Neural Network Thermal Bridge Analysis for Building Structures

The patent applies preliminary action by pre-processing thermal analysis data using finite element method to create training datasets before actual use. The neural network is trained in advance with pre-processed building structure data, thermal conductivity information, and corresponding thermal characteristic results, so that during actual analysis, the pre-trained network can directly predict thermal characteristics without requiring repetitive finite element pre-processing and post-processing steps.

The patent replaces the complex mechanical finite element analysis system with an artificial neural network system. Instead of using traditional finite element method requiring pre-processing, solving, and post-processing operations, the patent trains a neural network to substitute these mechanical processing steps. The trained network directly predicts thermal characteristics from building structure data and thermal conductivity information, eliminating the need for complex repetitive finite element operations.

The patent changes the parameters of the measurement system by transitioning from simple thermal imaging camera measurements to a neural network-based prediction system. The neural network is trained with parameters including building structure data, thermal conductivity information, and thermal characteristic results obtained through finite element method. This parameter transformation enables the system to achieve high measurement precision while maintaining operational simplicity through the trained network's direct prediction capability.