The Atacama Large Millimeter/submillimeter Array (ALMA) observatory, with its 66 individual telescopes and other central equipment, generates a massive set of monitoring data every day, collecting information on the performance of a variety of critical and complex electrical, electronic and mechanical components. This data is crucial for most troubleshooting efforts performed by engineering teams. More than 5 years of accumulated data and expertise allow for a more systematic approach to fault detection and diagnosis. This paper presents model-based fault detection and diagnosis techniques to support corrective and predictive maintenance in a 24/7 minimum-downtime observatory.