Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Madalina Carbureanu has a total of 21 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Quantifying and Improving Stereo Camera Calibration Robustness: An Outlier-Aware Algorithm for Digital Twin Data Acquisition

    by Madalina Carbureanu, Florin-Stefan Zamfir

    As calibration errors have a direct impact on epipolar consistency, rectification accuracy, and metric 3D reconstruction performance, stereo camera calibration is a fundamental requirement for high-accuracy 3D modeling and reliable digital twin data acquisition. Because current calibration workflows (based on pairwise calibration methods) lack systematic data-quality checks mechanisms, there is a clear need for more robust data selection strategies. The novelty of the approach consists in the development of a new outlier-aware stereo calibration algorithm (OutAw) that introduces a unified multi-stage approach that integrates hard geometric selection, candidate subset generation, multi-criterion ranking, bootstrap stability analysis, and triangulation assessment into a comprehensive and systematic calibration framework. Unlike conventional approaches, OutAw (through its mechanism of detecting and rejecting inconsistent pairs) redefines the calibration strategy from arbitrary to criterion-based data selection. Also, the proposed algorithm is compared with BSC (a baseline OpenCV all-pairs calibration algorithm) and InterFil (an intermediate filtered variant) using 49 stereo pairs (at 1280 × 720 resolution) captured using a planar checkerboard. OutAw algorithm achieved (using only nine image pairs) superior results (epipolar error 0.5119 px, stereo RMS 0.7666 px) to the BSC ones (epipolar error 1.3687 px, stereo RMS 1.9385 px), representing statistically significant improvements (60.5%, respectively 62.3%). OutAw geometric consistency was validated by triangulation-based metrics (square-length standard deviation 0.1140 mm and square absolute error 0.1097 mm). Contamination analysis revealed that as the outlier rate increases, the calibration process degrades progressively. Also, the results obtained highlight that geometric quality-driven image selection is critical for achieving a reliable stereo calibration for DT applications.

  • Open Access

    A Machine Learning-Based Data-Driven Model for Predicting Wastewater Quality Parameters in the Industrial Domain

    by Madalina Carbureanu, Catalina Gabriela Gheorghe

    This study proposes HGBRCond, a machine learning model for conductivity prediction in controlled biodegradation processes. Eight regression algorithms were evaluated using experimental data (n = 424) from a micro-pilot treatment system. HGBRCond, based on Histogram-Gradient Boosting Regression (best performing ML model), achieved optimal performance (R2 = 0.877 ± 0.011, RMSE = 10.235 ± 0.54 µS/cm) through 10-fold cross-validation. Unlike standard HGBR and previous conductivity models that lack comprehensive validation frameworks, HGBRCond integrates rigorous statistical validation (cross-validation, sensitivity analysis, confidence intervals) with multi-level interpretability (Morris screening, SHAP analysis, feature importance), achieving a 6.8% performance improvement over standard gradient boosting approaches while addressing mechanistic interpretability gaps present in prior work. However, limitations constrain direct potential industrial applicability: limited dataset (n = 424), narrow conductivity range (285–360 µS/cm), strong dissolved oxygen dependence, sensitivity across two critical parameters, constant flowrate, and validation restricted to controlled conditions. These constraints require model recalibration for potential industrial application. Future work will focus on model validation across extended operational ranges using industrial samples and full-scale testing to establish applicability beyond controlled experimental settings.


