Deep Learning Based Digital Twin Framework for Predictive Analytics in Industrial Iot Environments

Over the past few years, Digital Twin (DT) technology has emerged as an integral tool for enhancing predictive analytics in the Industrial Internet of Things (IIoT) by virtually replicating physical systems and enables real time condition monitoring. However, traditional DT frameworks often lack the integration of Deep Learning (DL) model which resulted in delayed response due to cloud-based processing, low accuracy in complex degradation conditions and lack of ability to model temporal dependencies. Therefore, a hybrid DT-driven predictive analysis which integrates 1-Dimensional Convolutional Neural Network with Long Short-term Memory (1D CNN-LSTM) is proposed for accurate degradation trend prediction in IIoT environments. Initially, time series sensor data is collected from IIoT sources such as vibration, pressure and temperature. Next this collected data is preprocessed using missing value imputation, min max normalization and sliding window segmentation. Then, this preprocessed data is passed through the 1D CNN to capture localized features. Lastly, these captured localized features are fed to the LSTM model which extracts long term temporal dependencies for prediction of future degradation states. Hence, the proposed 1D CNN-LSTM based DT framework achieved 95.6% prediction accuracy, 99 ms average feedback latency and 210 ms of end-to-end response time