|
| Welding type | Target output | Machine learning model used | Parameter considered | Result | Ref. |
|
| Resistance spot welding | Prediction of nugget size | Finite element model (FEM) | Temperatures, electrode displacements, and electric voltage drops | Much simulation offers many labeled data due to physics-based simulation data. There is no measurement sensor precision issue in data generated from a simulation model | [119] |
|
| Laser beam welding | During deep penetration welding, the capillary depth of the keyhole is calculated | Optical coherence tomography (OCT) | Laser power, welding speed, radiation mode, angle of incidence | Discrete wavelet transformation helps to classify the weld quality based on height parameter, and low pass Chebyshev helps in reducing the amount of process | [120, 121] |
|
| Resistance spot welding [52] | Nugget width prediction | Deep neural network | Strength and weight of the material, cost, manufacturability, and environmentalism | In RSW joints, DNN has much higher precision and lower variation for nugget width prediction because RSW data is highly nonlinear and complex | [122] |
|
| Ultrasonic metal welding | Prediction of USMW joint strength | Regression model, ANN, and ANFIS | Weld pressure, vibration amplitude, weld time, and T-peel failure loads | The number of neurons, learning rate, and momentum facture of the ANN set to 10, 0.08, and 0.6. MFs are also counted as 4 in the ANFIS system | [123] |
|
| Submerged arc welding | Prediction of transient temperature in SAW process | Multilinear regression | Voltage, trolley speed, stick out, current | Proves to be the robust method since computation cost was less and the estimation of transient temperature was good. Also, help to a swift resolution of several decision processes | [124] |
|
