A. Perera

An intelligent detector based on temperature modulation of a gas sensor with a digital signal processor

Abstract An intelligent detector based on a hot-plate gas sensor and a digital signal processor (DSP) is presented. The work comprises sensor measurements and gas identification with a pattern recognition (PARC) system along with a systematic verification of both stages, thanks to clustering validity methods and performance tests. Commercial silicon micromachined tin-oxide sensors have been used to capture dynamic measurements modulating the sensor heater at different temperatures, waveforms and frequencies. Feature extraction is based on the spectral and transient analysis of the sensor output signals. The PARC systems are based on self-organizing maps (SOM) and recent variations of these well-known neural networks. The proposed hardware is in charge of the whole system: the sensor temperature modulation and signal processing.

On-line event detection by recursive Dynamic Principal Component Analysis and gas sensor arrays under drift conditions

Leakage detection is a common chemical-sensing application. Leakage detection by thresholds on a single sensor signal suffers from important drawbacks when sensors show drift effects or when they are affected by other long-term cross sensitivities. In this paper, we present an adaptive method based on a recursive dynamic principal component analysis (RDPCA) algorithm that models the relationships between the sensors in the array and their past history. In normal conditions, a certain variance distribution characterizes sensor signals, however, in the presence of a new source of variance the PCA decomposition changes drastically. In order to prevent the influence of sensor drift, the model is adaptive, and it is calculated in a recursive manner with minimum computational effort. The behavior of this technique is studied with synthetic and real signals arising by oil vapor leakages in an air compressor. Results clearly demonstrate the efficiency of the proposed methodLeakage detection is an important issue in many chemical sensing applications. Leakage detection by thresholds suffers from important drawbacks when sensors have serious drifts or they are affected by cross-sensitivities. Here we present an adaptive method based in a Dynamic Principal Component Analysis that models the relationships between the sensors in the array. In normal conditions a certain variance distribution characterizes sensor signals. However, in the presence of a new source of variance the PCA decomposition changes drastically. In order to prevent the influence of sensor drifts the model is adaptive and it is calculated in a recursive manner with minimum computational effort. The behavior of this technique is studied with synthetic signals and with real signals arising by oil vapor leakages in an air compressor. Results clearly demonstrate the efficiency of the proposed method.

Fuzzy inference system for sensor array calibration: prediction of CO and CH4 levels in variable humidity conditions

Abstract An optimized fuzzy inference system for carbon monoxide and methane concentration estimation is presented and compared to the three most common linear methods: PLS, PCR and MLR, and also to nonlinear extensions of PLS. The system optimization includes: rule pruning, membership function optimization by Solis–Wett algorithm, rule consequents optimization and sensor selection by sequential floating feature selection (SFFS) algorithm. An extensive data set obtained from a sensor array composed of five metal oxide gas sensors operated at two working temperatures in different humidity conditions is used for the method evaluation. Advantages and drawbacks of both linear methods and fuzzy systems are discussed and compared.

Finding the best calibration points for a gas sensor array with support vector regression

Electronic noses and gas alarm systems use chemical sensor arrays for the detection of gas mixtures. These sensing devices typically have a high degree of collinearity and nonlinear responses which makes their calibration difficult. Support vector regression was used to select a minimal number of calibration points for a dataset generated from laboratory measurements of a twelve element metal oxide sensor array exposed to ternary mixtures of CO, CH/sub 4/, and ethanol. The results indicate that the prediction accuracy of the model generated with kernel regression methods is better than that of partial least squares even when the number of calibration points is small.

Poisoning fault diagnosis in chemical gas sensor arrays using multivariate statistical signal processing and structured residuals generation

Chemical gas sensors are a cheaper and faster alternative for gas analysis than conventional analytic instruments. .However they are prone to degradation because of sensor poisoning and drift. Statistical methods like principal component analysis (PCA) and partial least squares (PLS) have been proved to be very useful in the task of fault diagnosis of malfunctioning sensors. In this work we test the effectiveness of several techniques based on PCA and PLS on faults caused by sensor poisoning These techniques will be evaluated on a dataset composed by the signals of 17 conductive polymers gas sensors measuring three analytes at several concentration levels. These techniques will be evaluated concerning their capabilities to detect the fault, identify the faulty sensor and correct their signal.

Potato creams recognition from electronic nose and tongue signals: feature extraction/selection and RBF neural networks classifiers

We describe the development of a potato cream recognition system based on radial basis function neural networks from electronic nose and electronic tongue signals. Exhaustive and systematic feature extraction and selection, which are needed because of high dimensionality of signals, are performed on both instruments using various feature selection algorithms. At the end, we design the classifier based on the RBF network, and compare the results obtained from different features.

Biologically inspired large scale chemical sensor arrays and embedded data processing

Biological olfaction outperforms chemical instrumentation in specificity, response time, detection limit, coding capacity, time stability, robustness, size, power consumption, and portability. This biological function provides outstanding performance due, to a large extent, to the unique architecture of the olfactory pathway, which combines a high degree of redundancy, an efficient combinatorial coding along with unmatched chemical information processing mechanisms. The last decade has witnessed important advances in the understanding of the computational primitives underlying the functioning of the olfactory system. EU Funded Project NEUROCHEM (Bio-ICT-FET- 216916) has developed novel computing paradigms and biologically motivated artefacts for chemical sensing taking inspiration from the biological olfactory pathway. To demonstrate this approach, a biomimetic demonstrator has been built featuring a large scale sensor array (65K elements) in conducting polymer technology mimicking the olfactory receptor neuron layer, and abstracted biomimetic algorithms have been implemented in an embedded system that interfaces the chemical sensors. The embedded system integrates computational models of the main anatomic building blocks in the olfactory pathway: the olfactory bulb, and olfactory cortex in vertebrates (alternatively, antennal lobe and mushroom bodies in the insect). For implementation in the embedded processor an abstraction phase has been carried out in which their processing capabilities are captured by algorithmic solutions. Finally, the algorithmic models are tested with an odour robot with navigation capabilities in mixed chemical plumes

Improving drift correction by double projection preprocessing in gas sensor arrays

It is well known that gas chemical sensors are strongly affected by drift. Drift consist on changes in sensors responses along the time, which make that initial statistical models for gas or odor recognition become useless after a period of time of about weeks. Gas sensor arrays based instruments periodically need calibrations that are expensive and laborious. Many different statistical methods have been proposed to extend time between recalibrations. In this work, a simple preprocessing technique based on a double projection is proposed as a prior step to a posterior drift correction algorithm (in this particular case, Direct Orthogonal Signal Correction). This method highly improves the time stability of data in relation with the one obtained by using only such drift correction method. The performance of this technique will be evaluated on a dataset composed by measurements of three analytes by a polymer sensor array along ten months.