2005 Fiscal Year Final Research Report Summary
Sensing and Control of weld pool in MIG welding with neural network arc sensor
| Project/Area Number |
16560626
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Material processing/treatments
|
|---|
| Research Institution | Saitama University |
|---|
Principal Investigator |
YAMANE Satoshi Saitama University, Graduate school of Science and Engineering, Associate Professor, 大学院・理工学研究科, 助教授 (10191363)
|
|---|
| Project Period (FY) |
2004 – 2005
|
| Keywords | Switch back welding / Back bead control in one side / Estimation of wire extension / CCD Camera / Neural network / Observation with high speed video / Welding robot / Pulsed MIG welding |
|---|
| Research Abstract |
In order to get a high quality of the welding, the stable back bead is required during the welding. By using conventional welding method, the backing plates are needed to support the weld pool. If the enough weld pool can not be obtained into the backing metal plates, there is some notch. If the strong force acts on the notch, the crack takes place. In order to avoid the crack, the control of the weld pool is important and depends on the arc length and the wire extension. For this purpose, the sensing of the arc length and the wire extension are carried out. Since the behavior of the wire extension is nonlinear, it is difficult to describe the differential equation. The neural network is suitable to describe the nonlinear characteristic. The arc length and the wire extension are estimated by using the knowledge of the skill welders and the fundamental experiment result, i.e., the neural network is constructed to estimate the wire extension. If the extension is given, the arc length is
… More
calculated by using the relationship between the current and the torch voltage. In order to train the neural network, the pulsed MIG welding is carried out on the flat plate. It is taken by the high speed video camera, every 1ms. The welding current, the voltage and the wire extension are measured, too. These data are used to the training data for the neural network. In the pulsed MIG welding, the melting of the wire depends on the Joule heat, i.e., the effective value of the voltage and the current are related to the wire melting. It depends on the current, wire extension length and the wire feed rate. By using this knowledge, the input variables are determined in the neural network for the estimation of the wire extension. Sampling periods is 10ms. The data during 50ms are used as the training data. The neuron in the input layer is 20. The output is 1. That is, the extension wire is estimated by using the current, the voltage and the wire feed rate. Moreover, the control of the weld pool is considered in the butt welding for the base metal with 12mm thickness and 45 degree groove without the backing plates. By carrying out the fundamental experiments, the welding conditions are investigated. The weld pool was control according to the gap, so as to get the high quality of the welding. Less
|
