Temperature Effects on Luminescent Properties of Sr2CeO4:Eu Nanophosphor: a Machine Learning Approach

DRAGUTIN ŠEVIĆ, University of Belgrade, Original scientific paper Institute of Physics Belgrade, Belgrade UDC: 535.37 ANA VLAŠIĆ, University of Belgrade, DOI: 10.5937/tehnika2003279S Institute of Physics Belgrade, Belgrade MAJA S. RABASOVIĆ, University of Belgrade, Institute of Physics Belgrade, Belgrade SVETLANA SAVIĆ-ŠEVIĆ, University of Belgrade, Institute of Physics Belgrade, Belgrade MIHAILO D. RABASOVIĆ, University of Belgrade, Institute of Physics Belgrade, Belgrade MARKO G. NIKOLIĆ, University of Belgrade, Institute of Physics Belgrade, Belgrade BRANKA D. MURIĆ, University of Belgrade, Institute of Physics Belgrade, Belgrade BRATISLAV P. MARINKOVIĆ, University of Belgrade, Institute of Physics Belgrade, Belgrade JANEZ KRIŽAN, AMI, d. o. o, Ptuj, R. Slovenia


INTRODUCTION
Nowadays, nano materials have more and more advantages over bulk materials. Nano science inevitably entered our world [1]. Thermographic nano phosphors are widely used in many applications [2][3][4][5][6][7]. They typically consist of a ceramic host and rare-earth dopant. The temperature dependency of their luminescence is used for remote temperature sensing. For obvious reasons, non contact measurements have many advantages. Thermographic remote monitoring of laser cleansing is described in [8].
Strontium cerium oxide (Sr2CeO4) nano phosphors doped with europium ions (Eu 3+ ), Sr2CeO4:Eu 3+ are described in many scientific papers. As shown in [9], emission color change in a wide range of temperatures proves a great potential of Sr2CeO4:Eu 3+ nanocrystals for industrial applications, particularly in nanothermometric technology. Moreover, additional application possibilities for this material are provided by the fact that the samples with different grain sizes are characterized by various luminescence colors [9]. The possibility of application of this nanophosphor in single-color and two-color fluorescence thermometry techniques in temperature range of 303-523 K has been proposed in [10]. In [11] it was shown that the Eu 3+ doped Sr2CeO4 phosphors emitting white light (by combining blue, green and red emissions) has potential applications not only in the fields of lamps and display devices under 280 nm excitation, but also in the field of LEDs under near UV (350 nm) excitation. Sr2CeO4:Eu 3+ considered as a source of anti-stokes white light generated under near infrared excitation was analyzed in [12]. Various methods of synthesis and studies of structural and luminescent characteristics of nanophosphors based on Sr2CeO4:Eu 3+ or nondoped Sr2CeO4 are reported in [9][10][11][12][13], and references therein.
In this study, we analyze Sr2CeO4:Eu 3+ nanopowders, efficiently prepared using a solution combustion synthesis (SCS) method [14,15]. The main characteristics of this process are simplicity and low cost. The structure of prepared materials has been confirmed and characterized using X-ray powder diffraction (XRD), scanning electron microscope (SEM) and photoluminescence (PL) techniques [15]. The most of europium luminescence comes from transitions from the 5 D0 and 5 D1 state; and they are usually used for fluorescence intensity ratio technique for remote temperature sensing.
In our recent publication [16] we have shown that Sr2CeO4:Eu 3+ made by solution combustion synthesis could be used as a red phosphor. In [15] we have studied the possibility of using the synthesized Sr2CeO4:Eu 3+ for temperature measurements, using usual approach of calculating the calibration curves.
However, availability of more and more fast computers, capable of machine learning, gave us an idea of different approach. Here, we analyze the possibilities of training the computer to recognize optical emission spectra of Sr2CeO4:Eu 3+ at different temperatures. So, this paper describes extension of our work presented in [15].
The prepared starting reagents were combusted with the flame burner at approximately 500 o C, yielding a voluminous foamy pink powder in an intensive exotermic reaction. After the solution combustion synthesis, the nanopowder was annealed for 2 hours, in air atmosphere, at 900 °C. The annealing of the material is needed to achieve optimal optical characteristics of synthesized material.

Experimental details
As an excitation source for photoluminescence measurements we used the output of the optical parametric oscillator (Vibrant OPO), continuously tunable over a spectral range from 320 nm to 475 nm. Laser pulse duration is about 5 ns, at a repetition rate of 10 Hz. Time-resolved streak images of the luminescence response of Sr2CeO4:Eu 3+ nanopowder excited by the OPO system were acquired by Hamamatsu streak camera equipped with a spectrograph.
Emission spectra of Sr2CeO4:Eu 3+ were also acquired using Ocean Optics USB2000 and AVANTES AvaSpec 2048TEC USB2 spectrometers and continuous laser diode excitation at 405 nm. The experimental setup for luminescence measurement as a function of temperature is described in [17].
For machine learning simulation experiments we have used Solo software package (Version 8.8, Eigenvector Research Inc, USA).

