Neural network evaluation of weld quality using FSW feedback data
- Boldsaikhan E. ,
- Corwin E. ,
- Logar A. ,
- et al
- Boldsaikhan E. ,
- Corwin E. ,
- Logar A. and
- Arbegast W.
Links
Repository
Description
A study was conducted into a proof of principle for the effectiveness of a neural network in identifying the presence of metallurgical defects and for evaluating tensile strength from the feedback provided during the friction stir welding (FSW) process, and thus, from the resulting classification, for certifying weld quality. The materials investigated were 7075 and 2024 aluminium alloys. The feedback parameters selected, X force, Y force, Z force and torque, were shown to be statistically correlated with weld quality. A discrete Fourier transform (DFT) was applied to the forces individually and, after feature extraction, feature vectors were generated for each weld segment. Multilayer perceptron neural networks trained using the back propagation algorithm were employed to classify welds with regard to defects and strength.
6th International Symposium, 10-13 Oct 2006, Session 4A: Quality I, Paper 02
Related Records
