NOVELTY DETECTION IN A CHANGING ENVIRONMENT: REGRESSION AND INTERPOLATION APPROACHES

The technique of novelty detection is now established as a means of performing the lowest level of damage identification. Data are accumulated while the system or structure is operating in normal condition and used to construct a reference model. During subsequent operation of the system, data are compared to the reference and any significant deviations are taken to indicate damage. This approach has potential problems if the system or structure is embedded in a changing environment. If the reference data are only characteristic of a limited range of the environmental parameters, measurements from the system in an undamaged condition but from a different environmental state, may cause the diagnostic to register novelty and thus falsely infer damage. This paper demonstrates a potential solution to the problem via the construction of a reference set parametrized by an environmental variable. Two approaches are considered: regression and interpolation.