Digital Twin-enabled failure prediction for indoor air quality

Abstract

Accurate failure prediction is critical to achieving Predictive Maintenance (PdM) for Indoor Air Quality (IAQ), which is highly related to resident well-being and operational effectiveness. However, most existing studies emphasise anomaly detection rather than prediction. To develop a precise and robust method for pre-emptive IAQ warning, this article integrated Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Muti-Head Attention (MHA) mechanism into a novel C-B-M model, synergistically incorporating feature extraction, temporal dependency analysis, and contextual weighting mechanisms. Additionally, a real-world dataset collected from various buildings in Singapore is employed in a detailed comparative experiment with other benchmark models for different prediction periods, dataset selection, and failure severity levels to illustrate the effectiveness and robustness of the proposed method. Finally, a Digital Twin (DT)-oriented failure prediction framework for the indoor climate is introduced and validated through the prototype system demonstrating the 3D building model and IAQ alert information.