Main Technical Achievements M36

Techedge activity on SMARTFAN

Last month’s Techedge efforts were focused on developing AI models ready to be deployed on ultra-low power IoT devices (as the Arduino BLE 33 Sense), working on real-time data acquired using several types of sensors (audio, roto-vibrational, video). A key aspect here is the fine tuning of the features extracted from data, since a proper features engineering allows to:

  • Decrease the amount of dirty data to be ingested by the AI, thus decreasing the computational cost of the AI itself
  • Increase the AI accuracy
  • Gives news insights on the problem which is under investigations.

Our principal feature engineering approach is based on spectral analysis techniques.

We tested ultra-low power AI application for anomaly detection of tests components using vibrational a sound data. This testing step was necessary in order to optimize neural network models in the perspective to be used with the data coming from SmartFan partners. Following this step, we plan to perform several tests on different architectural solutions, using different neural network configurations as:

  • RBM
  • Autoencoders
  • Convolutional 1D NN

Contact [email protected] for further details.

Self-healing capsules (FORTH)

·      One of the main tasks of Forth during this period was the production, characterization and provision to SMARTFAN partners of Self-healing capsules.

·      Initially the optimal sequestration method for the healing agent and polymerizing agent has to be determined. This sequestration can be achieved through encapsulation or phase separation.

·      The standard recipe for the preparation of PUF/DCPD microcapsules was adapted from that of Brown et al. [1].

·      In order to confirm that DCPD monomer was enclosed in the UF capsules DSC measurements were taken place.

·      To evaluate the healing effect of self-activated samples fracture toughness test was also held, similarly to GFRP system up to 50mm displacement. The modified CFRP samples presented the same mechanical behaviour as the neat (virgin) samples. For the modified specimens, at the end of the first loading, they restored to their initial status (zero loading) and left to heal for 48 hours. The results were promising as the modified specimens reached an average 80% of their initial maximum load.

The abovementioned research was carried out during the M12-M24 period of the project, mainly as a part of Work Packages 1 & 2.

[1]Brown, E.N., Kessler, M.R., Sottos, N.R. & White, S.R. 2003. In situ poly(urea-formaldehyde) microencapsulation of dicyclopentadiene. Journal of Microencapsulation 20: 719-730.