Research interests and Projects

Research Interests

Artificial Intelligence and Machine Learning:

  • Artificial Neural Networks and Deep Learning (e.g. Feed-forward networks, Deep neural networks, Recurrent networks)
  • Bayesian modelling (Bayesian single level/hierarchical models)
  • Committee Machines (e.g. Ensemble Learning, Mixture of Experts)
  • Clustering methods (e.g. K-means clustering, Gaussian Mixture Models and Self-Organising Maps)
  • Dimensionally Reduction (e.g. Principal Component Analysis, Kernel PCA)
  • Fuzzy Logic and Adaptive Neuro-Fuzzy Inference Systems (ANFIS)
  • Inference approximations (e.g. Expectation Maximisation, Variational Bayesian)
  • Information theory (e.g. Mutual Information)
  • Kernel machines (e.g. Support Vector Machines, Relevance Vector Machines, Gaussian Processes)
  • Recursive Bayesian estimations (e.g. Kalman Filter, Extended Kalman Filter, Particle Filter)
  • Sampling methods (e.g. Monte Carlo, Importance Sampling, Gibbs sampler)

Health Monitoring Technologies:

  • Feature Extractions
  • Diagnostics
  • Prognostics

Modelling and Control Design:

  • Intelligent Control Systems
  • Systems Identification
  • Modelling of Complex Systems

Applications:

  • Bearing condition monitoring (diagnostics & prognostics)
  • Bio-medical robotics (modelling & control)
  • Machine learning for Applied Physics:
    • Friction process at nanoscale (modelling)
    • Atmosphere science (data analysis and modelling)
  • Gas turbine engines health monitoring (diagnostics & prognostics)
  • Lithium-ion batteries condition monitoring (diagnostics & prognostics)
  • Robot manipulator (modelling & control)
  • Twin rotor helicopter (modelling & control)

Research Projects

Institute for Atmospheric and Earth system Research (INAR), Helsinki University, Finland

  • Role: Postdoctoral Researcher
  • Project 1: Developing machine learning classifier for automatic determination of new particle formation days
    • Bayesian neural network classifier based on time and frequency domain features as well as multi-modal log-Gaussian distribution.
    • Convolutional neural network for images classification of new particle formation days.
  • Project 2: Developing non-linear correlation analysis on atmospheric variables
    • Developing method based on nearest neighbour mutual information to understand how aerosol particles, gases concentration, meteorological and radiation variables affect new particle formation.
  • Project 3: Developing machine learning based emulators for approximating complex meteorological models
    • Gaussian process emulator to speed up meteorological models
  • Project 4: Intelligent Health Monitoring on engineering systems diagnostics and prognostics
    • Battery is a device consisting of one or more electrochemical cells with external connections provided to power a large variety of electrical devices such as smartphones, robots, electric cars and scientific instruments.
    • Developing Batteries Prognostics using Variational Bayesian Hierarchical Models with Adaptive Priors
    • Developing Bearing diagnostics using Fault-classifier based on Deep Neural-Networks
    • The health data are extracted from bearing and batteries benchmark datasets

Aalto Science Institute (AScI) & Centre of Excellence in Computational Nano-physics (COMP), Aalto University, Finland

  • Role: Postdoctoral Research Fellow
  • Project 1: Developing advanced Machine Learning strategies for material sciences
    • Modelling friction process at the nano-scale using a committee machine of Mixture Bayesian Neural Networks.
    • High Performance Computing (HPC) implementation to parallelise the construction of committee machines based algorithms.

SUPPORTED AND COLLABORATE WITH:

Aalto Science Institute (AScI)

Centre of Excellence in

Computational Nanoscience (COMP)

  • Project 2: Machine Learning for Atmosphere Science
    • Big database is extracted from Smart SMEAR (Stations for Measuring the forest Ecosystem-Atmosphere Relationships), where large database has been gathered through real-time sensor measurements since 1996 to date.
    • Developing tool to visualise particle concentrations and model their distributions.
    • Developing Machine Learning strategies to automatise the judgement of event / non-event days (the days with/without aerosol formation).
    • Developing Clustering and Classification based Machine Learning methods to understand the way of aerosol formed in the atmosphere.

