Percorrer por autor "Ahmadi, Mahdia"
A mostrar 1 - 4 de 4
Resultados por página
Opções de ordenação
- Anomaly detection in gas sensor data using LSTM autoencoder and latent space analysisPublication . Ahmadi, Mahdia; Izidorio, Felipe; Igrejas, Getúlio; Rodrigues, Pedro João; Lopes, Rui PedroAnomaly detection in gas sensor data is crucial for food quality control, environmental monitoring, and industrial safety, yet traditional supervised approaches require labeled anomalous data that is often impossible to obtain. This paper presents a single-class LSTM autoencoder for BME688 gas sensor anomaly detection using latent space distance analysis. Training exclusively on normal samples (Anis estrellado), the model detects anomalies by measuring Euclidean distances in the learned 8-dimensional latent space. Treating the 10-step heater profile as a temporal sequence enables the capture of sequential dependencies in gas resistance patterns. Evaluation across seven compounds achieves 100% detection for olive oil and 50.4% for air while maintaining false positive rates at or below 5% for normal classes (coffee: 0.0%, tea: 0.4%, cocoa: 5.0%). Compared to reconstruction-based methods, our approach provides 3.7× better separation, faster inference (6.8ms vs 12.3ms), and improved interpretability, offering an efficient solution for real-time anomaly detection where only normal operational data is available.
- Exploring data analysis and insights on volatile organic compounds for hazardous waste detectionPublication . Ahmadi, Mahdia; Méndez-Pérez, Natalia; Cruz, Helena; Igrejas, Getúlio; Rodrigues, Pedro João; Lopes, Rui PedroVolatile Organic Compounds (VOCs) are critical indicators of environmental contamination, particularly in hazardous waste contexts. While gas chromatography–mass spectrometry (GC–MS) provides high specificity, it struggles with scalability and pattern discovery in large, complex datasets. This study presents a data-driven framework integrating Exploratory Data Analysis (EDA) techniques — including principal component analysis (PCA), hierarchical clustering, and correlation mapping — to uncover emission patterns in compost-derived VOC data. Using the LCSC VOC 2022 Compost Dataset (141 variables, 90 samples), we identified strong co-emission clusters (e.g., D-Limonene and ω-Pinene) and a temperature-dependent ethanol emission pattern unique to food-and-yard waste samples. Pearson correlation analysis revealed shared emission behavior, and regression confirmed a positive slope (25.6) for ethanol versus temperature. These findings highlight EDA’s potential to enhance VOC dataset interpretability and source identification. The proposed framework supports practical applications such as early-warning systems, sensor deployment, and data-informed environmental policy.
- Multi-sensor pattern recognition and real-time data processing for autonomous smart waste managementPublication . Jvarsheishvili, Mariam; Ibrahim, Ahmad; Ahmadi, Mahdia; Igrejas, Getúlio; Soares, Caio; Izidorio, Felipe; Lopes, Rui Pedro; Rodrigues, Pedro JoãoUrban waste management systems require intelligent monitoring solutions that can process multi-modal sensor data in real-time while operating autonomously. This paper presents RAICYCLE, a comprehensive smart waste management system that integrates advanced pattern recognition techniques with real-time operating systems for autonomous urban deployment. The system employs eight BME688 environmental sensors with distinct heater profiles (50°C to 350°C) for volatile organic compound (VOC) pattern classification, combined with VL53L0X Time-of-Flight sensors and GPS tracking. The embedded architecture utilizes FreeRTOS dual-core task scheduling on ESP32 microcontrollers, enabling concurrent sensor data processing, LoRaWAN communication, and system monitoring. Data serialization through Protocol Buffers achieves 70% payload reduction compared to JSON formats, while kinetic energy harvesting from container lid movements enables autonomous operation. The system demonstrates effective real-time processing of 32 dimensional feature vectors for waste classification and environmental monitoring in urban deployments.
- Synthetic data generation for volatile organic compounds recognitionPublication . Ahmadi, Mahdia; Ibrahim, Ahmad Gamal; Jvarsheishvili, Mariam; Igrejas, Getúlio; Izidorio, Felipe; Lopes, Rui Pedro; Soares, Caio; Rodrigues, Pedro JoãoThe fact that machine learning (ML) models to recognize volatile organic compounds (VOC) are typically developed with limited datasets and can be expensive to gather scaled sensor data is an obstacle in their development. The Bosch BME688 is a multi-gas sensor that can give detailed environmental data, but needs large experimental campaigns to construct representative data sets. To overcome this issue, we introduce a Python library on synthetic data generation to the BME688. The tool uses the Kernel Density Estimation (KDE) to generate an empirical gas resistance distribution according to various heater profiles and uses mathematical gas mixing to generate self-configurable multi-gas simulations. Experiments by validation on coffee and oil gases show that the resulting datasets retain the statistical characteristics of actual measurements, both at the stepwise level of gas resistance distributions and at the multivariate level with Principal Component Analysis (PCA). The library generates machine learning reproducible experimentation, machine learning algorithm prototyping on mixtures of percentages, and provision of systematic evaluation of VOC recognition systems. The contribution of the work is a modular and lightweight framework to address the problem of the lack of data, facilitate the reproducible research and speed up the creation of air quality monitoring solutions based on ML.