@

초록

The electrode tip is a critical component, responsible for applying pressure and current in the robotic spot-welding process. As the electrode tip degrades, its contact area increases, resulting in the smaller weld nuggets that adversely affect the welding quality. Therefore, it is essential to have a monitoring system that can predict the electrode tip area and estimate its remaining useful life (RUL). In this study, we propose a novel degradation index such as the Hurst Exponent Degradation Indicator (HEDI) And develop the 2-Stage Long Short-Term Memory (LSTM)-based RUL prediction model. First, various sensor data are collected, and electrode tip areas are measured in the robotic spot-welding experimental testbed. After the extraction of time-frequency features from the sensor data, the HEDI is computed, and the relevant features are then selected. A first-stage LSTM model, referred to as the Tip Area Prediction Model, is trained with the features that were selected to predict the current tip area. A second-stage LSTM model, denoted as the Tip Area Prognosis Model, is developed to predict future tip areas using previous measurements. The final RUL prediction can be realized by combining these two models, with the Tip Area Prediction Model providing estimated tip areas as inputs to the Tip Area Prognosis Model for RUL prediction. Finally, we evaluate the effectiveness of the HEDI-based model by comparing its RUL prediction capabilities with those of the conventional Feature Goodness Metrics Degradation Indicator (FGMDI).

키워드