Showing papers by "College of Engineering, Pune published in 2022"

A Bayesian Optimized Discriminant Analysis Model for Condition Monitoring of Face Milling Cutter Using Vibration Datasets

Abstract: With the advent of Industry 4.0, which conceptualizes self-monitoring of rotating machine parts by adopting techniques like Artificial Intelligence (AI), Machine Learning (ML), Internet of Things (IoT), data analytics, cloud computing, etc. The significant research area in predictive maintenance is Tool Condition Monitoring (TCM) as the tool condition affects the overall machining process and its economics. Lately, machine learning techniques are being used to classify the tool's condition in operation. These techniques are cost-saving and help industries with adopting future-proof solutions for their operations. One such technique called Discriminant analysis (DA) must be examined particularly for TCM. Owing to its less expensive computation and shorter run times, using them in TCM will ensure effective use of the cutting tool and reduce maintenance times. This paper presents a Bayesian optimized discriminant analysis model to classify and monitor the tool condition into three user-defined classes. The data is collected using an in-house designed and developed Data Acquisition (DAQ) module set up on a Vertical Machining Center (VMC). The hyperparameter tuning has been incorporated using Bayesian optimization search, and the parameter which gives the best model was found out to be ‘Linear’, achieving an accuracy of 93.3%. This work confirms the feasibility of machine learning techniques like DA in the field of TCM and using Bayesian optimization algorithms to fine-tune the model, making it industry-ready.

Management of IoT Devices Security Using Blockchain—A Review

Abstract: Internet of things (IoT) refers to a network where the devices included in that network are connected with each other through a common medium, in this case, the Internet, in order to share or exchange data with other devices in the network. The paper concentrates on Management of Security of these devices in the IoT network along with their maintenance, accessibility, etc. The main problems faced are data leakage, data alteration/modification, access to private data or important transactions, data loss, etc. This review paper refers to various ways to improve the existing IoT system with the use of different consensus algorithms and techniques. It also covers the security and data privacy of systems like smart homes and smart cities through modified blockchain systems.

Botnet Detection Using Bayes Classifier

Abstract: In today’s connected world, risk of getting attacked over the internet is increased, which plays a major role in infecting the devices over the internet. The internet is flooded with different malwares, but we have focused on the harmful effects of Botnet. Botnet is a group of devices controlled by a single device to attack and infect other devices over the internet. The devices are called bots and these can be any internet-connected device and the single device controlling these can be called as a botmaster or a bot driver. It is crucial to detect them at a faster rate since they can perform various malicious activities. We performed different experiments to detect Botnet. For experimentation, we used CICIDS2017 dataset and different machine learning algorithms from Weka. With the ML algorithms, we achieved the highest accuracy of 98.9146% for NaiveBayesMultinominalText algorithm.

Deep Learning for COVID-19: COVID-19 Detection Based on Chest X-Ray Images by the Fusion of Deep Learning and Machine Learning Techniques

Abstract: The proposed book chapter focuses on development of an efficient AI based medical imaging solution for COVID-19 by leveraging the easily available COVID X-Ray Images (CXR). For this the experimentation with different deep learning and machine learning algorithms is performed. A convolution network (CNN) is one of the widely used deep learning algorithms used for medical imaging systems. In this chapter different variants of CNN are used. Data augmentation and dropout techniques are used to avoid overfitting. Among these different variants, CNN with ten convolutional layers and 15 epochs has given the best performance of 88.23% training 85.94% validation accuracy. This is followed by the use of Alexnet for feature extraction from CXR images and the extracted features are given as the input to different machine learning classifiers including Gaussian Naive Bays, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Gradient Boost, Ada Boost for classification of input images among the classes of Covid-19, No Finding, Pneumonia. Among these machine learning classifiers, SVM has given the best performance of 86% testing accuracy. Thus, the deep learning algorithms have proven to give satisfactory performance. This performance can still be improved with the help of more images and by fine tuning the pre trained models. In addition, this chapter highlights the importance of transfer learning, brief description about medical imaging.

Addressing Concept Drifts Using Deep Learning for Heart Disease Prediction: A Review

Abstract: Heart disease is definitely among the many most significant triggers of morbidity and fatality amid the populace among the globe. Prediction of cardiac disease can be considered as one particular among the most crucial topics in the sector of medical info evaluation. The quantity of data through the medical industry is very large. Deep learning becomes the huge range of natural medical care data straight to data which usually may support to identify possibilities and forecasts. This paper reveals the novel algorithm and performance methodology that can forecast the heart disease by ways of CNN modeling. The parameters evaluation will be done for accuracy, sensitivity, specificity, and positive prediction value (PPV). Such parameters can be used in a user-friendly manner by doctors to trace out the possibility of diseases.

