Optimal Sensor Placement for heat transfer digital twin with Denoising Autoencoder and discrete optimization

Heat transfer digital twins are essential for monitoring, simulating, and optimizing thermal performance in a wide range of engineering systems. However, their effectiveness is highly dependent on the quality of sensor data, the efficiency of sensor placement, and robustness against environmental noise. Traditional Optimal Sensor Placement (OSP) methods, such as PCA-based SVD-QRP and POD-DEIM, often underperform in the presence of nonlinear dynamics and high noise levels. To address these limitations, this study proposes an advanced OSP framework that integrates a Denoising Autoencoder (DAE) with discrete optimization. The DAE is utilized to reconstruct clean temperature fields from noisy sensor data, thereby improving placement reliability and noise resilience. The discrete optimization procedure, coupled with a projection operator, identifies cost-effective sensor configurations while minimizing reconstruction errors. The proposed approach is validated across diverse thermal scenarios – including a heat plate, a battery module, and a brushless DC motor (BLDC) – under varying levels of measurement noise. Comparative evaluations confirm the superiority of the proposed method in terms of reconstruction accuracy, robustness to noise, and sensor efficiency. The results highlight its capability to achieve high-fidelity simulation with significantly fewer sensors, demonstrating strong potential for scalable deployment in real-world digital twin applications. • Proposes a noise-robust sensor placement framework using a Denoising Autoencoder. • Employs projected gradient descent to select sensors under budget constraints. • Validated on heat plate, battery module, and BLDC motor thermal datasets. • Outperforms SVD-QRP and POD-DEIM in noisy and sensor-sparse environments. • Achieves high-fidelity reconstruction using significantly fewer sensors.