Domain-adapted Gaussian mixture models for population-based structural health monitoring

Abstract

Transfer learning, in the form of domain adaptation, seeks to overcome challenges associated with a lack of available health-state data for a structure, which severely limits the effectiveness of conventional machine learning approaches to structural health monitoring (SHM). These technologies utilise labelled information across a population of structures (and physics-based models), such that inferences are improved, either for the complete population, or for particular target structures — enabling a population-based view of SHM. The aim of these methods is to infer a mapping between each member of the population’s feature space (called a domain) in which a classifier trained on one member of the population will generalise to the remaining structures. This paper introduces the domain-adapted Gaussian mixture model (DA-GMM) for population-based SHM (PBSHM) scenarios. The DA-GMM, infers a linear mapping that transforms target data from one structure onto a Gaussian mixture model that has been inferred from source data (from another structure). The proposed model is solved via an expectation maximisation technique. The method is demonstrated on three case studies: an artificial dataset demonstrating the approach’s effectiveness when the target domain differs by two-dimensional rotations; a population of two numerical shear-building structures; and a heterogeneous population of two bridges, the Z24 and KW51 bridges. In each case study, the method is shown to provide informative results, outperforming other conventional forms of GMM (where no target labelled data are assumed available), and provide mappings that allow the effective exchange of labelled information from source to target datasets.

Measuring data similarity in population-based structural health monitoring using distance metrics

“…As a result, a given metric can give an indication of the type of features present in a signal without manual inspection. The results presented here using the simulated dataset are satisfactory pointers towards generalisation, because in a population-based setting -particularly within a homogeneous population where the behaviour of the data within a given window is assumed to be the same/very similar (Even in heterogenous population examples, feature similarity is important to avoid negative transfer 23,26 ) -the metrics will not identify differences between general trends as D(x, y) = 0 when x = y. As a result, the metrics are expected to identify behaviours/features that are anomalous to the general population trends.…”

“…Methods such as normal condition alignment (that standardises and aligns data when structures are under normal operating condition) has proven incredibly useful for PBSHM in transferring knowledge across structures. 22,23 Therefore, the behaviour of metrics as data are standardised is an important consideration for testing their effectiveness for PBSHM.…”

“…In the current paper, a different approach is considered where time-domain features are explored as, typically, the data are available in the time domain at the start of the analysis chain. 25 Usually, many studies in PBSHM literature acknowledge that similarity measurement is vital, and conducts it visually 23,26,27 or using a specific metric. 24,25 This paper attempts to find a principled method of assessing feature similarity in PBSHM that provides a guideline on which metrics are most suitable for a given commonly occurring feature found in SHM and PBSHM.…”

Vehicle Detection Using Machine Learning Model with the Gaussian Mixture Model (GMM)

“…f. Ultimately, the choice of motion tracking method depends on the specific requirements of the application and the trade-offs between accuracy, computational complexity, and robustness. According to statistics, GMM is one of the most used statistical models and is most commonly utilized when adaptive background subtraction in movies is needed [24]. When used in settings with relatively little background movement, this approach performs exceptionally well.…”

“…Figure 2: Proposed ModelGaussian Mixture Model (GMM) is one of the ways to get rid of the background. GMM models the time series of pixel values[23][24][25]. The accuracy of the Gaussian Mixture Models (GMM) method in separating the background and the tracked object can be evaluated using several metrics, such as: a. Intersection over Union (IoU): This metric measures the overlap between the predicted object bounding box and the ground truth bounding box, divided by their union.…”