John M. Andrews

Neural network approach to qualitative identification of fuels and oils from laser induced fluorescence spectra

Abstract A series of software implemented, three-layer, backpropagation neural networks is trained to identify seven classes of petroleum hydrocarbon based fuels and oils from their fluorescence emission spectra. The classes include: 87 and 92 octane unleaded gasoline, diesel fuel No. 2, diesel fuel marine, JP5, and lube oils 2190 and 9250. The nitrogen laser induced fluorescence spectra of multiple samples of each class are collected through an optical fiber incorporated into a multichannel detection system. Thirty-six spectra examples of the seven fuel/oil classes are partitioned into seven separate paired combinations of training and test spectra. Each combination uses 29 spectra to independently train a network. The spectra from the remaining seven samples are used to test the trained networks ability to make generalized classifications. Networks trained with data sampled directly from the normalized fluorescence emission spectra accurately identify 96% of the test spectra. An additional series of networks is trained and tested using the same spectra but with principal component analysis (PCA) employed as a preprocessor. Networks trained using PCA processed spectral data achieve somewhat lower performance, successfully identifying only 90% of the test spectra.

Rapid in situ determination of total oil concentration in water using ultraviolet fluorescence and light scattering coupled with artificial neural networks

Real-time measurement of total oil concentration in complex samples is required in wastewater discharge streams from ships and processing industries. A novel technology has been developed for the accurate quantification of a variety of single oils and their mixtures. Four major types of oils (lube oils 2190 and 9250, diesel fuel marine (DFM), and jet fuel (JP5)), each of which consisted of a dozen subtypes of oil samples, were examined to obtain both fluorescence and light scattering spectra as a function of concentration of single oils and mixtures. Tremendous variations in both fluorescence and scattering were observed among oil types, subtypes, and mixtures. The spectral response of an oil mixture was not the simple summation of respective single oils. To account for all these variations, a multivariate, nonlinear calibration method is applied to associate instrumental responses with oil concentrations using artificial neural networks (ANNs). The neural network architecture has been established by optimizing network parameters such as epochs, the number of neurons in the hidden layer, and learning rates in order to achieve the maximum accuracy of oil concentration measurements. It is demonstrated that the simultaneous, combined use of fluorescence and light scattering significantly improves the accuracy of measurement for oil samples. The newly developed technique permits the reliable, real-time determination of the total concentration of various oils and mixtures in water.

Multispectral fluorometric sensor for real time in-situ detection of marine petroleum spills

This paper describes the development of a fluorescence based in-situ sensor system for real time monitoring and detection of petroleum hydrocarbon contaminants in the marine environment. The system consists of an array of underwater sensors deployed just below the water surface. The sensors can detect floating product (surface sheen) from below the surface as well as detect emulsified or dissolved phase petroleum in the water column. Data from each of the sensors is transmitted to a central base station computer for display, logging, and analysis. The primary intended use of the system is to protect marine facilities from accidental petroleum discharges by providing responding authorities with immediate notification of the occurrence of a leak or spill. The detection of petroleum is based upon the fluorescence of polycyclic aromatic hydrocarbons found within petroleum derived products. The sensors utilize broadband ultraviolet excitation from a pulsed xenon lamp to generate fluorescence in contaminated sea water. The intensity of the resulting fluorescence emission is proportional to both the oil concentration in water, and/or the oil film thickness on the water surface. Multispectral fluorescence emission information is used to distinguish between several possible petroleum classes and eliminate false positive interference from non-petroleum based fluorophores such as chlorophyll. Real time qualitative identification yields an important advantage in terms of rapidly resolving questions of spill origin or in determining an appropriate response. To enable long term underwater deployment, the optical energy of the ultraviolet excitation source also serves to prevent the occurrence of biofouling on the surface of the optical Window. The results of initial testing in San Diego Harbor and at the Ohmsett wave tank facility in New Jersey demonstrate the systems ability to detect petroleum products under a variety of conditions, including the presence of strong harbor chop.

Fiber optic fluorometer based on a dual-wavelength laser excitation source

A remote fiber optic fluorometer system which incorporates a dual wavelength UV laser excitation source is described. The system provides increased specificity for detection of multiple fluorophores without sacrificing real time sensing capability. Limitations imposed by UV transmission in fused silica fibers are discussed.

Use of a neural network for the analysis of fluorescence spectra from mixtures of polycyclic aromatic hydrocarbons

The use of a software implemented backpropagation neural network is reported for the qualitative and quantitative analysis of the fluorescence emission spectra from multicomponent mixtures of polycyclic aromatic hydrocarbons (PAHs) in solution. Analysis of two types of data is described. First, a backpropagation network is developed to determine the component concentrations in a ternary mixture of PAHs. The input data provided to the network consists of sampled two-dimensional (intensity vs. emission wavelength) fluorescence spectra. A second backpropagation network is investigated for the analysis of three-dimensional time resolved fluorescence emission spectra for a binary PAH mixture. Both of the networks are trained to recognize preselected compounds. Each trained network is then used to evaluate unknown emission spectra and to determine the presence and relative concentration of the compounds it has learned to recognize. Results from analysis of two-dimensional emission spectra show that the trained network was able to successfully identify the individual components and their concentrations in solutions containing mixtures of anthracene, chrysene, and acenapthene. Analysis of three-dimensional time resolved fluorescence emission data showed that individual components could be resolved in mixtures of two spectrally similar components (anthracene and chrysene). Although a network could also be trained to recognize anthracene and chrysene in binary mixtures using their two-dimensional emission spectra, use of three-dimensional time decay spectra reduced the learning time required to train the network by a factor of three.