In order to meet this objective a laboratory test bench for the installation of blind fasteners was set up. The system allows to acquire and store the installation torque-angle diagrams. Using this equipment an experimental study was made to understand the causes that lead to defect installations and how they affect to the torque-angle evolution. After analysing a set of preliminary diagrams and their installation outcomes (i.e. formed heads) a full test campaign was used to design and implement a data-driven monitoring solution for detecting fault installations. The main outcomes of this work are:
* The method for conditioning the torque-angle signals and prepare them for data extraction has proved effective under different types of installation conditions and fastener references
* A set of descriptors (i.e. statistics, regression coefficients, representative points) describes the torque-angle signal evolution along the installation despite varying conditions and the quality of the installation
* By using the signal descriptors both shallow and deep learning algorithms (CSVM, logistic regression, LSTM-FCNN) enable to differentiate with reasonable accuracy correct installations from in-air ones, too high/high/short formed heads and double formed heads. Shallow learners classify with very high accuracy and in a simple way in-air installations from the correct ones.
* The monitoring solution runs in the laboratory test bench for the automatic evaluation of installations
Some additional information about the project results is available at public access publications
https://doi.org/10.1007/978-3-319-59650-1_17(odnośnik otworzy się w nowym oknie) and
https://doi.org/10.3390/ma12071157(odnośnik otworzy się w nowym oknie).
