Adaptive anomaly detection in time-series data

An anomaly detection system, comprising at least one processor and a memory storing a plurality of programming instructions that, when executed by the at least one processor, cause the at least one processor to:

• receive uniformly spaced digital data from a plurality of sensors monitoring one or more equipment components of a monitored system;
• receive physical model parameters associated with the monitored system;
• generate, based at least in part on the physical model parameters, a set of orthogonal kernel basis functions;
• correlate the uniformly spaced digital data with the set of orthogonal kernel basis functions to extract a plurality of feature components and to generate an energy map;
• detect a signature anomaly based on a deviation of the energy map from a baseline, the signature anomaly being associated with at least one of the equipment components;
• for the detected signature anomaly:
• when a number of orthogonal kernel basis functions in the set is less than a number of feature components associated with the detected signature anomaly, update the set of orthogonal kernel basis functions by adding one or more additional orthogonal kernel basis functions and reprocess the uniformly spaced digital data using the updated set;
• and
• when a width of at least one feature component associated with the detected signature anomaly is less than a width of at least one orthogonal kernel basis function in the set, modify the set of orthogonal kernel basis functions by narrowing the width of one or more orthogonal kernel basis functions and reprocess the uniformly spaced digital data using the modified set;
• and
• output, to a damage detector, data associated with the detected signature anomaly.