Novel data-to-image method for heating ventilation and air conditioning fault detection and diagnosis in the built world

Tabular data obtained from the heating ventilation and air conditioning (HVAC) systems is not directly suitable for analysis with image-based deep learning models such as convolutional neural networks (CNNs), limiting the usage of advanced deep neural networks in the built world. To overcome this challenge, a novel feature association graph (FAG) generation method is proposed to convert HVAC tabular data into images automatically. In FAG, each feature is converted to a grid in the image, with the feature values conveyed through the grid’s grayscale. The proposed method fully considers the correlation between features, allowing the grouping of highly correlated features. This aids the CNN in extracting the feature interactions from the image. Besides, an image grid adaptation algorithm is proposed to resolve coordinate conflicts, ensuring that features are densely distributed in the two-dimensional grid. Three popular deep CNN-based fault detection and diagnosis (FDD) models were implemented and their performance was evaluated. To verify the generalizability of FAG, three realworld HVAC system FDD datasets were used. The results show that the proposed FAG, in combination with CNN models, achieves superior overall FDD performance compared to existing tabular data-to-image methods. The proposed attention residual neural network trained on FAG outperforms traditional machine learning models typically used for processing tabular data.