A Physics-Informed IIoT Digital Twin Framework for Automotive Cooling Predictive Maintenance and Odometer Fraud Detection using LSTM-Attention

Abstract

        The shift to Industry 4.0 in the automotive industry requires an approach which moves away from passive monitoring and towards physics-based predictive maintenance. This paper introduces a Physics-Informed IIoT-Enabled Digital Twin (PI-DT) framework based on an Adaptive Physics-Informed Long Short-Term Memory with Attention (PI-LSTM-Attention) model to forecast the Remaining Useful Life (RUL) of Engine Cooling Systems (ECS). The proposed architecture integrates the thermodynamic Monotonicity constraints into the learning process using Physics-Informed, meeting the 100% target of Monotonicity compliance, 99.64% Failure-Point Prediction Accuracy, and a self-healing rate of 0.00%. The framework is a Dual-Layer Edge-Cloud Continuum for ultra-low latency and complete offline operational resilience. In a Zero Trust cybersecurity paradigm, the accumulated ledgers of thermal damage are continually compared with the digital reading on the odometers to identify tamper and warranty fraud. The proposed PI-LSTM-Attention network is compared with Transformer architectures with lower latency and higher prediction accuracy, which proves its unique optimization for the automotive Edge, especially for time-critical applications.