Data-Driven Flexible Electronic Processing Equipment Performance Degradation Trend Forecast

In response to the issues of a single construction method for flexible electronic processing equipment performance indicators, lack of performance indicator information, and low accuracy in predicting performance degradation trends, this article proposes a prediction method based on the degradation trend of equipment performance using self-encoding coders and multi-header attention mechanisms. This method first extracts multiple types of mechanism characteristics from the original vibration signal, taking into consideration their monotonicity, robustness, and correlation, and selects virtual features that are conducive to performance prediction. Secondly, to address the issue of redundant characteristic information, a self-encoding coder is utilized to integrate these features and construct performance indicators. Additionally, the combination of a multi-header attention mechanism and long short-term memory (LSTM) network is investigated to enhance the weighting of important degradation information and disregard redundant information, thus addressing the limitations of existing models. Finally, the effectiveness of the proposed performance prediction method is validated through experiments conducted on a bearing dataset.