1208 IEEE SENSORS JOURNAL, VOL. 8, NO. 7, JULY 2008 A Method of Examination of Liquids by Neural Network Analysis of Reflectometric and Transmission Time Domain Data From Optical Capillaries and Fibers Michal Borecki, Michael L. Korwin-Pawlowski, and Maria Beblowska Abstract— This paper presents the construction and working principles of an intelligent fiber-optic sensor used for liquid ex- amination using time domain data. The sensing elements consisted of a length of optical fiber or a short section of optical capillary and worked either on the reflection intensity basis or on transmis- sion intensity basis. The changes of the monitored signal are caused mainly by the variation in light propagation conditions at the in- terfaces of liquid and gaseous phases and formation of drops of liquids or lenses at liquid-vapor interfaces. The physical effects on which depends the formation of a drop of liquid or a lens are sur- face tension, viscosity, boiling point, vapor pressure of liquid and its heat capacity. They provide information allowing determining the type of the liquid by a procedure which includes submerging, submersion, emerging and emergence of the sensing head from the examined liquid, or by local heating of the liquid sample. The mea- sured data were analyzed using neural networks. Index Terms—Analysis of liquids, fiber optic sensor, intensity sensor, neural network, optical capillary, optoelectronic test method, photonic measurements, sensor data analysis. I. MEASUREMENT SYSTEMS T HE NOVEL fiber optical heads of sensors of liquids pro- posed by us generate a specific task: complex data exami- nation. The artificial neural network can be used to classify data based on either a priori knowledge or on statistical information extracted from the patterns. The patterns to be classified are usu- ally groups of measurements or observations, defining points in an appropriate multidimensional space. The classification or description scheme implemented in sensor systems is usually based on the availability of a set of patterns that have already been classified or described. Up to today, there exist effective examples of neural network usage for processing data coming Manuscript received July 31, 2007; revised October 31, 2007; accepted November 10, 2007. Published July 16, 2008 (projected). This work was supported in part by a Discovery Grant of the Natural Sciences and Engineering Research Council of Canada, by a grant from the Canada Foundation for Innovation, and by research grants from the Polish Government. The associate editor coordinating the review of this paper and approving it for publication was Dr. Andrea Cusano. M. Borecki and M. Beblowska are with the Institute of Microelectronics and Optoelectronics, Warsaw University of Technology, 00-662 Warszawa, Poland (e-mail: borecki@imio.pw.edu.pl; beblowska@imio.pw.edu.pl). M. L. Korwin-Pawlowski is with the Centre de Recherche en Photonique, Université du Québec en Outaouais, Gatineau, QC, Canada J8X 3X7 (e-mail: michael.korwin-pawlowski@uqo.ca). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSEN.2008.926182 from fiber optic sensors heads. For example, the classification of spectra of the reflected signal coming from a FBG head [1] and the data interpreting from a three-head ethanol concentra- tion sensor system [2]. The commonality of sensors presented in [1] and [2] is that the processed data are connected with spectra of index of refraction and attenuation. In the proposed method, we examine the possibility classification of liquids by extending the data collected for the liquids with additional information of viscosity and surface tension, and reducing the data of the spectra information. The opto-electronic subsystem of our sensor consisted of a light source, a fiber optical link, a sensing head, and a photo- detection unit controlled by an intelligent detection and control system which included a computer with a data acquisition and control card (DAQ). An in-house developed software package was used for internal data exchange, control, and an artificial neural network (ANN) analysis [3]. The sensor heads for reflectometric measurements had the form of an end of optical fiber, or an end of optical fiber in- troduced into an optical capillary, or of optical fibers connected to the wall of an optical capillary [4]. The transmission mea- surements were done with fibers introduced into the hole of the optical capillary. We used mostly thin wall and small inner di- ameter optical capillary tubes made from fused silica with an index of refraction of 1.458 [5]. The capillaries had inner di- ameters from 50 to 2 mm and wall thickness from 20 to 1 mm. Such capillaries allow relatively simple sampling of small volumes of liquids. For hydrophilic liquids, sampling can be done without need of a pump, since the capillary can be filled by capillary forces. Such capillaries allow precise determination of the sample volume which can be also calculated using the sur- face tension values for a given liquid [6]. We used three detection schemes, two in a reflectance con- figuration and one in transmission. The optical signal changes which have to be induced to be the subject to time domain anal- ysis were achieved by changing the position of the head with a mini-lift device or by local heating of the liquid in the capil- lary [7]. The first experimental observation setup used a mini-lift in what is one of the forms of fiber drop analysis (FDA) [8]. It con- sists (Fig. 1) of an optical sensing head that is sequentially made to submerge, is in submersion, made to emerge and is in emer- gence from the examined medium [3]. In these experiments, we used a large core polymer optical fiber (POF) PFM-22E-750 1530-437X/$25.00 © 2008 IEEE