The Internet of Things (IoT) is an interconnection of millions of devices of various types, sensors and surveillance systems to collect and transmit the real-time. There has been a tremendous surge in the IoT with technological advancements in the last decade. The IoT has transformed and is now becoming a part of almost every sector including healthcare, manufacturing, smart cities, agriculture, home appliances and many more. The rapid growth in IoT can be attributed to several factors such as increased connectivity, advancement in technology, proliferation of IoT devices, etc. Although the rapid growth has resulted in several benefits, there has been an increase in the security vulnerabilities including botnet attacks, Denial of Service (DoS) attacks, data privacy concerns, IoT specific malware, etc. To ensure security and reduce the effect of risks associated with it, several solutions have been proposed such as security by design, IoT software regular update and patch management, advanced authentication and encryption mechanisms, network segmentation, Intrusion Detection Systems (IDS), etc. We propose an intrusion detection model based on Deep Learning (DL) strategies to detect the potential intrusions in the IoT network in real-time. In this study we implemented three DL based algorithms which include Generative Adversarial Networks (GAN), Graph Convolution Network (GCN) and a hybrid model implemented by integrating the GAN and GCN. Among the three implemented algorithms the proposed hybrid model achieved the highest performance with an accuracy of 97.5 % followed by GCN gaining an accuracy of 96.3 % and GAN with 89.6 % accuracy. The other performance metrices include precision, recall, f1-score, False Positive Rate (FPR) and False Negative Rate (FNR).
Industries encounter challenges in managing wastewater treatment systems, particularly in controlling aeration, which is vital for maintaining optimal oxygen levels for effective treatment. Frequent malfunctions in aerator control can lead to system damage and operational inefficiencies, impacting environmental compliance and increasing costs. To address these issues, this study examines the potential of an IoT-based monitoring system for aerators. The paper discusses specific IoT applications for aerator monitoring and explores enhancements in monitoring efficiency. The primary objective is to develop a system that maintains optimal oxygen levels by monitoring and controlling aerator operations. The system includes automatic notifications for oxygen level adjustments via the LINE application and manages data through Microsoft SQL Server for centralized storage. The methodology utilizes digital innovation and transformation techniques, integrating IoT technology through sensors, control boards, and Supervisory Control and Data Acquisition (SCADA) software, significantly enhancing monitoring and control efficiency. Automatic notifications facilitate timely issue resolution for operators, reducing their workload in wastewater management. Results indicate substantial improvements in system efficiency and operator workload reduction, highlighting the practical implications of this IoT solution for optimizing wastewater treatment processes in industrial settings.
The Philippines, situated along the Pacific Ring of Fire, faces heightened earthquake vulnerability. To mitigate electrical accidents during earthquakes, it's advised to switch off circuit breakers on distribution panels. However, amidst tremors, people often prioritize seeking shelter or evacuating rather than immediately turning off circuit breakers. This study focuses on designing an earthquake-sensitive circuit breaker with notification features. Its goal is to detect seismic motion, triggering the device to cut off specific loads based on seismic activity and intensity. Notifications serve as safety alerts, mitigating electrical damage and hazards during earthquakes. An automatic calibrating prototype was developed, integrating a 3-axis accelerometer with an Arduino microcontroller and a GSM module to test reliability and accuracy. Observed values aligned with expected sensitivity measurements, validating the prototype's functionality. The reliability of SMS notifications via the GSM module was also tested, yielding a 100% reliability rate in response to seismic movements. Results demonstrate successful automatic tripping across low, medium, and high-risk seismic movements, with a 100% reliability rate for all 30 trials.
Sugarcane is a tropical grass that grows worldwide and is mainly utilized in sugar production. Even in the Philippines, it is one of the contributors for the country's revenue and can be found in most regions, spanning about 422,500 hectares. Sugar production in the Philippines is lower than in most producing countries, which presents a challenge for local sugarcane farmers. One of the reasons for this drop, according to the related studies, is that sugarcane farmers in provinces continue to employ conventional growing methods. Several studies have found that technology is being incorporated into agricultural equipment. utilizing this information, the researchers built a device with a Raspberry Pi and a web camera to classify and detect the quality of sucrose crystals with Yolov5.