Numerical Analysis of Steel Buckling-Restrained Braces with Varying Lengths, Gaps, and Stoppers

Abstract: The conventional buckling type of brace members offers lateral stiffness to the structural system; however, such members are likely to yield under tension and buckle under compression when ...

Analysis of Feature Selection Techniques to Detect DoS Attacks Using Rule-Based Classifiers

Abstract: Denial of Service (DoS) attacks are emerging as a security threat, which, when ignored, may result in enormous losses for the organizations. Such attacks lead to the unavailability of the services provided by the organizations to legitimate users. The detection of such attacks with lower computation and minimization of errors is an ongoing research area. This paper focuses on analyzing different feature selection methods for feature selection in the detection of DoS attacks. The analysis of feature selection methods provides relevant and noisy feature subsets based on the score obtained by each method. The obtained relevant feature subset is tested on the CICIDS-2017 DoS dataset and achieves higher accuracy of 99.9591% with the PART classifier.

Design of a SLAM Map Building System for a Ground-Based Robot Using ROS

Abstract: Robots operate autonomously to perform services for human well-being. Intelligent machines capable of handling tasks on their own without human intervention are autonomous robots. The knowledge of autonomous robots is required for service robots. Service robots are becoming more and more popular nowadays. They can be used in various sectors such as health care, agriculture, hotel, military, etc. So the aim of this paper is to design and develop a mechanical ROS robot on which further a SLAM system for the robot is built, which will be capable of building maps of the environment. Further, this system can also be used in various robotics applications like obstacle avoidance, navigation, etc.

System for Analyzing Crime News by Mining Live Data Streams with Preserving Data Privacy

Abstract: Data stream mining is an emerging field of data science. It is a process of extraction of knowledge from streaming data using incremental algorithms. Mining streaming data comes with different challenges [3] like concept drift, handling incomplete and delayed information, skewness of data, and privacy preservation. Privacy of streaming data should be maintained during the process of its mining and processing to protect sensitive information from attackers and also to preserve user-sensitive personal data that is vulnerable to malicious purposes. In this paper, we proposed a system for mining crime news data streams along with privacy preservation of sensitive data using K-anonymization and Apache Spark. The knowledge gained through the process of mining streaming data is visualized in the form of real-time updating charts which provides the end-user with useful insights about current crime rates and statistics in the popular cities of India.

A Complete Crop Guide Using Machine Learning and AI-Based Chat Bot

Abstract: We all know that India is one of the largest growing economies and the contribution of agriculture here is among the top. Almost 60% of rural India’s income is based on agriculture. The GDP contribution of agriculture is nearly 17%. The main purpose of our project here is to create awareness among farmers as well as to provide a complete agricultural solution to farming. Here, we are going to implement four different modules mainly scheduling an appointment using AI-based chatbot, crop suggestions to the farmers using ML, complete crop plan, and finally predicting the yield. Here, we have used historical data such as soil nutrients, weather, rainfall, etc. In the future, the live data can be used to predict the output.

Multiregional Segmentation of High-Grade Glioma Using Modified Deep UNET Model with Edge-Detected Multimodal MRI Images

Abstract: With newly emerging technologies in the field of computer science, there is rising awareness about its applications in the medical sciences. One of such important applications is early and accurate diagnosis of cancer tumor. Brain tumor is a deadly disease and needs to be diagnosed accurately on time. Among many types of brain tumors, high-grade glioma (HGG) is the most belligerent type, and the aggressive nature of tumors affects the survival outcomes of patients. Image processing and deep learning techniques have helped a lot in this endeavor. The proposed work deals with a hybrid methodology that combines simple edge detection technique with deep convolutional neural network to achieve state-of-the-art results. The Brain Tumor Segmentation (BRATS) 2018 dataset is used, which is provided under BRATS contest organized by MICCAI international conference (Medical Image Computing and Computer-Assisted Intervention). The Sobel operator is used for edge detection, and such edge-detected images are further trained using the modified deep UNET (md-UNET)-based model to segment glioma tumor regions into three classes namely—enhancing tumor, non-enhancing necrotic tumor, and edema. The data of 210 patients is used for training the proposed model using train-validate-test split. The model has achieved an accuracy of 98.94%, dice score of 99.03%, and dice loss of 0.0096 for separate test cohort of 6720 multi-modal MRI images of 40 glioma patients.

