Abstract
The development of reliable laser welding processes within a short time and with minimum experimental effort is an important aspect for small batch-size manufacturing. A physics-informed hybrid model was applied for the prediction of the threshold of deep-penetration laser welding. A “residual model” approach was used where a machine learning model was applied to learn and compensate for the deviations of an analytical model to the experimental results. Gaussian processes were used for the machine learning part. The results show an increase in model accuracy by using such a hybrid model compared to only using the analytical model. In comparison to only using a black-box machine learning model, the amount of required training data can be reduced and the extrapolation capability can be improved.
Users
Please
log in to take part in the discussion (add own reviews or comments).
