A conceptual digital twin framework for supply chain recovery and resilience

Amidst escalating global supply system risks and interruptions, the imperative for fortified supply networks is evident. Organizations striving for competitiveness and resilience must adeptly recognize, comprehend, and address disruptions. This study presents a three-phase digital supply chain twin framework, leveraging discrete event simulation and neural networks to anticipate floods—a typical natural catastrophe and disruptive event—and predict recovery indicators. This aids supply chain (SC) managers in making informed decisions. In the first phase, machine learning algorithms, including logistic regression and Long Short-Term Memory (LSTM), were trained on Kerala India's precipitation data to predict floods. LSTM outperforms logistic regression, achieving flood prediction with 73 % recall, 75 % accuracy, and 84 % Area Under Curve-Receiver Operating Characteristics score. In the second phase, simulations replicate value chain breakdowns. A process flow logic-driven discrete event simulation within a real-world SC network emulates operational disruptions. FlexSim is employed to model service-level failures, influencing SC model performance based on the distribution center service level. The third phase employs simulated case scenario data to train a multilayer neural perceptron network (MLPNN) for predicting production network recovery post-disruptions. The MLPNN monitors the mean squared error (MSE) and disruptive inputs throughout training and validation, revealing consistent MSE reduction over recovery periods. The number of epochs needed to achieve a minimum MSE is used as a recovery indicator to predict service restoration time. Consequently, this study introduces a conceptual digital twin framework for catastrophic operations chain breakdowns and recovery prediction. The framework's output assists SC planners in shaping robust strategies by foreseeing disruptions and facilitating recovery.