APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS:

  • Sensors in Chemical Engineering
  • Financial Data Modelling
  • Medical Data Classification
  • Civil Engineering Structures Modelling
  • Prosthetic Hand Control
  • Autonomous Wheelchair Trajectory Control
  • Applications of ANN are mainly driven by requests from inside or outside the University.
  • Applications comprise a research component in the sense that every problem needs its specialized solution, usually not available off the shelf. Sometimes they reveal fundamental scientific problems that need to be addressed.
  • They also provide useful teaching material.

    • Sensors in Chemical Engineering.

    This was work done in the Swiss Federal Institute of Technology (EPFL) with Prof. Urs von Stockar and Dr. Jo Lister, and with financial support of NESTLE SA.
    The problem was to relate values produced by ultrasound sensors with actual physical characteristics of air bubbles in fermenter.
    This was a mapping problem and an MLP was used.
    As there were very few calibrated data available for the net training, a simulation of the physical system was developed. The data produced by the simulation were used as training set. Subsequent comparisons between measurements done with the ultrasound / ANN system and measurement done on images of the bubbles showed an excellent fit.
    This work also revealed that MLP nets are not robust against the loss of neurons, which led to the development of a new training method.

    Publications:
    "Characterizing bubbles in bioreactors by ultrasound" Bugmann, G. and vonStockar, U. (1989) Trends In Biotechnology (TINS) 7, 166-169 "Characterizing bubbles in bioreactors using light or ultrasound probes: Data analysis by classical means and by neural networks" Bugmann, G., Lister, J.B. and vonStockar, U. (1991) Canadian Journal of Chemical Engineering, 69, 474-480 "Direct approaches to improving the robustness of multilayer neural networks" Bugmann, G., Sojka, P., Reiss, M.,Plumbley, M. and Taylor, J.G. (1992) In: Alexander, I. and Taylor, J.G. (ed) "Artificial Neural Networks II", Elsevier, Proc. of ICANN'92, Brighton, UK, p. 1063-1066.

    Top of the Page

    Financial Data Modelling:

    This work was done with Dr. Sue Farrar and Dr. Jon Tucker of the Business School or the University of Plymouth. The problem was to predict if a company would raise funds by issuing shares or making debts. The data comprised 18 parameters describing the financial profiles and financial decisions of several hundred companies over several years.
    This was a difficult classification problem and a number of techniques were tried, such as MLP, linear regression and NRBF.
    All techniques found that, overall, the data are consistent with accepted economical models. However, they also showed that a majority of companies would not base their decisions on such economical models.
    On the financial side, further research should investigate the true basis for financial decisions. On the neural network side, this work has drawn our attention to the problem of inconsistent data and has led us to develop tools for assessing the quality of data.

    Publications:
    "Modelling the Marginal Capital Structure Decision using Logistic Regression and Neural Networks"
    Tucker J., Farrar S. and Bugmann G. (1997)
    Abstract book of British Accounting Association National Conference, University of Birmingham, UK, 24-26 March 1997

    Top of the Page

    Medical Data Classification

    This work is ongoing with Dr. Kate Burn-Thornton and Simon Thorpe of the Database Research Group of the School of Computing of the University of Plymouth. There are no published results yet.

    Top of the Page

    Civil Engineering Structures Modelling

    This work is done with Dr Yaqub Rafiq and Dave Easterbrook in the Department of Civil Engineering of the UoP.
    The problem is to produce neural network models of optimal dimensions of structures such as beams, slabs, etc. The aim is to reduce computation time in GA based design support tools. This is ongoing work.

    Publications:
    "Artificial Neural Networks for Modelling some of the Activities of the Conceptual Stages of the Design Process" Rafiq, Y., Bugmann, G. and Easterbrook, D.J. (1998), In Wang K.C.P (ed.) "Computing in Civil Engineering: Proceedings of International Computing Congress, Boston, Massachusetts, October 1998", pp. 631-643. Published by the American Society of Civil Engineers (ASCE), 1801 Alexander Bell Drive, Reston, Virginia 20191-4400, USA. ISBN 0-7844-0381-3

    Top of the Page

    Prosthetic Hand Control

    This work was done with Paul Robinson of the School of Electrical and Electronic Engineering of the UoP and Peter Nurse, Roland Burns and Ralf Richter from the School of Mechanical Engineering of the UoP.
    The problem was to control a prosthetic hand by using EMG signals produced by remaining forearm muscles.
    A MLP was used to classify 1 second of signal into one of three possible hand postures. The net proved to be remarkably robust and could generalize across users.
    Currently funding is sought for further work aiming at i) reducing the reaction time of the system by analysing shorter portions of the signal and ii) providing an analog output enabling continuous control of hand postures.

    Publications:
    "Single Site Myoelectric Control of a Complex Robot Hand" Robinson P., Nurse P., Roberts S., Richer R., Bugmann G. and Burns R. (1997) Proc. International Workshop on Advanced Robotics and Intelligent Machines, University of Salford, Manchester, UK, 25-26 March 1997, ISSN 1363-2698, paper no. 8.

    Top of the Page

    Autonomous Wheelchair Trajectory Encoding

    This work was done at the request of the artist Donald Rodney, and with the support of Kheng Lee Koay, Mike Philips and Nigel Barlow of the School of Computing of the SOC.
    The problem was to modify an electric wheelchair so that it would perform safely a predefined sequence of movements within a limited environment accessible to spectators. The whole project was realised in two month and is described in more details in a     dedicated web page.
    For encoding the trajectory, an new type of neural network was designed.
    Currently funding is sought i) to provide the wheelchair with a Natural Language user interface and ii) to develop a vision system for outdoor navigation and indoor navigation.

    Publications:
    "Stable Encoding of Robot Trajectories using Normalised Radial Basis Functions: Application to an Autonomous Wheelchair" (83,209) Bugmann G., Koay K.L., Barlow N., Phillips M. and Rodney D. (1998) Proc. 29th Intl. Symp. Robotics (ISR'98), 27-30 April, Birmingham, UK, pp. 232-235. (ISBN 0 9524454 7 6)

    Top of the Page

    Back to Home Page
    Guido Bugmann, December 1998 email