A methodology on interpretable novelty detection

Artur Movsessian, David Garcia Cava, Dmitri Tcherniak, Rimas Janeliukstis

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

Vibration-based Structural Health Monitoring (VSHM) systems continuously gatherdata from an array of sensors mounted on a structure. Features are constructed from the datameasured. The aim is to monitor the vibration responses in the search for changes that mayhint to damage. The continuous data acquisition generates high-dimensional feature spacesthat require Data-Driven approaches to make inferences concerning the integrity of the struc-ture. In recent years, machine learning has played an increasingly important role in VSHM.Data-driven algorithms have been successfully used to construct models capable of detectinganomalies such as damage in the features derived from the vibration signals. Mahalanobis Dis-tance based novelty detection is a common used method to detect damage. Yet, the resultingmodels have been labelled “black box models” given that they lack interpretability. This be-comes a relevant challenge in the presence of high-dimensional feature spaces. Using machinelearning algorithms that can be interpreted would enable a more reliable novelty detection pro-cess, building trust in these methods and easing the decision-making process. Decision Trees(DT) is a widely used interpretable machine learning algorithm. The hierarchical structureof this algorithm enables the prioritisation of features that are used as predictors in the dam-age detection models. Furthermore, the nature of the algorithm enables the user to track thedecisions and understand the classification process in detail. In this paper, we introduce thecomplementary use of so-called “black box models” and DT for novelty detection. The pro-posed damage detection approach is tested on an experimental setup with a 14.3m wind turbineblade (WTB) equipped with 24 accelerometers. A pseudo-damage was simulated by addingmasses to several locations of the WTB. The pseudo-damage was detected by means of a semi-supervised novelty-detection. The novelties were later studied in detail with decision-trees tomake inferences on their potential causes.
Original languageEnglish
Title of host publicationInternational Conference on Structural Dynamics (EURODYN 2020)
Pages922-935
Number of pages14
DOIs
Publication statusPublished - 23 Nov 2020

Fingerprint

Dive into the research topics of 'A methodology on interpretable novelty detection'. Together they form a unique fingerprint.

Cite this