RESULTS AND DISCUSSION
The structure of material was confirmed by XRD patterns and SEM images, see [15].
The streak image of the time resolved photoluminescence spectrum of the Sr2CeO4:Eu 3+ using the 330 nm excitation is presented in Figure 1. Horizontal scale of streak image corresponds to wavelength, vertical scale shows development of spectra in time. Images are presented in pseudocolor, where different colors mean different optical intensities.
The 5 D1-7 F3 transition (583 nm), located closely between the 5 D0-7 F0 (582 nm) and the 5 D0-7 F1 (587 nm) transitions is easy to identify on the time resolved image. Its time integrated peak has a comparable intensity to the intensities of peaks originating from nearby 5 D0 levels (see the line profile denoted by a red curve in Fig. 4).
The luminescence spectra presented in publications usually do not have the time resolution, so it is hard to guess which transitions are short lived. Streak image presented in Figure 1 shows clearly that the 5 D1-7 F3 transition (583 nm) has a much higher intensity and a much shorter lifetime than nearby transitions from 5 D0 state. The temperature dependency of intensity ratio of spectral lines The luminescence of samples was measured both using pulsed (OPO) and continuous excitation. The measured luminescence spectra of Eu 3+ doped Sr2CeO4 at various temperatures are presented in Figure 2. The spectra were obtained by using continuous laser diode excitation at 405 nm.

Figure 2 -Luminescence spectra of Eu 3+ doped
Sr2CeO4 at various temperatures. (Continuous laser diode excitation at 405 nm). The fluorescence intensity ratio (FIR) of spectral line intensities from the 5 D0 and 5 D1 transitions depend on two physical processes: the thermalization of the 5 D1 level with rising temperature, where the energy difference to populate the 5 D1 level from the 5 D0 level is fully covered by phonons; and the nonradiative quenching of the 5 D0 and 5 D1 levels through the charge transfer state. Looking at Figure 2 we can see that intensity ratio between lines at 614 nm and 537 nm is temperature dependent.
The Principal Component Analysis of temperature dependent Sr2CeO4:Eu 3+ spectra Principal component analysis (PCA) finds combinations of variables, or factors, that describe major trends in the data [18].
If X is a data matrix with m rows and n columns, each variable being a column and each sample a row, PCA decomposes X as the sum of rti and pi, where r is the rank of the matrix X: The ti, pi pairs are ordered by the amount of variance captured. The ti vectors are known as scores and contain information on how the samples relate to each other. The pi vectors are known as loadings and contain information on how the variables relate to each other.
For analysis presented here, we use luminescence spectra of Eu 3+ doped Sr2CeO4 at temperatures between 300 and 400 K, measured with the step of 5 K. About a half of the spectral data (measured at 300, 310, 320 ... K) are used to train the PCA algorithm. Another half of the spectral data (measured at 305, 315, 325 .... K) are used to test the obtained PCA model.
Scores on first two principal components of measurement data of temperature dependence of luminescence of Sr2CeO4:Eu 3+ nanophosphor are shown in Figure 3. It could be seen that scores on PC 1 gradually move along the x axis, while scores on PC 2 oscillate along the y axis. Figure 3 -Scores on first two principal components of measurement data of temperature dependence of luminescence of Sr2CeO4:Eu 3+ nanophosphor. We see that the predictions for future measurements are well within the 95% confidence level. However, it should be noted that this approach is different, and not necessarily better, than usual method based on fitting the temperature sensing calibration curve. Namely, the machine is trained on restricted size of training data set. The remote temperature estimation is based on classification of newly obtained data in regard the calibrated data. The classification is implemented by comparing the scores distances between the calibrated data and the newly measured ones. So, the larger the training set, the better is resolution of the remote temperature sensing.

CONCLUSION
In this paper we have applied the Principal Component Analysis, the basic machine learning algorithm, on temperature dependent Sr2CeO4:Eu 3+ spectral data. We have shown that the machine could be trained to differentiate spectral data obtained on different temperatures. However, the resolution of this remote temperature sensing technique depends on the size of spectral data training set. For relatively small training set, the predicted data are well within the confidence level of 95 %.

ACKNOWLEDGEMENT
This work was financially supported by funding provided by the Institute of Physics Belgrade, through the grant by the Ministry of Education, Science, and Technological Development of the Republic of Serbia.