Federated learning has gained increasing attention in social applications in recent years. However, concerns have emerged regarding potential threats to the internal information security of learning models. This research addresses these security challenges through several key contributions: We propose a differential privacy implementation for the Federated Averaging (FedAvg) algorithm. This approach aims to create a secure federated learning environment while preserving the benefits of FedAvg's weighted aggregation. We also investigate attack scenarios involving malicious participants in the federated learning process, with a particular focus on model-reversal attacks, where an adversary attempts to extract learning data from the global model. Finally, we evaluate the effectiveness of security measures designed to counter these model-reversal attacks. Through extensive experimentation, we analyze both the ability to protect differential privacy and the potential vulnerabilities in collaborative learning systems. Our findings provide valuable insights into how to safeguard collaborative learning systems while preserving the utility of real-world applications.
This paper presents AgentMILO, an LLM-based conversational agent designed to assist non-expert users in modelling complex Mixed-Integer Linear Programming (MILP) problems, with a focus on production planning and singletraveller routing problem types. By incorporating expert designed knowledge graphs, AgentMILO improves the problem modelling process by guiding users through asking relevant questions to elicit problem specifications from users and formulation of precise MILP models. Through experiments with ten distinct autoanswering agents acting as users, we compared AgentMILO against a general LLM agent without knowledge graphs. The results show that AgentMILO consistently outperforms the general LLM model in clarity in guiding users, question quality, and ease of interaction, while also delivering more precise MILP formulations. In contrast, the general model struggled with inconsistent performance and frequent failures in guiding users effectively. AgentMILO's architecture allows for integration into various domains, with the potential to utilise expert knowledge to design new knowledge graphs tailored to specific scenarios. The complete development framework for AgentMILO, including the prompts, knowledge graphs, tested problem instances, generated conversations, and full experimental setup, is available at https://github.com/arc2022-deakin/AgentMILO.
The experiments presented in this paper aim to assess the effect of proper loading of the end of the robot's arm on its pose repeatability and working cycle time. The measurement is carried out on a FANUC LR Mate 200iC industrial robot using a digital indicator. The experiment was carried out at five different robot loads. For each single load, a measurement in three axes was performed. The experiment also includes an assessment of the influence of the detailed definition of the end of the robot's arm load in its control system on the measurement results. The paper provides insight into the possibility of investigating the performance criteria of industrial robots and highlights the essence of the correct setting of the robot load including the moments of inertia and the position of the center of gravity.
The biosystem is a crucial system grounded in classification and detection, utilizing Artificial Intelligence (AI) approaches or metaheuristic techniques. Currently, cancer of the blood cells is among the deadliest cancers in the world. Acute lymphoblastic leukemia (ALL) is a cancer of blood cells that causes excessive proliferation of lymphocytes. It is extremely time-consuming and expensive to conduct diagnostic calculations. The number of platelets in a patient's blood is computed by a platelet count. A lacking number of platelets can indicate cancer, infection, or other health problems. A patient with too many platelets is at risk for blood strokes. A single drop of blood includes tens of thousands of platelets. The main goal of this paper is how to detect the features of blood cells and classify with predicting cancer type based on platelets analysis by using Bee Colony followed by Reptile Search Optimization (BCRSO) technique. According to the results, BCRSO algorithm performed better in terms of classification efficacy and accuracy rate than other algorithms. Based on simulation results, the proposed method is more effective than previously published research for classification optimization.
This research involves the creation and evaluation of a system that allows for text extraction and automatic question generation (AQG) using a T5 and TrOCR pipeline. With the use of a Raspberry Pi 5, web camera, and a touchscreen display, factoid- type questions are created from image captures of single-column handwritten notes that only contain textual information. The T5 large language model (LLM) used was finetuned using the Stanford Question Answering Dataset (SQuAD) for facilitating question generation. The system had a word error rate (WER) of 0.40, a ROUGE-1 score of 0.358, and a question validity rate of 68%. This research helps to promote the ease of creation of learning materials in learner education.
Vehicle recognition systems are used in various sectors of our modern society. Environmental factors greatly affect the efficiency of these systems. This study aims to determine the effect of weather conditions such as direct sunlight, cloudy weather, and rain on a YOLO based system's ability to detect a vehicle's model, logo, and license plate.
Wildfires pose significant ecological, economic, and societal challenges in Australia, with their frequency and intensity exacerbated by climate change. Accurate wildfire prediction is critical for effective fire management strategies. This study investigates the performance of machine learning models incorporating time-series data for wildfire prediction across seven Australian regions. Comparisons between models with and without SMOTE and ANOVA revealed that these techniques did not consistently improve performance. Without SMOTE and ANOVA, CatBoost achieved the highest overall accuracy (84.35 %), followed closely by HXGBoost (84.32%}) and LightGBM (84.16%). After applying SMOTE and ANOVA, all models experienced slight declines in accuracy, sensitivity, and F1 scores, indicating that oversampling and feature selection may have introduced noise rather than meaningful improvements. Regionspecific models significantly outperformed a single model for all of Australia, with exceptional accuracies observed in Queensland (96.36%) and Western Australia (97.95%). However, challenges persisted in regions like South Australia, where accuracy dropped further after applying SMOTE and ANOVA (from 70.16% to (65.03%), highlighting the need for more targeted feature selection and data-balancing techniques.
