Attention-based BiLSTM-XGBoost model for reliability assessment and lifetime prediction of digital microfluidic systems

Traditional methods for reliability and lifetime testing of digital microfluidic systems heavily rely on real-time monitoring data. This often leads to evaluation lag and limits their application, especially for complex droplets. To address these issues, this study proposes a novel prediction model for digital microfluidic (DMF) devices. The model combines an attention-based bidirectional long short-term memory (BiLSTM) with eXtreme Gradient Boosting (XGBoost) using a Stacking approach. This integrated model efficiently identifies the health state and predicts the failure time of digital microfluidic devices. This approach overcomes the limitations of traditional methods, such as over-reliance on sensor feedback and detection hysteresis. Experimental results demonstrate high prediction accuracy. The model achieved a mean absolute percentage error (MAPE) of 1.6464, Root mean squared error (RMSE) of 0.3667, mean absolute error (MAE) of 0.2557, and a coefficient of determination (R-squared) of 0.9949. Compared to baseline methods, the proposed BiLSTM-XGBoost model achieves the highest prediction accuracy, enabling effective health monitoring, problem identification, and failure prediction. Ultimately, this improves system reliability and lifetime with greater timeliness and accuracy.