Ilija Kamenko

Ilija Kamenko

The Institute for Artificial Intelligence Research and Development of Serbia, Novi Sad, Serbia
orcid

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 August 2012 in Expert Systems with Applications
Verified
Journal Article
Closed access
DMDragan Matić
Dragan Matić
FKFilip Kulić
Filip Kulić
MPManuel Pineda-Sanchez
Manuel Pineda-Sanchez
IKIlija Kamenko
Ilija Kamenko

10 December 2017 in Advances in Production Engineering & Management
Verified
Journal Article
Closed access
NSN. Simeunovic
N. Simeunovic
IKIlija Kamenko
Ilija Kamenko
VBV. Bugarski
V. Bugarski
MJM. Jovanovic
M. Jovanovic
BLB. Lalic
B. Lalic

23 February 2022 in Automatika
Verified
Journal Article
Open access
IKIlija Kamenko
Ilija Kamenko
VČVelimir Čongradac
Velimir Čongradac
FKFilip Kulić
Filip Kulić
This paper presents a novel method for PID (proportional–integral–derivative) controller auto-tuning based on expert knowledge incorporated into a fuzzy logic inference system. The proposed scheme iteratively tries to improve the performance of the closed-loop system. As performance measures, the proposed scheme uses the characteristics of the step response (rise time, overshoot, and settling time). PID parameters in the first iteration can be calculated based on the basic open-loop step response experiment or it is possible to use current parameters. In each successive iteration, step response characteristics are measured and the relative changes expressed in the percentage of value in the first iteration are calculated and converted into linguistic values. The fuzzy expert system computes fuzzy values that are used after defuzzification as multiplying factors for current PID parameters. To achieve a balance between the aggressive and robust closed-loop response, as well as between the slower and the faster one, the fuzzy expert system works in three operating modes: the one for speeding up the system, the one for reducing the overshoot, and the one for a balanced reduction of rise time and overshoot. The performance and robustness are verified by computer simulation using an extensive range of different processes.

22 January 2024 in Foods
Verified
Journal Article
Open access
PDF file
MPMilana Pribić
Milana Pribić
IKIlija Kamenko
Ilija Kamenko
SDSaša Despotović
Saša Despotović
MMMilan Mirosavljević
Milan Mirosavljević
JPJelena Pejin
Jelena Pejin
Triticale grain, a wheat–rye hybrid, has been reported to comply very well with the requirements for modern brewing adjuncts. In this study, two triticale varieties, in both unmalted and malted forms, were investigated at various ratios in the grist, applying different mashing regimes and concentrations of the commercial enzyme Shearzyme® 500 L with the aim of evaluating their impact on wort production. In order to capture the complex relationships between the input (triticale ratio, enzyme ratio, mashing regime, and triticale variety) and output variables (wort extract content, wort viscosity, and free amino nitrogen (FAN) content in wort), the study aimed to implement the use of artificial neural networks (ANNs) to model the mashing process. Also, a genetic algorithm (GA) was integrated to minimize a specified multi-objective function, optimizing the mashing process represented by the ANN model. Among the solutions on the Pareto front, one notable set of solutions was found with objective function values of 0.0949, 0.0131, and 1.6812 for the three conflicting objectives, respectively. These values represent a trade-off that optimally balances the different aspects of the optimization problem. The optimized input variables had values of 23%, 9%, 1, and 3 for the respective input variables of triticale ratio, enzyme ratio, mashing regime, and triticale variety. The results derived from the ANN model, applying the GA-optimized input values, were 8.65% w/w for wort extract content, 1.52 mPa·s for wort viscosity, and 148.32 mg/L for FAN content in wort. Comparatively, the results conducted from the real laboratory mashing were 8.63% w/w for wort extract content, 1.51 mPa·s for wort viscosity, and 148.88 mg/L for FAN content in wort applying same input values. The presented data from the optimization process using the GA and the subsequent experimental verification on the real mashing process have demonstrated the practical applicability of the proposed approach which confirms the potential to enhance the quality and efficiency of triticale wort production.

1 January 2026 in IEEE Access
Verified
Journal Article
Open access
IKIvan Kipelkin
Ivan Kipelkin
IKIlija Kamenko
Ilija Kamenko
JIJovan Ivošević
Jovan Ivošević
AFAlina Fedorova
Alina Fedorova
GZGrigory Zharkov
Grigory Zharkov
MTMax Talanov
Max Talanov
This article presents a unified benchmark of energy consumption per synaptic event across five classes of neuro-computational substrates: graphics processing units, neural processing units, field-programmable gate arrays, digital spiking processors, and in-memory/memristive devices, with biological synapses used as reference bands. The main contribution is methodological as well as empirical. Methodologically, we introduce a common event-level metric that makes heterogeneous systems directly comparable despite major differences in architecture, coding scheme, and learning dynamics. Empirically, we combine measurements obtained by the authors with carefully normalized literature data to map the present energy landscape of artificial and biological neural computation. The results reveal three robust regimes. Dense von Neumann ANN implementations on GPUs, NPUs, and FPGA operate mainly in the $10^{5}$ to $10^{8}$ fJ range per synaptic event. Digital spiking processors reduce this requirement to about $10^{3}$ to $10^{4}$ fJ per event. Memristive and transistor-based artificial synapses span a much broader interval, from array-level values near $10^{6}$ fJ down to a few femtojoules in the most efficient device-level realizations. In particular, the best organic PANI memristive samples approach ~2 fJ per synaptic event, entering the biological vicinity and, in some cases, surpassing the rat reference range. Taken together, the benchmarked landscape spans up to ten orders of magnitude across the full set of systems considered in this study. This result clarifies where current AI hardware stands relative to biological efficiency, identifies event-driven and in-memory computation as the most promising routes toward sustainable AI, and provides a quantitatively grounded reference framework for future neuromorphic benchmarking. INDEX TERMS Artificial neural networks, energy consumption, energy efficiency, memristive devices, neuromorphic computing, spiking neural networks, sustainable AI.

1 October 2012
Verified
Conference Paper
Closed access
DMD. Matic
D. Matic
FKF. Kulic
F. Kulic
IKIlija Kamenko
Ilija Kamenko
VBVladimir Bugarski
Vladimir Bugarski
PNP. Nikolic
P. Nikolic
This paper proposes method for broken bar detection in induction motors at very low slip. The proposed method consists of extracting reliable discriminative feature from a steady state one-phase current signal and design of optimal classifier via a support vector machine. The fault related features are extracted from frequency spectra of a modulus of a motor phase current Hilbert transform series. The features are fed to the support vector machine input and the output indicates rotor condition in respect of broken bar appearance independently of a slip value. Tests are conducted on 1.1kW two poles induction machine in an industrial environment. It is shown that proposed method is accurate, fast, reliable, not hardware costly.