  • Top Cited Papers
    Open Access

    Application of Machine Learning Models in Optimizing Wastewater Treatment Processes: A Review

    by Florin-Stefan Zamfir, Madalina Carbureanu, Sanda Florentina Mihalache

    The treatment processes from a wastewater treatment plant (WWTP) are known for their complexity and highly nonlinear behavior, which makes them challenging to analyze, model, and especially, to control. This research studies how machine learning (ML) with a focus on deep learning (DL) techniques can be applied to optimize the treatment processes of WWTPs, highlighting those case studies that propose ML and DL methods that directly address this issue. This research aims to study the ML and DL systematic applications in optimizing the wastewater treatment processes from an industrial plant, such as the modeling of complex physical–chemical processes, real-time monitoring and prediction of critical wastewater quality indicators, chemical reactants consumption reduction, minimization of plant energy consumption, plant effluent quality prediction, development of data-driven type models as support in the decision-making process, etc. To perform a detailed analysis, 87 articles were included from an initial set of 324, using criteria such as wastewater combined with ML, DL, and artificial intelligence (AI), for articles from 2010 or newer. From the initial set of 324 scientific articles, 300 were identified using Litmaps, obtained from five important scientific databases, all focusing on addressing the specific problem proposed for investigation. Thus, this paper identifies gaps in the current research, discusses ML and DL algorithms in the context of optimizing wastewater treatment processes, and identifies future directions for optimizing these processes through data-driven methods. As opposed to traditional models, IA models (ML, DL, hybrid and ensemble models, digital twin, IoT, etc.) demonstrated significant advantages in wastewater quality indicator prediction and forecasting, in energy consumption forecasting, in temporal pattern recognition, and in optimal interpretability for normative compliance. Integrating advanced ML and DL technologies into the various processes involved in wastewater treatment improves the plant systems’ predictive capabilities and ensures a higher level of compliance with environmental standards.


  • Open Access

    Data-Driven Approaches for Predicting and Forecasting Air Quality in Urban Areas

    by Cosmina-Mihaela Rosca, Madalina Carbureanu, Adrian Stancu

    Air quality (AQ) is one of the most important urban environment indicators for the quality of life. The paper proposes a software solution for predicting and forecasting the air quality index (AQI) in urban areas. The study integrates pollutant factors (CO, NO2, SO2, PM2.5), meteorological parameters (temperature, humidity, wind speed), and traffic data to determine air quality. For this purpose, 19 predictive models were developed and compared: 12 machine learning algorithms, 7 deep learning, and 1 forecasting model based on structural component analysis. The Random Forest Regression model, customized within the study, achieved the best results, with an R2 score of 99.59%, an MAE of 0.22%, an MAPE of 0.68%, and an OP (Overall Precision) score of 95.61%. It was subsequently validated on unseen data and recorded a mean deviation of 0.58%. For short-term AQI forecasting (5 days), the AQIF model achieved an R2 of 71.62%, an MAE of 0.4%, and an MAPE of 0.9%. The proposed solution was integrated into a web application with IoT infrastructure and real-time alert mechanisms. Future directions include expanding the dataset and optimizing hyperparameters for the deep learning models to increase accuracy, as well as integrating PM10 and O3 factors, along with the degree of industrialization and demographic level.


  • A COMPARATIVE STUDY OF DEEP LEARNING METHODS APPLIED FOR WASTEWATER pH NEUTRALIZATION PROCESS MODELLING

    by Madalina Carbureanu

    In a wastewater treatment plant (WWTP) there are several different, intricate processes with a dynamic and nonlinear behaviour. The fact that these processes are nonlinear, some of them having a high degree of nonlinearity, as is the wastewater pH neutralization process, comes with a number of problems related to their modelling and control. The identification of any method that can be used to simplify the modelling and control of such a high nonlinear process, it is a desideratum to ensure a quality effluent of the plant, because its water quality is affected by the treated wastewater discharged into it. The Deep Learning (DL) and Machine Learning (ML) techniques offer incredible solutions that can be explored in order to find out the optimal tool that can be used for wastewater pH treatment process modelling. In the present paper, seven DL solutions were implemented and tested in order to identify the most appropriate DL method for modelling this type of process, method that ensures the best result. The analysed DL methods are Feedforward Neural Networks (FNNs), Temporal Convolutional Networks (TCNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), General Regression Neural Networks (GRNNs), Time Delay Neural Networks (TDNNs) and Deep Belief Networks (DBNs), being implemented using Python 3.9 software and Tensorflow. The analysis made with the mentioned DL methods, was based on knowing the flowrate of the acid reactant (which was maintained constant), the initial alkaline reactant flowrate in the treated solution, the initial pH level, and the desired pH level, with the final goal of predicting the required quantity of alkaline reactant flowrate necessary to obtain a neutral pH.

