Results 2021-2 BRIDGES Call

Abstract

Filament winding is a fast and cost-effective manufacturing process used to create lightweight and high-performance fiber reinforced composites. At GRADEL, the robotic endless filament winding xFKin3D is an automated process producing ultra lightweight 3D composites used for instance in transportation. The next step of development is to free the process from conventional petroleum-based resins. An ideal approach would be to employ bio-based resins which could heal, be reshaped and recycled, to decrease the global carbon footprint of the composites.

Vitrimers are polymers that could fulfill this objective and provide exceptional opportunities of enhancing the manufacturing process.

The project idea of GREENSHAPEr is to develop bio-based vitrimer formulations and to adapt the xFKin3D process to produce 3D composite parts made of these vitrimers. It will be carried out from the lab scale to a TRL 4 demonstrator, by a close cooperation between GRADEL and LIST and their joint lab.

Abstract

Modern technology such as 5G telecommunications, automotive radar or the ‘Internet of Things’ relies on ever faster data treatment and transmission in the frequency range above 30 GHz, putting new demands to the electrodes: at low frequencies, electrical loss in the electrodes depends mainly on the electrode thickness, but surface roughness or grainy microstructure become more important as the frequency increases. The fact is empirically known, but there is yet no numerical model to connect structural parameters, resistivity and loss. This project investigates different commercial copper foils for microelectronics and compares them with copper films produced by sputtering, evaporation, electroplating and ink-jet printing. Structural properties of the electrodes are determined e.g. by X-ray Diffraction, Electron and Atomic Force Microscopy, and electrical properties up to 110 GHz are measured using a Vector Network Analyzer. A numerical Finite Element Model is developed to connect the parameters. Based on this model, optimized structures will be proposed to minimize electrode loss.

Abstract

The current market of electronics is increasingly linked to the miniaturization of components, contributing to the digitalization of society while addressing environmental concerns by reducing raw material consumption. In this context, ultrathin copper foil is a material of choice to produce miniaturized printed circuit boards for smartphone, tablet and computer applications. To reinforce the competitiveness of Circuit Foil Luxembourg (CFL), one of the last European manufacturers of ultrathin copper foil, the development of new products is essential. The ultrathin copper foil processed by CFL (thickness between 1.5 and 5 μm) is supported by a carrier copper foil (thickness of 18 μm) that is peeled off by the customers after lamination to produce printed circuit boards. The peelability of the carrier foil is ensured by an inorganic release layer (thickness of 14 nm) made of nickel-based alloys and derived oxides. To our best knowledge, the interfacial adhesion mechanisms of this release layer are still unknown, and hence, are not controlled. This is mainly due to the nanometric thickness of this release layer requiring cutting-edge characterization techniques with a nanosized probe to accurately determine its physic-chemical structure. This BRIDGES project conducted by the Luxembourg Institute of Science and Technology (LIST) in close collaboration with CFL has two objectives: (i) identifying the structure of this nanometric release layer and the corresponding adhesion mechanisms, and (ii) developing new product demonstrators with a controlled interfacial adhesion based on the gathered scientific knowledge. To address the first objective, a methodology employing atomic force microscopy, transmission electron microscopy and atom probe tomography will be developed. For the second objective, a constant peel strength comprised between 15 and 25 N/m is targeted, relying on optimizing the actual process, modifying it, or developing a post-processing treatment.

Abstract

The project OptiSimCVD proposes a data-driven framework for prediction, sensitivity analysis and uncertainty quantification in industrial-scale processes used to produce hard coatings and wear protection. The core of the production process is Chemical Vapor Deposition (CVD) reactors with different set up but common goal: uniform coatings with strict quality requirements. With the proposed computational framework, different clusters of reactors will be identified, with different set-up but similar qualitative characteristics of the coating. Then, in each one of the clusters, predictive models will be developed, able to correlate the inputs of the process to the output. Eventually, efficient and accurate process models will be implemented in the context of uncertainty quantification and sensitivity analysis with the ambition to contribute to process efficiency by reducing scrap rate (30%) and improve quality by enhancing homogeneity (15%).