Electrocardiograms (ECGs) are essential tools for diagnosing cardiac conditions. However, analyzing 12-lead ECG signals manually is time-consuming, making automated classification crucial for efficient and accurate diagnosis. This study investigates both handcrafted and deep learning (DL)-based feature extraction techniques for classifying 12-lead ECG signals. The aim is to enhance diagnostic accuracy and efficiency. We extracted QRSTP peaks and computed various time-domain features, including heart rate, heart rate variability ($H R V$), $R R$ intervals, median $R-R$ intervals, SDNN, RMSSD, and PNN60, based on the $R$-peaks during the handcrafted feature extraction process from the 12 -lead ECG signals. Additionally, we applied several DL algorithms, including CNN, ResNet18, VGG16, and DenseNet83, to extract new global features from the raw ECG signals. We used a merged dataset from five different sources, including the “CPC Database” and “CPC Database Extra” from the China Physiological Signal Challenge 2018 (CPSC2018), the “St Petersburg INCART 12-lead Arrhythmia Database,” the “PTB Diagnostic ECG Database” and “PTBXL” from the Physikalisch-Technische Bundesanstalt (PTB), the “Georgia Database,” and an undisclosed American database. The dataset comprises 22,797 12-lead ECG recordings. We have conducted a series of experiments for performance evaluation for feature extraction and training models. The experimental results show that combining handcrafted and DL features outperforms the DL-based methods in improving classification performance. Both quantitative and qualitative studies, along with ablation experiments, are conducted to further validate our approach.
This research investigates the impact of music reflexology on brain activity, utilizing Electroencephalogram (EEG) features to analyze neurological responses. The study addresses two primary objectives: Identifying the effectiveness of music listening on humans and analyzing the relationship between different signals released by the different activity and music reflexology using EEG feature (Alpha, Beta, and Theta Bands). EEG data were collected from various subjects during induced stress (Cognitive task English comprehension test) and listening to the music in the reflexology sessions. EEG signal comparisons across various activities reveal that stress or cognitive demands increase Beta wave activity while decreasing Alpha and Theta waves. Conversely, relaxing activities like listening to calming music increase Alpha and Theta waves and reduce Beta waves. In conclusion, music significantly influences human emotions and physiological states, particularly through compositions with slow tempos, harmonic structures, and acoustic instruments that enhance relaxation and reduce stress.
Hate speech detection is essential in combating online toxicity, particularly in mixed-language or code-switched texts prevalent on social media. Traditional natural language processing (NLP) models often struggle with these complex linguistic structures due to blending multiple languages. This paper investigates the effectiveness of BLOOM (BigScience Large Open-science Open-access Multilingual language model) and XLM-RoBERTa, two powerful multilingual models, in addressing these challenges. BLOOM's extensive pre-training across diverse languages and XLM-RoBERTa's robust capabilities allow a nuanced understanding of context in mixed-language environments. We fine-tune both models on an English and Indonesian text dataset containing instances of mixed-language hate speech and evaluate their performance against state-of-the-art benchmarks. Our findings highlight the effectiveness of these models in recognizing hate speech in mixed-language scenarios, with the fine-tuned BLOOM (bloom-560m) performing better than XLM-RoBERTa (xlm-roberta-base).
In laser powder bed fusion 3D printing, defects can be created in less than 100 mu s, making it difficult and expensive to monitor and control the process in real time. Existing techniques either rely on the use of simple, high-speed sensors for control purposes (e.g., photodiodes) or high-resolution, low-speed sensors (e.g., cameras) to analyze print defects. Since both techniques have limitations (either lack of accuracy or speed in defect detection), we propose that fusing both sensor types can provide detailed print quality indicators at a high enough speed for real-time control. Our data fusion works in a hybrid fashion by first processing images from a high-speed camera (5 kHz) to compute detailed print quality indicators (melt pool size & intensity, number of spatters). These print quality indicators are then fused with very high speed photodiode values (100 kHz) to extrapolate the print quality indicators to 100 kHz. An adaptive weight network is used to perform the fusion and extrapolation, allowing the low-speed print quality indicators to provide context for the interpretation of the photodiode values. Our results on three print data sets show that the proposed data fusion produces print quality indicators within 10 mu s that are significantly more accurate than using only one sensor type alone. The speed, accuracy, and relatively low cost of the proposed data fusion make it a credible option for low-cost monitoring and control of such high-speed processes.
