This is a continuation of a previous post about one of the goals of the MAT4.0 project, the introduction of machine learning algorithms in the manufacturing processes related with composite materials. In that post it was presented the challenge of improving the evaluation of components to ensure that the industrial parts fulfill quality standards. In this case the focus is in presenting the advances in introducing a data-approach in the field of material science.

Machine learning and deep learning models have experienced a huge success in several fields ranging from business and IT to medicine. New applications are developed every day due to the massive disruptive efficacy improvements they undertake. The introduction of these techniques in the material science presents obstacles that have been addressed during the project: the data scarcity, data curation and preparation, the shift to a data-driven methodology and the interpretation of the models.

If in the previous industrial revolution, the electricity and oil were the fuel of the innovations such as the cars, light bulbs, and trains; data is today´s fuel. However, data is scarce in the field of manufacturing. During the MAT4.0 project one of the key advances is the creation, augmentation and curation of the data needed to fuel the models used in the evaluation of manufacturing components.

The digitalization of the processes is more than using new technology to solve the same problems. Society needs of relearning and tackle traditional problems from a new approach. In the MAT4.0 this shift takes form in the developed data-driven methodology. An end-to-end approach that starts in the experimental tests, covers the data cleaning, the modeling and its evaluation considering relevant factors like the scalability and its industrialization for production environments.

Last but not least, the evaluation of manufacturing components is critical for the safety of the airplanes, buildings and cars. The machine learning models employed to improve this evaluation must not be “black-boxes”. Extraordinary estimations would be distrusted if field professionals cannot explain those results. This is a reason why during the MAT4.0 project many efforts are employed in data visualization and new approaches for model validation.

The work carried out has so far resulted in estimations of the porosity content that improves the ones of the state of the art. Since porosity is the most frequent fabrication defect in composite materials advances towards this direction no doubt will be useful for the sector. Results obtain will be subject of a scientifical publication summarizing all the advances. The future work includes the assessment of porosity relevant factors, mostly related to geometry