Abstract:
Resistance spot welding (RSW) is one of the most widely used welding processes
for sheet metal joining, especially in the automotive industry. One of the challenges
facing RSW is inconsistencies in quality of the weld. This challenge can be
addressed by implementing an on-line weld quality assessment and control. In this
study an on-line quality assessment and control model based on Learning Vector
Quantization Neural Network (LVQ-NN) system and Adaptive Neuro-Fuzzy
Inference System (ANFIS) is developed.
The ANFIS model is realized for identifying the RSW dynamical system based on
given input output data. It can be used to approximate nonlinear systems with
minimum training data, quicker learning speed and higher precision.
An indirect estimation of the weld quality employing an LVQ-NN type classifier
was designed to provide a real time assessment of the weld quality.
Experiments were conducted to establish the effects of various parameters, such
as the welding time and the welding current on the quality of the weld produced.
A set of important parameters was then selected as the input data to train the
proposed LVQ-NN type classifier and ANFIS controller. Once the monitoring and
control system had been trained, it was then tested to evaluate their validity in
the RSW process.
The results show that the classifier based on LVQ-NN was able to classify the weld
quality into normal, cold and expulsion based on their corresponding dynamic
resistance curves obtained as a function of welding time. The recognition rate was
82 percent for the test data. The proposed control algorithm based on ANFIS
demonstrated robust performance reducing the number of expulsion welds by 30%
xiii
compared to the conventional controller, while increasing the number of normal
welds by 31% in RSW process.