ICT: 2 projects 

Abstract

Thanks to the recent advances in Artificial Intelligence (AI), it is now possible to mimic the creativity of humans. Some examples include the recent tools for automatically generating drawing portraits or sketches. Another important example with a massive impact in industrial applications is the generation of Computer-Aided Design (CAD) models directly from physical objects. CAD models are now the standard option for designing objects ahead of manufacturing. Generating CAD models from physical objects is often called “3D reverse engineering” and consists of scanning objects with appropriate 3D scanner devices and applying algorithms to infer CAD models. In this project, we will develop learning-based algorithms to automatically infer CAD procedures from 3D scans.

Abstract

Effective maintenance is key for any manufacturing company, as it extends equipment life and contributes to improve the OEE (Overall Equipment Effectiveness). A maintenance process consists of several steps, among which two important ones: (1) the prediction of equipment failures/anomalies based on different types of (sensor) data; (2) the planning/scheduling of maintenance actions to be conducted. While a number of predictive maintenance (PdM) frameworks have been proposed in the literature to properly address these steps, they still suffer from limitations. First, regarding step (1), existing frameworks often implement Machine Learning (ML) algorithms without properly analyzing how vulnerable those algorithms are to possible adversarial perturbations. Adversarial perturbations here refer to cases where time series (sensor) signals/data** in the operational phase may (unexpectedly) differ from the ones used for ML model training, which, as a consequence, may lead to incorrect output or misclassification. Regarding step (2), the majority of the frameworks proposed in the literature to optimize maintenance task scheduling and operator assignment make use of metaheuristic algorithms, which perform extremely well under “deterministic” environments (i.e., where information is known a priori such as the number of tasks to be performed, the task duration and level of criticality, the operator availability, etc.), but quickly underperform in “stochastic” (dynamic) environments due to uncertainty about the future (e.g. sudden equipment failures, variable maintenance times, resource unavailability).

The UPTIME4.0 project, which stands for “robUst PredicTIve MaintEnance for Industry 4.0”, is aiming at addressing these two limitations under the industrial environment of our private partner, Cebi Luxembourg S.A. At step (1), UPTIME4.0 investigates new strategies to discover, in an efficient and timely manner, possible adversarial perturbations that may occur during the manufacturing process. Once identified, the goal is to robustify the ML model against such perturbations. At step (2), UPTIME4.0 investigates how reinforcement learning can be used to improve existing approaches for the dynamic maintenance task scheduling and operator assignment in an “opportunistic” manner (i.e., by identifying cost-effective time windows of opportunity for maintenance that mitigate the impact of rescheduling on the production). As both contributions rely on ML methods, an efficient ML pipeline aims at being created to help Cebi Luxembourg S.A. to significantly reduce deployment and maintenance efforts to put a model from research into production.

All the strategies/algorithms developed in UPTIME4.0 will be deployed, tested, refined and validated based on both (i) a PdM use case scenario defined at the shop-floor level of Cebi’s factory, and (ii) existing scientific datasets, along with comparative studies with state-of-the-art approaches. Overall, the project outcomes and results will not only benefit Cebi Luxembourg S.A., but the broader scientific community, as well as any company that deals with asset maintenance and repair.

** Time series data is widely used in industrial applications to express data coming from sensors/assets.

Abstract

CAMPHIBIAN will push forward the development of a novel approach to improve amphibian monitoring based on an underwater wildlife camera trap (NEWTRAP) prototyped during a previous FNR-funded project. NEWTRAP is the first underwater device producing high-resolution time-series of automatically recorded images of amphibians. Images can be analysed with artificial intelligence to inform on species presence or abundance. CAMPHIBIAN will fix some technical defects, improve the user-friendliness of NEWTRAP, and broaden its scope of application for monitoring a wide range of amphibian species. Field and lab experiments will assess the added value created by such a technological innovation. CAMPHIBIAN will bring NEWTRAP closer to a fully-fledged version via a partnership with a world-leading company producing wildlife sampling equipment (NHBS). CAMPHIBIAN will open new market opportunities for NHBS and pave the way for the development of a living lab on underwater camera trap data integration.

Abstract

Modern medicine is highly dependent on the rapid diagnosis of patient specimen for the selection of an appropriate treatment to have the best possible outcome for the patient. Current pathological laboratories are using histology to characterize the specimen. Here the pathologist provides based on the use of a number of biological marker the pathological diagnosis. In this project we will develop a mass spectrometry-based methodology, which will expand the number of biomarkers to allow a more precise diagnosis by the pathologist. The method will be compatible with existing techniques to preserve the biological specimen by fixation in formalin.