labull/EngineeringPatternRecognition

Code to reproduce paper results (or as close as possible, depending on data-availability). Each publication has a Jupyter notebook. Mostly probabilistic/Bayesian ML for engineering applications, particularly performance and health monitoring.

EPR

Code to reproduce paper results (or as close as possible, depending on data-availability). Each publication has a Jupyter notebook.

Mostly probabilistic/Bayesian ML for engineering applications, particularly performance and health monitoring. Scripts are provided to test and demonstrate the EPR module.

Notebooks for papers

• Hierarchical Bayesian modelling for knowledge transfer across engineering fleets via multitask learning (CACAIE, 2022)

  • Hierarchical regression models of engineering populations, allowing knowledge transfer between subgroups.
  • Applications to truck fleet survival analysis and wind farm power prediction.
  • Jupyter notebook demo based on truck-fleet survival analysis.

• On the transfer of damage detectors between structures: an experimental case study (JSV, 2021)

  • Domain adaptation to transfer novelty detectors between aircraft tailplane ground-tests.
  • The TCA code used in the papers.

• Towards semi-supervised and probabilistic classification in structural health monitoring (MSSP, 2020)

  • Semi-supervised learning of GMMs via (MAP) expectation maximisation, applied to Gnat aircraft ground-test data to utilise both labelled and unlabelled data.
  • Jupyter notebook demo; MATLAB demo.

• Probabilistic active learning: An online framework for structural health monitoring (MSSP, 2019)

  • Uncertainty sampling (GMMs) to direct inspections from online data streams from engineering systems. Applied to bridge, lathe, and aircraft monitoring datasets.
  • Jupyter notebook demo; MATLAB demo.

• Active learning for semi-supervised structural health monitoring (JSV, 2018)

  • Hierarchical sampling for active learning (the DH active learner) applied to learn a classifier for ground-test vibration data from a Gnat aircraft.
  • MATLAB demo.

Algorithms

• Multitask Learning

  • Hierarchical regression (Stan)

• Domain Adaptation

  • Transfer Component Analysis (TCA)

• Partially-supervised learning

  • Active learning by uncertainty sampling in Gaussian Mixture Models (GMMs)
  • Semi-supervised learning of mixture models via (MAP) expectation maximisation
  • Hierarchical sampling for active learning (the DH active learner)

Figures

Multitask learning (MTL)

MTL for knowledge transfer between tasks

Compared to independent models

Active learning

Semi-supervised learning

(blue ellipse shows the prior)

TCA domain adaptation (transfer learning)

Archived MATLAB functions/scripts are available in the matlab folder.