas. In this way, interdisciplinary approaches were successfully used to push research further towards tailored (development and) application of ML methods within SmartPro. Each of the four subprojects served as a starting point for the current cross-sectional impulse project BEYOND with a focus on Machine Learning.

The subproject MagTwin (assigned to the impulse project MagNetz to energy converters) focused on the development of a digital twin of a permanent magnet test bench. Most importantly, the aging processes of permanent magnets were simulated.

Based on the subproject DigitEL on energy storage systems (assigned to the impulse project LiMaProMet) the application of Machine Learning in the analysis of microstructures of electrode material in lithium-ion accumulators was improved with respect to performance parameters. A particular focus was on the prediction of current rate capability.

The subproject with respect to lightweight construction (and to the impulse project InDiMat) investigated the surface properties of adhesive-bonded multi-material systems based on carbon fiber-reinforced plastics (CFRP). For this purpose, the systems were analyzed microscopically; images were collected using 2D and 3D systems. Based on these and supplementary data, Machine learning approaches were used to predict the mechanical strength of the bonded joints.

In SmartPrint (assigned to the AddFunk impulse project), the relationships between the quality of 3D-printed optical components and their process parameters as well as material properties were investigated using Machine Learning.

  • Project management

Prof. Dr. Ricardo Büttner, Business Informatics

  • Project partner

Prof. Dr.-Ing. Sebastian Feldmann, Digitale Systemintegration im Maschinenbau

Prof. Dr. Ulrich Klauck, Machine Learning and Data Analysis

Prof. Dr. Manfred Rössle, Business Informatics

SmartPro // FH-Impuls:
Strong universities of applied sciences – impulses for the region

With SmartPro, Aalen University of Applied Sciences has positioned itself in the top group of universities of applied sciences nationwide. SmartPro is one of ten partnerships funded by the funding measure “FH-Impuls” of the Federal Ministry of Education and Research with around 10 million euros from 2017 until 2025. Core objectives are the sustainable expansion of the regional transfer and cooperation network, the strengthening of research, and innovative power. SmartPro makes contributions to social challenges such as climate protection and digitization.