Designing a framework for a data-driven digital twin model to predict remaining useful life

Purpose Digital twin (DT) is a technology used in Industry 4.0 to enhance the efficiency of the industry. Therefore, data-driven DT should be implemented to determine the remaining useful life (RUL) of the component/system to prevent sudden breakdowns. Design/methodology/approach This study uses a secondary dataset to develop a data-driven DT model to predict the RUL of mechanical systems, such as turbo engines. Principal component analysis technique was used to reduce the features of the given dataset. Additionally, different machine learning algorithms such as Random Forest (RF), Convolutional Neural Network (CNN), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting and Long Short-Term Memory (LSTM) were used to develop an accurate data-driven DT model. A conceptual framework was discussed to develop the data-driven DT model for predicting RUL. Findings This study shows that LSTM is the most suitable model for predicting turbo engine RUL on the secondary dataset, achieving the lowest root mean squared error. Additionally, the receiver operating characteristic curve shows 98% accuracy for LSTM. Whereas AdaBoost, RF and CNN are not very accurate for the given dataset. Practical implications This study helps industry professionals to understand the role of a data-driven DT model in determining the RUL of mechanical components/systems to reduce downtime. It is helpful to implement a data-driven DT model for RUL prediction within a conceptual framework. Originality/value This study used a secondary dataset to develop a data-driven DT model for RUL prediction, accompanied by a conceptual framework for such a model.