With automated fault detection and diagnostics tools poised to address many of the barriers to good commissioning and maintenance practice in building energy systems, techniques from the artificial intelligence and machine learning domains are emerging as viable approaches where rules-based techniques can be less suitable. This paper describes a novel dynamic, machine learning-based technique for detecting faults in commercial air handling units. Preliminary results showing the performance of the proposed technique based on real world fault data obtained from a large scale building laboratory facility are presented, with discussion on current research and future research direction also provided.