![]() |
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ЦЭМИ РАН |
||
Scientific interest in carbon dots (CD) as potential nanosensors is high due to the simplicity and cheapness of their synthesis, stable fluorescent properties, and low toxicity. A lot of research efforts are channeled to the development of CD-based optical sensors for various substances [1-3]. The main obstacle to their creation is the lack of analytical models describing the spectral curves of CD fluorescence. In such circumstances, the development of a multimodal nanosensor (i.e., a sensor capable of simultaneously measuring several parameters) is significantly complicated by the nonlinear nature of the CD fluorescence dependence on several parameters of the medium. In this paper, it is proposed to solve this problem using artificial neural networks (ANN) since they do not require the construction of any analytical model of the research object, but only the collection of data reflecting the dependence of CD fluorescence on the salt composition of the medium. Thus, the research is devoted to the development of a fluorescent CD-based nanosensor for recognizing and determining the concentration of heavy metal ions in liquid multicomponent media using artificial neural networks. A database consisting of 1000 fluorescence spectra of CD aqueous suspensions in the presence of Cu(NO3)2, Ni(NO3)2, Cr(NO3)3 salts, their concentrations ranging from 0 to 4.95 mM, was obtained experimentally. The fluorescence signal of the samples was excited by radiation at 41 wavelengths. The spectra were processed using fully connected neural networks (perceptrons) and convolutional neural networks. The application of a neural network approach for measuring the concentration of heavy metal ions in aqueous suspensions via CD fluorescence ensured the simultaneous determination of three ions' concentrations. Application of ANN can potentially increase the number of ions, while in most publications the number of target parameters does not exceed two [1-3]. The application of even simple ANN architectures (perceptrons) to a part of the database (fluorescence spectra corresponding to only 3 excitation wavelengths, and not 41) provided the following accuracy in determining the (average absolute error) ion concentration: 0.27±0.03 mM for Cu2+, 1.12±0.08 mM for Ni2+, 0.19±0.01 mM for Cr3+. The obtained results are comparable with similar data of carbon sensors for diagnostics of media with two ions [3]. This study has been supported by Russian Foundation for Basic Research No. 20-32-70150 (K.A.Laptinskiy). The contribution of O.E. Sarmanova (programming and training of neural networks) was supported by noncommercial Foundation for the Development of Science and Education "Intellect" (Project No. ASP-11-NS_FF/2021). References: 1. M. Zulfajri, G. Gedda, C.J. Chang, et. al, Cranberry beans derived carbon dots as a potential fluorescence sensor for selective detection of Fe3+ ions in aqueous solution. ACS omega 2019, 4(13), 15382-15392. 2. S. Wang, H. Chen, H. Xie, et. al, A novel thioctic acid-carbon dots fluorescence sensor for the detection of Hg2+ and thiophanate methyl via S-Hg affinity. Food Chemistry 2021, 346, 128923. 3. Y. Gong, H. Liang, Nickel ion detection by imidazole modified carbon dots. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 2019, 211, 342-347.