ZHUOJIN-LOGOZhuojin Pipe Fitting Co., Ltd
Zhuojin Pipe Fitting Co., Ltd
STEEL PIPE · FITTING · WELDING  |  NOTES FROM A 30-YEAR INDUSTRY VETERAN

Curve Fitting Method for Dual-Pulse MIG Welding Expert Database

Literature Overview and Core Findings

This paper by Lin Fang, Chen Xiaofeng, Wei Zhonghua, Cui Longbin, Gao Liwen, and Xue Jiaxiang addresses a practical problem in the development of digital welding machines: how to build an expert database that links the welding current to the wire feed speed in a way that is accurate, complete, and suitable for one-knob control. The authors propose a new data fitting and determination method. They first obtain 50 sets of data matching a given current to the corresponding wire feed speed through process tests, then fit these data in MATLAB using curve fitting methods of different orders, and evaluate the linearity and error control capability of the fits using three metrics: R-square, Adjusted R-square, and RMSE. The optimal order of the fitting curve is selected based on these metrics, and the complete expert database is generated from the optimal fitting curve. The experimental verification shows that the expert database generated by this method can effectively achieve the one-knob control function of the digital welding machine. The parameters selected for trial welding result in a stable welding process with no wire sticking or burn-back, and good weld quality. This is a highly practical paper that addresses a real production need.

Technical Interpretation and Data Analysis Insights

The core challenge in building an expert database is the relationship between the welding current and the wire feed speed. In a constant-voltage welding system, the current is determined by the wire feed speed and the arc voltage. For a given wire diameter and shielding gas, the melting rate of the wire is proportional to the current, and the wire feed speed must match the melting rate to maintain a stable arc length. The relationship between current and wire feed speed is not linear over the entire range; it is typically linear at low currents but becomes nonlinear at high currents due to the increased resistive heating of the wire and the change in the arc characteristics. The authors use curve fitting to capture this relationship. The choice of the polynomial order is critical: a low-order polynomial may not fit the data well, resulting in large errors, while a high-order polynomial may overfit the data, introducing oscillations and reducing the predictive accuracy. The authors use R-square, Adjusted R-square, and RMSE to evaluate the fits. R-square measures the proportion of the variance in the dependent variable that is predictable from the independent variable, Adjusted R-square adjusts for the number of predictors in the model, and RMSE measures the average deviation of the fitted values from the observed values. The optimal order is the one that maximizes the Adjusted R-square while minimizing the RMSE. In practice, this is a sound approach, but it must be complemented by a physical understanding of the process. The relationship between current and wire feed speed is not purely mathematical; it is governed by the physics of the melting rate, and the fitting curve should be consistent with the physical model.

Engineering Practice Integration and Lessons Learned

Let me bring this to the production floor. The one-knob control function is a key feature of modern digital welding machines. It allows the operator to set a single parameter, typically the current, and the machine automatically adjusts the wire feed speed and the voltage to maintain a stable arc. This simplifies the operator's task and reduces the risk of setting incorrect parameters. However, the accuracy of the one-knob control depends entirely on the quality of the expert database. If the database is inaccurate, the machine will set the wrong wire feed speed, resulting in an unstable arc, spatter, or burn-back. The authors' method of building the database from process tests and curve fitting is sound, but it has a critical limitation: the database is specific to a particular wire diameter, wire material, shielding gas, and torch configuration. In a production environment, the wire diameter and the shielding gas may change, and the database must be updated accordingly. The authors do not address this, but it is a practical concern. Another limitation is that the database does not account for the effect of the torch height, or the contact tip to work distance. The melting rate depends on the stick-out, and if the stick-out varies, the relationship between current and wire feed speed changes. A comprehensive expert database should include the stick-out as a parameter. The authors also do not address the effect of the pulse parameters in dual-pulse MIG welding. In dual-pulse MIG, the current is modulated at two frequencies: a high frequency for the pulse transfer and a low frequency for the weld pool stirring. The relationship between the average current and the wire feed speed is different from that in constant-current welding, and the database must be built specifically for the dual-pulse mode.

Key Questions and Reflections

The paper raises several questions that are worth considering. First, how does the curve fitting method perform when the data are noisy? The authors used 50 data points from process tests, but in practice, the measurements may have significant scatter due to variations in the wire feed, the arc voltage, and the shielding gas flow. The fitting method must be robust to this noise. Second, what is the effect of the wire material on the relationship between current and wire feed speed? The authors used a specific wire, but the melting rate depends on the wire's resistivity, which varies with the alloy composition. A database built for a steel wire may not be accurate for an aluminum wire. Third, how can the expert database be updated when the welding conditions change? The authors do not address this, but in a production environment, the wire diameter, the shielding gas, and the torch configuration may change, and the database must be updated accordingly. A practical solution is to build a library of databases for different conditions and to select the appropriate one automatically based on the machine settings.

Study Insights and Practical Recommendations

The most important takeaway from this paper is that the curve fitting method is a practical and effective way to build an expert database for digital welding machines, and that the use of R-square, Adjusted R-square, and RMSE provides a rigorous basis for selecting the optimal fitting order. For the production floor, this is a valuable tool that can improve the accuracy and reliability of one-knob control, reducing the risk of welding defects and improving productivity. In practice, I would recommend building the expert database using the method described in the paper, but with several enhancements. First, the database should be built for each combination of wire diameter, wire material, and shielding gas that is used in production. Second, the database should include the effect of the stick-out, either by building separate databases for different stick-out values or by incorporating the stick-out as a parameter in the fitting model. Third, the database should be validated over a wide range of operating conditions, including the full range of currents and wire feed speeds that are used in production. Fourth, the database should be stored in a format that can be easily updated and that is accessible to the welding machine's control system. The final thought is this: the expert database is the brain of the digital welding machine, and its accuracy determines the quality and consistency of the weld. The curve fitting method described in this paper is a solid foundation for building this database, but it must be applied with a clear understanding of the physical process and with a view to the practical requirements of the production environment. That is the lesson from decades on the shop floor, and this paper provides the tools to implement it.