This study quantitatively evaluates how improving delivery time prediction accuracy constructed using real-world traffic information impacts economic efficiency through computational simulation. It proposes a practical framework to optimize delivery operations and support data-driven decision-making. The results show that improving delivery time prediction accuracy led to a 2% increase in revenue and a 21% improvement in customer satisfaction, highlighting the economic and operational benefits of integrating real-world traffic information. This indicates that enhancing delivery time prediction accuracy can lead to economic efficiency gains and improved customer experience, enabling businesses to optimize delivery operations and improve competitive advantage. This research strengthens the foundation of data-driven management and provides a practical framework for establishing a competitive advantage.
Skin diseases are one of the most widespread global health issues, impacting millions of people worldwide. Being the most sensitive part of the body, it requires accurate and quick pre-diagnosis to prevent severe outcomes. The traditional diagnostic methods used by dermatologists are often proven insufficient, which motivated researchers to focus on developing deep learning (DL) systems that can provide early diagnosis and treatment for patients. However, current DL systems face challenges with limited, low-quality, and imbalanced datasets, leading to overfitting and limited generalization. Our proposed method leverages transfer learning with pre-trained models and is tested on DenseNet121, DenseNet201, VGG16, VGG19, and ResNet152, respectively, with modified top layers for skin lesion classification. These models use pre-trained weights from ImageNet as primary values and fine-tune when training on the skin lesions dataset. We use the most challenging ISIC 2019 public dataset, which is popular for its complexity in skin lesion classification. To improve image quality and facilitate more accurate feature extraction, we applied a hair removal algorithm to reduce visual obstacles in dermoscopic images. We evaluated the impact of class weights and data augmentation techniques, including shifting, rotation, zoom, and shear, to improve model generalization. We provide Grad-CAM visualization for model interpretability and qualitative evaluation. Experimental results show that the DenseNet201 achieved the highest accuracy of 96%, demonstrating its effectiveness in our proposed hair removal and class imbalance solutions for skin lesion classification tasks.
Conventional authentication systems often rely on alphanumeric or graphical passwords, or token-based methods. The disadvantages of these systems include the risk of forgetfulness, loss, and theft. Biometric authentication which is a solution to these issues is quickly taking the place of traditional methods and becoming a ubiquitous part of daily life. The electrocardiogram (ECG) is one of the most recent traits considered for biometric purposes. A notable contribution of this work is the introduction of a novel ECG time-slicing technique that outperforms other ECG-based methods. By leveraging machine learning algorithms and tailor-made compact data learning techniques, this research presents a more robust, reliable biometric authentication system. Upon evaluation, the proposed system showed up to 95% identification accuracy when using the optimal machine learning model. These findings could lead to substantial advancements in network information security, with potential applications across various internet and mobile services.
Verifying the safety of cyber-physical systems is widely recognized as an important concern, especially the safety of autonomous driving systems. In this paper, we focus on the PID controller of a robot car and verify its safety requirements. The verification of such systems is done through reachability analysis. However, this problem is undecidable. Therefore, we check the socalled delta-reachability, which is a weaker notion of reachability, to verify the safety requirements. We model the car as a hybrid system and utilize a delta-reachability solver for verification. Several combinations of PID gains are tested and evaluated.
Intelligent traffic signal control is a critical solution for optimizing urban transportation networks by reducing congestion and enhancing traffic flow efficiency. This study introduces a deep reinforcement learning (DRL) framework integrated with probabilistic graph models (PGMs) to address the challenges of adaptability and interpretability in traffic management. The PGMs enable a structured representation of complex dependencies within traffic systems, allowing for interpretable insights into dynamic traffic conditions. The proposed model is trained and validated in the SUMO simulation environment, accurately reflecting real-world traffic scenarios. Results demonstrate substantial improvements in throughput, reduced travel times, and minimized intersection delays compared to traditional control methods. The integration of PGMs with DRL not only enhances decision-making processes but also improves scalability and robustness in real-time applications. This innovative approach lays a solid foundation for implementing adaptive and intelligent traffic management systems, paving the way for smarter urban infrastructures capable of handling evolving transportation demands.
