OMA Clustering

Python package to cluster the outputs of an Operational Modal Analysis. The clusters serve as a basis for automatically setting the configuration of further mode tracking. The clustering is done using the DBSCAN or HDBSCAN algorithm. The hyperparameters of the clustering algorithm are adapted based on physical knowledge of the structure and interprator supervision as illustrated in the figure below.

• Flowchart illustrating the automatic setting of the configuration using the clustering results:

Features

• ModeClusterer_DBSCAN
• ModeClusterer_HDBSCAN

Installation

Before using ModeClusterer, ensure you have the oma_clustering package installed. If not, you can install it using pip:

$ pip install git+https://github.com/WEILMAX/oma_clustering.git

Usage

• Importing the class

from oma_clustering.oma_clustering_dbscan import ModeClusterer_DBSCAN
from oma_clustering.oma_clustering_hdbscan import ModeClusterer_HDBSCAN

• Initializing the class with set of hyperparameters

  • The hyperparameters are used to cluster the dataframe.
  • The multipliers are used to scale the columns of the dataframe before clustering.
  • Ensure the multipliers have the same keys as the columns of the dataframe.
  • The index_divider is used to divide the index of the dataframe to get the index of the dataframe used for clustering.

clusterer_DBSCAN = ModeClusterer_DBSCAN(
        eps = 5,
        min_samples = 100,
        multipliers = {"frequency": 35, "size": 0.5, "damping": 1},
        index_divider = 20000,
        cols = ['frequency', 'size', 'damping'],
        min_size = 5.0,
        max_damping = 5.0
)

clusterer_HDBSCAN = ModeClusterer_HDBSCAN(
        min_cluster_size = 100,
        min_samples = 100,
        multipliers = {"frequency": 500, "size": 0.5, "damping": 1},
        index_divider = 20000,
        cols = ['frequency', 'size', 'damping'],
        min_size = 5.0,
        max_damping = 5.0
)

• Data description

  • The dataframe should contain the following columns: frequency, size, damping, index.
  • The index should be the index of the dataframe used for clustering.
  • The dataframe should contain the modes of the structure.

• Fitting the class with a dataframe

  Fit the model using your OMA result data:

clusterer_DBSCAN.fit(oma_result)
clusterer_HDBSCAN.fit(oma_result)

• Getting the clusters

  Predict clusters and filter out the noise:

clustered_modes_DBSCAN = clusterer_DBSCAN.predict(min_cluster_size=200)
clustered_modes_DBSCAN = clustered_modes_DBSCAN[clustered_modes_DBSCAN.labels != -1]

clustered_modes_HDBSCAN = clusterer_HDBSCAN.predict(min_cluster_size=200)
clustered_modes_HDBSCAN = clustered_modes_HDBSCAN[clustered_modes_HDBSCAN.labels != -1]

Example

• Here is a complete example of using ModeClusterer_DBSCAN:

from oma_clustering.oma_clustering_dbscan import ModeClusterer_DBSCAN

clusterer_DBSCAN = ModeClusterer_DBSCAN(
        eps = 5,
        min_samples = 100,
        multipliers = {"frequency": 35, "size": 0.5, "damping": 1},
        index_divider = 20000,
        cols = ['frequency', 'size', 'damping'],
        min_size = 5.0,
        max_damping = 5.0
)

clusterer_DBSCAN.fit(oma_result)

clustered_modes_DBSCAN = clusterer_DBSCAN.predict(min_cluster_size=200)
clustered_modes_DBSCAN = clustered_modes_DBSCAN[clustered_modes_DBSCAN.labels != -1]

• Here is a complete example of using ModeClusterer_HDBSCAN:

from oma_clustering.oma_clustering_hdbscan import ModeClusterer_HDBSCAN

clusterer_HDBSCAN = ModeClusterer_HDBSCAN(
        min_cluster_size = 100,
        min_samples = 100,
        multipliers = {"frequency": 500, "size": 0.5, "damping": 1},
        index_divider = 20000,
        cols = ['frequency', 'size', 'damping'],
        min_size = 5.0,
        max_damping = 5.0
)

clusterer_HDBSCAN.fit(oma_result)

clustered_modes_HDBSCAN = clusterer_HDBSCAN.predict(min_cluster_size=200)
clustered_modes_HDBSCAN = clustered_modes_HDBSCAN[clustered_modes_HDBSCAN.labels != -1]

Resulting clustering

• Example of a resulting clustering using ModeClusterer_DBSCAN:

• Example of a resulting clustering using ModeClusterer_HDBSCAN:

Credits

This package was created in the context of the NWE Interreg 'Smart Circular Bridge' project. For more information, see https://vb.nweurope.eu/projects/project-search/smart-circular-bridge-scb-for-pedestrians-and-cyclists-in-a-circular-built-environment/