Automated defect and contaminant inspection of HVAC duct

摘要

To sustain acceptable indoor air quality in a building, it is essential to frequently inspect and clean the Heating, Ventilation and Air-Conditioning (HVAC) ductwork. Nowadays the condition inspection is mostly conducted manually according to the video acquired by a pipeline robot. This situation has been significantly resulting in subjectivity, high-cost and inefficiency for HVAC ductwork cleaning and maintenance. In this paper an automatic defect and contaminant inspection system of HVAC duct is developed. The system consists of an infrared-CCD diagnosis device and a novel supervised method for duct inspection by cascading seeded k-means and C4.5 decision tree. The seeded k-means feature-clustering method first partitions the features of training instances into k clusters using Euclidean distance similarity. C4.5 decision tree is then used to refine the decision boundaries by learning the subgroups within the cluster. Finally the decisions of the k-means and C4.5 methods are combined to achieve the inspection results. To improve the classification performance on the minority classes as well as reduce the computation load during the process, Tabu search is employed for the feature selection and the cost-sensitive function is introduced into Tabu search. Experimental results on real-world data sets demonstrate that the proposed system is effective and efficient in inspecting the condition of HVAC ductwork.