Skip to main content

Evaluation of the application of neural networks to distributed optical fibre sensors


Research objectives and content
The proposed project is a feasibility study into the application of neural networks to novel multi-point on distributed optical fibre sensors. Distributed optical fibre sensors produce complex signals at the point of signal detection. This complexity is cross modulation due to interference from parameters other than those being measured. A classical example of this is pressure changes affecting shifts in wavelength of the detected light in distributed optical fibre temperature sensors. Such effects can be minimised using wavelength referencing, but this is not always the optimum solution and can be expensive when used in conjunction with optical time domain reflectometry (OTDR). It is the intention of this project to investigate data from a previously developed sensor of relatively simple construction. Such an example is is one to detect the presence contamination of water supplies at multiple points in a system and requires the e application of OTDR techniques to detect the point on a long loop of fibre at which the contamination has occurred Effects other than fouling at a single point will then be modelled into the system e.g. change in coloration of the water (which itself is possibly a source of contamination), damage to the fibre e.g. microcracking, ageing effects of sources and detectors. It is envisaged that such extraneous effects will give rise to optical signals with characteristic temporal signatures and that the neural network may be trained to recognise these characteristic changes. The inclusion of neural network software requires relatively modest computing power and thus may easily be adapted to existing sensor systems which utilise PCs or dedicated computers or microcontrollers for the complex signal processing involved with OTDR. This study will provide valuable information for the identification of future research directions in this area including a full scale sensor implementation in an industrial or environmental application.
Training content (objective, benefit and expected impact)
The proposed project is designed to draw together expertise in the areas of advanced sensing techniques and neural network based pattern recognition. Both of the participating institutions will benefit from he collaboration with each gaining access to the extensive research laboratories and expertise in the case Of University of Limerick and the pattern recognition methods expertise applied to other sensing techniques in the case of the applicant. The proposed exploratory phase is anticipated to lead to further research programmes (PhDs) and ultimately to integrated sensor systems providing solutions to existing industrial and environmental problems.
Links with industry / industrial relevance (22)
The proposed project is a feasibility study of 3 months duration and although direct industrial linkage is not possible at such an early stage. it is likely (as outlined in the detailed proposal information) that the work will be of direct relevance to industry in terms of manufacturing and application to existing industrial sensing problems.

Funding Scheme

RGI - Research grants (individual fellowships)


Plassey Park, Castletroy
61 Limerick