LIBRARY ı6ıul 0) - Cranfield University
LIBRARY ı6ıul 0) - Cranfield University
LIBRARY ı6ıul 0) - Cranfield University
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
for predicting possibility of bad arc ignition and possibility of undercut were applied<br />
directly, without modification.<br />
In adapting the models, new welding trials were carried out using the<br />
BDH550. The range of welding parameters used in the trials was selected to cover the<br />
whole range of conditions normally used for gas metal arc welding thin sheet steel.<br />
The voltage levels were chosen to produce a stable process in all the trials. Models<br />
were then developed to map the behaviour of the BDH550, based on the set-up<br />
welding parameters and on the statistical features (see equations 2.25 to 2.32)<br />
extracted from each welding trial. The welding data was analysed using the multiple<br />
regression analysis tools provided by Statgraphicss and the model structures proposed<br />
by Ogunbiyi [ref. 51] were adopted. In order to reduce the number of experimental<br />
trials, the welding speed was fixed at 500 mm/min.<br />
Simple linear regression models were used to calibrate the leg length and weld<br />
penetration prediction models developed by Ogunbiyi [ref. 51] so that they could fit<br />
the corresponding measurements obtained in this work. This model calibration was<br />
necessary' to compensate or account for differences in the heat sink provided by the<br />
jigging system, the welding position and the welding power source. It should be noted<br />
that Ogunbiyi's models were developed for horizontal-vertical fillet joints whereas in<br />
this work the flat position was used. Due to the different welding positions and in<br />
order to reduce the masking effect caused by bead misalignment, the calibration<br />
functions were obtained considering the average values of the geometrical features of<br />
interest (i. e. average leg length and average penetration).<br />
The BDH550 models and the bead geometry calibration models (see section<br />
6.1) were successfully used in the welding parameter generator. Although calibrated,<br />
the penetration model was still imprecise, a fact which was also observed by Ogunbiyi<br />
[ref. 51]. However, considering that the quality criteria for penetration are quite<br />
flexible [refs. 107,190] and that penetration is normally assessed as either adequate or<br />
inadequate, the output of the model was found to be acceptable.<br />
The models developed were used to predict bead geometry and welding<br />
parameters in both dip and spray metal transfer modes. Although the bead and the<br />
penetration profile in both modes are essentially different, a single model was used to<br />
map welds from both modes of metal transfer. This was due to the fact that it is<br />
difficult to establish a limit above which a determined mode of metal transfer should<br />
predominate and, in addition, it is sometimes beneficial to use mixed mode transfer.<br />
8.3.2 Stand-off estimation models<br />
The stand-off estimation model proposed by Ogunbiyi [ref. 51] (see equations<br />
2.21 to 2.24) was initially implemented. Taking into consideration the differences in<br />
power sources, a new welding current model was built for the BDH550 using the<br />
model structure utilised by Ogunbiyi [ref. 51]. This model was a function of the power<br />
source set-up voltage, the stand-off and the wire feed speed (see equation 6.4).<br />
However, during the stand-off model (see equation 2.22) validation trials (see<br />
5 PC software for statistical analysis<br />
6A similar approach has been previously used by Doherty and Plummer (rcf. 1981 to calibrate a<br />
welding rig.<br />
192