SUPPORTED AND COLLABORATE WITH:

Aalto Science Institute (AScI)

Centre of Excellence in

Computational Nanoscience (COMP)

Centre of Excellence in Atmospheric Science

Center for Advanced Life Cycle Engineering (CALCE), University of Maryland, USA

  • Role: Postdoctoral Research Associate
  • Developing a deterministic Bayesian approach for Lithium-ion batteries prognostics
  • Assisting in PHM software development.

SUPPORTED AND COLLABORATE WITH:

Rolls-Royce University Technology Center for systems & control, The University of Sheffield, UK

  • Role: Doctoral researcher
  • Developing solutions for aerospace gas turbine engine prognostics using data-driven approaches.
  • Developing deterministic Bayesian prognostic technique for capturing degradation uncertainty in complex system prognostics (the graphical models are shown on the first two left hand side Figures below).
  • Developing Bayesian hierarchical model based on sampling (Gibbs sampler) and approximate inference (variational Bayesian) methods to accommodate, optimally, multiple asset data for enhancing RUL estimation of assets (the graphical models are shown on the first two right hand side Figures below).
  • Developing an integrated prognostic approach combining Bayesian hierarchical model and information theoretic change-point detection algorithm to deal with irregular events occurring during the life cycle of an asset, e.g. fault & maintenance events.

SUPPORTED AND COLLABORATE WITH:

Dept. of Automatic Control & Systems Engineering, The University of Sheffield, UK

  • Role: Research Assistant
  • Developing systems modelling and control techniques using an online learning based Adaptive Neuro-Fuzzy Systems (ANFIS) for Twin Rotor MIMO System (TRMS).

Dept. of Automatic Control & Systems Engineering, The University of Sheffield, UK

  • Role: MSc student
  • Developing adaptive control strategy based on artificial neural networks (ANN) on Functional Electrical Stimulation (FES) to cope with muscle fatigue phenomena on paraplegic rehabilitation.

Dept. of Electrical & Electronics Engineering, Trisakti University, Indonesia

  • Role: BEng student
  • Developing electronic design and microcontroller programming on a manipulator robot for griping and moving objects.

Experimental Rigs

My research fields are mainly in artificial intelligence, machine learning, modelling of complex systems and control theory which maybe very "abstract" to "common people" because these research foundations are based on advanced mathematics and statistics as well as computational based research. They are mostly implemented through software, such as C/C++, Python and MATLAB/Simulink, embedded into computing platforms. However, their real applications are still visible (via hardware) in the form of experimental rigs. Here some of experimental test rigs which I have worked related to my research and teaching activities.

  • Transient response analysis (e.g. RLC circuit)
  • Mechanical & electro-mechanical control systems (e.g. Modular Servo System, Digital Pendulum, Magnetic Levitation System)
  • Process control (e.g. Level control and Flow control systems)
  • Intelligent control systems on twin rotor dynamics (e.g. Twin Rotor Multi-Input-Multi-Output System)
  • Intelligent control systems of biomedical-robotics - humanoid orthotics (e.g. Functional electrical stimulation FES)
  • Performance & Life Testing (e.g. Arbin BT2000 Battery test system)

Some examples of used experimental test beds:

The top pictures (from the left to the right) show Twin rotor MIMO system (feedback instrument), magnetic levitation (feedback instrument), level control (GUNT Hamburg) and flow control (GUNT Hamburg). The bottom picture shows experimental setup of Modular Servo System (feedback instrument), inverted pendulum (feedback instrument) and Functional electrical stimulation (FES) experiment facilities.

For low cost computations (such as simple control algorithms and sensing), I used to program in assembly language for microcontroller in the past (such as Atmel AT89 series). However, since the advancement of community based hardware, I have used Arduino-based microcontroller (such as Atmel ATmega328P, also known as Arduino/Genuino UNO).

I am also trying to familiarise myself with a Raspberry Pi (which may accommodate more expensive computations). Recently, MATLAB based Simulink Coder™ (formerly Real-Time Workshop®) is able to generate automatically C/C++ codes to handle more complex algorithm implementations.

Recently, I have also started working on Python and R languages for machine learning implementations. I have just started depositing some of my codes into my GitHub.