  • MACHINE LEARNING METHODS APPLIED IN AIR QUALITY PREDICTION

    by Romania, Mihai-Claudiu Vieru, Mădălina Cărbureanu

    Air quality is an important environmental component that has a significant influence on public health and well-being. Poor air quality can cause a variety of health problems, including respiratory and cardiovascular disorders. Therefore, there is a growing demand for air quality prediction tools to enable consumers and authorities to take the best decisions and to implement the necessary actions to reduce air pollution. The present paper describes an innovative application that uses machine learning techniques to supply to the users real-time air quality predictions made on past data from their unique location. The Scikit-learn Python package was used to implement five machine learning algorithms, including K-Nearest Neighbors, Random Forest, Gradient Boosting, Support Vector Regression (SVR) and AdaBoost. To achieve robust model performance, compatibility with cross-validation approaches was evaluated. The obtained results indicate that these machine learning techniques are successful at forecasting air quality. The AdaBoost method emerged as the best accurate predictor after extensive investigation, closely followed by Gradient Boosting, SVR, Random Forest, and K-Nearest Neighbors. Furthermore, the investigation also focused on the adapted handling of inaccurate data and providing graphical visualizations to highlight the algorithm's efficacy.


  • INTEGRATED CONVOLUTIONAL NEURAL NETWORKS APPLICATION FOR ROAD TRAFFIC EFFICIENCY

    by Romania, Liviu-Andrei Ștefan, Mădălina Cărbureanu

    Road traffic is a problem that every country has to face for the benefit of its population. Everything related to the transportation of people, goods, and services can be delayed due to a deplorable infrastructure. From minor tasks like individual shopping to important services like ambulances and fire trucks, all vehicles in Romania will eventually pass through a traffic light-controlled intersection that is not managed in real-time and that is also highly inefficient. At such a location, it is critical for traffic flow to be as smooth as possible to prevent time losses, and for traffic information to be stored into a cloud for process optimization. The purpose of the developed application is to reduce the real-time traffic jams to allow the optimal vehicles movement in safe conditions for all traffic participants. Using deep learning (DL) and artificial intelligence (AI) techniques, the application is based on two convolutional neural networks (CNN’s), respectively one for real-time detection of cars and people, and another one for real-time detection of cars license plates. The cars detections with CNN’s is important for road traffic optimization, for the efficient pedestrian detection necessary for crosswalks, and the car licence plates detection necessary for preventing traffic violations. The developed application contains several modules that operate based on deep learning and devices that communicate with each other, ensuring efficient data saving in cloud, data visualization, real-time traffic light control, and the information processing from traffic cameras.

  • INTEGRATED ARTIFICIAL INTELLIGENCE TECHNIQUES APPLICATION FOR WASTEWATER pH CONTROL

    by Romania, Maria-Elisabeta Oprea Grigore, Madalina Carbureanu

    The control of the pH neutralization process has a decisive role in many industrial branches, as: wastewater treatment, biotechnology, pharmaceutical industry, chemical processing, etc. However, the pH neutralization process is a difficult one to control, due to time-varying process parameters, its high non-linearity and to the sudden variation of the pH around the equivalence point. The paper presents a device and an integrated application that uses artificial intelligence (IA) developed for dosage of the necessary reactants in wastewater pH control. The AI techniques were integrated into the application by developing an expert system (ES). For this purpose, were used the necessary if-then rules (rules that make up the inference engine (IE) of the ES) related to the flowrates of the reactants necessary for wastewater pH neutralization. The ES knowledge base (KB) was built based on the recorded pH values. The hardware component (the device) was created using the Arduino equipment.