16/01/2023

TEMACON Project: closure, development and our contribution as members of the consortium

The TEMACON project concludes after a period of work focused on applying Artificial Intelligence and Machine Learning to improve defect prediction in aerospace manufacturing processes. In this final phase, the consortium has worked on developing predictive models to anticipate the deviation of pre‑drilled holes in the Wing Lower Cover (WLC) of the Airbus A350 in the RJA area, a critical point for structural assembly.

 

Project development
To model the real behaviour of the process, an exhaustive analysis of the original dataset was carried out, consisting of thousands of observations and more than one hundred variables. Intensive data‑cleaning work was performed, including selection of relevant columns, noise removal, and consolidation of redundant information.
Subsequently, models from different families were trained and optimized: XGBoost, CatBoost and Deep Learning networks, combining traditional tuning techniques with advanced Bayesian optimization methods using Optuna, which enabled the identification of more accurate model configurations.

 

General conclusions
The developed models achieved a reduction of the error (MAE) to less than one third of the mean estimator, reaching values close to 0.03. Although this improvement is significant, the results highlight a key limitation:
the dataset is insufficient and unbalanced, especially in cases with large deviations—precisely the most important ones for predicting drilling failures.
The analysis concludes that the methodology is valid, but to reach industrial‑level precision it will be necessary to increase both the volume and the representativeness of the data, particularly in the most critical ranges of the process.

 

Idaero’s specific contribution
Within the consortium, Idaero has been responsible for the development of the AI‑based predictive software, leading:

• The preparation and cleaning of the dataset, removing duplicated, irrelevant or inconsistent columns and correcting structural issues.
• The training and optimization of XGBoost, CatBoost and neural‑network models, applying advanced large‑scale tuning and optimization methodologies.
• The automation of studies and validation through scripts that enable reproducible parameter adjustment and analysis of each variable’s impact.
• The strengthening of its R&D capabilities by expanding the technical team and consolidating expertise in AI applied to aerospace manufacturing processes.

 

We would like to thank all the companies and entities in the consortium for their collaboration throughout the project, as well as the joint effort that has made these results possible.

 

 

“TEMACON is a project under the Innovation Hub calls of the Community of Madrid and receives funding from the Community of Madrid.”

 

“Call for 2018 for grants co‑funded by the ERDF to contribute to improving public‑private cooperation in R&D&I through support for tractor‑effect technological innovation projects developed by open innovation hubs in the Community of Madrid.”

 

“Subsidised at 50% by the ERDF and the Community of Madrid with a contribution of €2,018,366.47.”