Hi,
There is a very interesting lab project that was in the project lists that we want to try doing. It's about audio processing to try to identify defects in materials during processing.
The issue is we have very limited experience in audio ML, the lab that proposes the project also has limited experience in machine learning techniques, and we have no idea of the feasability of such a project.
Here is an exerpt from the explanations of the lab for the project :
The dataset is organized in terms of the different alloys that are being studied, and the different defects that are being created during the print. The acoustic sensors are positioned close to the process zone and the emitted signal is recorded. The idea is to compare signals that were acquired in the same conditions, in one case the defect or any transformation takes place and in the other case it doesn’t. We would use ML in order to classify the different types of defects and transformation depending on the alloy, by creating them on purpose. For the moment, we are studying recrystallization in 316L and austenitic decomposition in Ti64. The final goal would be to be able to identify what happens during the process thanks to the ML model, based only on the signal and without needing to cut the sample after the experiment to verify if the defect really exists or not.
Their dataset also is kind of small, which is again an issue because we have no idea if huge datasets are needed for audio processing (data from a few experiments). The question is, at our current level, working a reasonable amount of time each week, can we actually do something out of this project? Is it okay if we don't manage to have interesting results?
Project feasibility
Hi,
There is a very interesting lab project that was in the project lists that we want to try doing. It's about audio processing to try to identify defects in materials during processing.
The issue is we have very limited experience in audio ML, the lab that proposes the project also has limited experience in machine learning techniques, and we have no idea of the feasability of such a project.
Here is an exerpt from the explanations of the lab for the project :
The dataset is organized in terms of the different alloys that are being studied, and the different defects that are being created during the print. The acoustic sensors are positioned close to the process zone and the emitted signal is recorded. The idea is to compare signals that were acquired in the same conditions, in one case the defect or any transformation takes place and in the other case it doesn’t. We would use ML in order to classify the different types of defects and transformation depending on the alloy, by creating them on purpose. For the moment, we are studying recrystallization in 316L and austenitic decomposition in Ti64. The final goal would be to be able to identify what happens during the process thanks to the ML model, based only on the signal and without needing to cut the sample after the experiment to verify if the defect really exists or not.
Their dataset also is kind of small, which is again an issue because we have no idea if huge datasets are needed for audio processing (data from a few experiments). The question is, at our current level, working a reasonable amount of time each week, can we actually do something out of this project? Is it okay if we don't manage to have interesting results?
Thanks,
Add comment