Temporal Extrapolation Generalization of Proper Orthogonal Decomposition (POD) and Radial Basis Function (RBF)...

Area of Science:

• Computational science and engineering
• Thermal analysis
• Machine learning applications

Background:

• Accurate transient temperature prediction is crucial for effective thermal management in electronic devices.
• Existing methods can be computationally expensive, limiting real-time applications.
• Electronic packages often involve complex thermal interactions from multiple heat sources.

Purpose of the Study:

• To develop an efficient and accurate reduced-order surrogate modeling approach for transient temperature fields.
• To enable robust temporal extrapolation beyond the training data horizon.
• To reduce the computational cost associated with thermal analysis in electronic systems.

Main Methods:

• Utilized proper orthogonal decomposition (POD) for dimensionality reduction.
• Employed radial basis function (RBF) neural networks for surrogate modeling.
• Mapped time-conditioned modal coefficients in a parameter-time space.

Main Results:

• The proposed POD-RBF surrogate model achieved high predictive accuracy with a global MRE below 3%.
• Demonstrated significant reductions in computational cost compared to traditional methods.
• Validated the approach using a multi-heat-source conduction model typical of electronic packages.

Conclusions:

• The POD-RBF surrogate modeling approach offers an efficient and accurate solution for transient temperature prediction.
• The method shows strong potential for real-time thermal monitoring and management in electronic systems.
• Robust temporal extrapolation capabilities enhance its applicability for dynamic thermal scenarios.