Applications of artificial neural network on signal processing of optical fibre pH sensor based on bromophenol blue doped with sol–gel film Faiz Bukhari Mohd Suah a , Musa Ahmad a,* , Mohd Nasir Taib b a School of Chemical Sciences and Food Technology, Faculty of Science and Technology, National University of Malaysia (UKM), 43000 Bangi, Selangor D.E., Malaysia b Faculty of Electrical Engineering, MARA University of Technology (UITM), 40450 Shah Alam, Selangor D.E., Malaysia Abstract In this paper, the applications of artificial neural network (ANN) in signal processing of optical fibre pH sensor is presented. The pH sensor is developed based on the use of bromophenol blue (BPB) indicator immobilized in a sol–gel thin film as a sensing material. A three layer feed-forward network was used and the network training was performed using the back-propagation (BP) algorithm. Spectra generated from the pH sensor at several selected wavelengths are used as the input data for the ANN. The bromophenol blue indicator, which has a limited dynamic range of 3.00–5.50 pH units, was found to show higher pH dynamic range of 2.00–12.00 and with low calibration error after training with ANN. The enhanced ANN could be used to predict the new measurement spectra from unknown buffer solution with an average error of 0.06 pH units. Changes of ionic strength showed minor effect on the dynamic range of the sensor. The sensor also demonstrated good analytical performance with repeatability and reproducibility characters of the sensor yield relative standard deviation (R.S.D.) of 3.6 and 5.4%, respectively. Meanwhile the R.S.D. value for this photostability test is 2.4% and it demonstrated no hysteresis when the sensor was cycled from pH 2.00–12.00–2.00 (acid–base–acid region) of different pH. Performance tests demonstrated a response time of 15–150 s, depending on the pH and quantity of the immobilized indicator. # 2003 Elsevier Science B.V. All rights reserved. Keywords: Artificial neural network; Optical fibre pH sensor; Signal processing; Sol–gel; pH indicator; Bromophenol blue 1. Introduction There are several kinds of pH sensors such as electro- chemical sensor, salt dependent sensor and lately optical fibre sensor [1]. However, the major disadvantage of an optical fibre pH sensor is that they determine pH indirectly by measuring the colour of the dissociated and undissociated forms of the indicators and their response is sigmoidal [2]. Regardless of the sigmoidal response showed by these sensors, a narrow linear range of the curve can be taken as linear (often 2–4 pH units only), in order to determine the pH by interpolation method [3]. Numerous attempts have been proposed in order to extend the pH range of these sensors by employing for example, multiple pH indicators or one indicator with multiple steps of acid dissociation, fluorescent indicators and multiplexing several optical pH probes [4]. A number of signal processing techniques, for instance polynomial curve-fitting [5] has also been applied for modelling the sensor response. Over the last several years, the number of studies on application of artificial neural network (ANN) for solving modelling pro- blems in analytical chemistry and especially in optical fibre chemical sensor (OFCS) technology, has increased substan- tially. ANN is a computing system made up of a number of simple and highly interconnected processing elements, which processes information by its dynamic state response to external inputs [6]. It is composed of many simple processing elements that usually do little more than take a weighted sum of all their inputs. The range of scope of applications of ANN comes from their capability to estimate complex functions that make them compatible for modelling non-linear relationships. The range of chemical applications of ANN is very large and it includes fields as diverse as modelling structure of protein, molecular dynamics, process control, interpretation of spec- tra, calibration, pattern recognition, optimisation of the linear signal range and signal processing [7–9]. Meanwhile in OFCS technology, ANN is used in signal processing, data reduction and optimisation, interpretation and prediction of spectra and calibration [10]. Sensors and Actuators B 90 (2003) 182–188 * Corresponding author. Tel.: þ60-3-8921-5438; fax: þ60-3-8921-5410. E-mail address: andong@pkrisc.cc.ukm.my (M. Ahmad). 0925-4005/03/$ – see front matter # 2003 Elsevier Science B.V. All rights reserved. doi:10.1016/S0925-4005(03)00026-1