The Application of LVQ Neural Network for Weld Strength Evaluation of RF-Welded Plastic Materials

Abstract

Radio-frequency welding is a material joining process based on the dielectric loss principle, which is well suited to join plastic materials used in biomedical applications such as drainage and solution bags. For these kind of applications, it is very important that welds are of good quality due to the sterility and biological hazard concerns. This usually means that welds must have a certain predetermined strength. In production environment, destructive testing of sample products is a commonly used method of quality control. This is, however, time consuming and the results obtained on sample testing can by no means be fully relied upon for all the products within a batch. Therefore, a decision was made to design an artificial intelligence monitoring system that could determine the quality of each weld based on the measurements performed in real time. The main signal being measured is the displacement of the upper electrode. Various parameters of the displacement curve were found out to be in relation with the weld strength. Due to disturbances in air supply that result in variable welding force, a force sensor was added as well. The measurements obtained from both the sensors were used to form the input vector for linear vector quantization neural network. This type of network is suitable to put the input vectors in different classes. In our case, this means if a certain measurement corresponds to a good quality weld or a bad quality weld. The experiments have shown that the proposed neural network performs really well and could be of great value in a production environment.