Particle Filter for Temperature Compensation

This Python package provides filter-based temperature compensation methods to remove the slow temperature trends from strain data with short-term events. An implementation of a particle filter is given, specifically designed for temperature compensation in various measurement scenarios.

Features

• Particle filter implementation for temperature compensation.
• Management of multiple particle filters with ParticleFilterBank.
• Support for SEP005 compliant measurements.
• Customizable particle filter settings including noise levels and thermal expansion coefficients.

Installation

Ensure you have Python 3.9 or more recent installed on your system. You can install this package using the following command:

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

Usage

• Example for using a single ParticleFilter for temperature compensation:

from filter_tc.particle_filter import ParticleFilter

# Create a particle filter
measurements = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]) # Replace with actual measurements
inputs = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]) # Replace with actual inputs (temperature data)

num_particles=1000

pf = ParticleFilter(
num_particles=num_particles,
r_measurement_noise=5e1,
q_process_noise=np.array([2, 1]),
scale=5e-3,
loc=-10)

# Run the particle filter
pf.filter(measurements, inputs, loading='tension')

• Example for using a ParticleFilterBank for managing multiple ParticleFilters:

from filter_tc.particle_filter import ParticleFilterBank

# Create a particle filter bank
measurements = [
        np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]), # Replace with actual measurements
        np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]) # Replace with actual measurements
]
inputs = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]) # Replace with actual inputs (temperature data)

num_particles=1000

pfb = ParticleFilterBank(
num_particles=num_particles,
r_measurement_noise=5e1,
q_process_noise=np.array([2, 1]),
scale=5e-3,
loc=-10)

# Run the particle filter
filtered = pfb.filter(measurements, inputs, loading='tension')

# Export the particle filter states to late use
pfb_states = pfb.export_states()

pfb2 = ParticleFilterBank().from_states(pfb_states)

Resulting filtering

Examples for applying the Particle filter for tension and compression loading measurements are given in the notebooks. The following figures show the concept of the Particle Filter and the resulting data after removing the PF output from the measurements to approximate events.

• Example of inputs, measurements, particle propagations (grey) and Particle Filter output:

  

• Example of the data after removing the PF output from the measurements to approximate events:

  

Credits

The implementation is based on the concepts from Kalman and Bayesian Filters in Python and adapted by Maximillian Weil for specialized use.

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/