Improving Topographic Features of DEM Using Cartosat-1 Stereo Data

Abstract: In the current study, we have generated a Digital Elevation Model (DEM) using high spatial resolution stereo images (2.5 m spatial resolution) of the Cartosat-1 satellite and examined the terrain's quantitative topographic features. Firstly, the DEM is generated through topographic features such as elevation, slope gradient, aspect, hill shade, and contour map. We performed a comparative evaluation of the accuracy of topographic features the DEM generated through stereo images and freely accessible Cartosat-1 DEM data (30 m spatial resolution) with other references DEMs such as Shuttle Radar Topography Mission (SRTM) DEM and ALOS global DSM (AD3D30). The visual analysis of all the DEMs is done through a surface profile map. The surface profile map of DEM generated through stereo images shows a good correlation with reference DEMs in all regions of the profile map. This study reveals that the Cartosat-1 DEM generated through stereo images gives better accuracy than freely accessible Cartosat-1 DEM.

Wind Speed at Hub Height (Using Dynamic Wind Shear) and Wind Power Prediction

Abstract: Prediction of hub-height wind speed with the ground-level (10 m) wind speed is difficult as the wind is chaotic. Several forecasters provide wind speed forecasts, but due to variations in hub heights, conversion of a hub-height wind speed is challenging. At present, lots of research is going on to predict the wind speed by using mathematical formulae and statistics, and biologically inspired computing have also been used to predict particular height wind speed. Weather parameter affects the accuracy and increases the error band. To solve this issue, the models have been created based on the Decision Tree Regressor/Keras Neural Network ML technique, which uses the weather parameter and ground-level wind speed to predict the wind shear. These attributes will help in predicting the wind particular hub height and wind speed for at least 1.5–3 h. Besides, there are also two power forecast models (Decision Tree Regressor/Keras Neural Network ML) which take the hub-height wind speed and weather parameters as input and forecast the power generation for the given power plant. It also provides brief information about the power-law method to calculate the wind shear coefficient. This model will help many wind power plants know about the present wind prediction model capabilities; it will also allow us to predict the particular hub-height wind speed and power generation for their specific wind farms.

Effective Text Comment Classification Using Novel ML Algorithm - Modified Lazy Random Forest

Abstract: Machine learning (ML) algorithms are methods used to classify data. The various patterns or classes can be classified with the help of these various ML algorithms. There are numerous areas where these algorithms can be used. One such area is to detect whether the comment, sms or text message is SPAM or Normal message. So the aim of this work is to identify the best machine learning algorithms to detect SPAM text message on two different dataset. The first dataset is collected from YouTube comment dataset and second is the SMS dataset. The Random Forest (RF) is the ensemble learning method for classification, regression and other tasks that operates by developing a multitude of decision trees at learning phase and outputting the class. Its one variant which performs well as compare to normal RF is Lazy RF, as the study shown and is the base for this research work. In this work, we have proposed one more novel variant of LRF and the different machine learning algorithms are compared in terms of accuracy with proposed Modified Lazy Random Forest. The results are compare with two techniques, first is the simple hold out, and second is the K-fold cross validation. For both cases the proposed algorithm performs well to detect the SPAM messages for the both datasets.

Code Generation from Images Using Neural Networks

Abstract: In this paper, we have proposed and demonstrated a method that generates graphical user interface front end code from user drawn screen images. We apply classical computer vision techniques viz. canny edge detection, contour box detection, dilation and erosion for image, and two deep learning techniques of text detector and convolutional neural networks. The process of creating an application starts with a graphical user interface designer, creating a layout that becomes the blueprint for the web developer. The web developer sends the layout back to the designer with desired changes. This process is iterative, and extra efforts are spent to recreate the user interface code. We provide a unique way to automate this iterative process, to give a template code from the layout image.

Experimental Exploration of Effect of Hydrogen Enrichment on the Performance and Emissions of Dual Fuel Diesel Engine Equipped with CRDI by Varying Injection Duration

Abstract: The hydrogen due to its carbonless structure is considered as a potential supplement fuel in near future for dual-fuel Internal Combustion engines. It reduces the burden of energy imports and reduces carbon containing tailpipe emission, thereby protecting the environment. Hydrogen has inimitable characteristics because of carbonless structure which is considered as better alternative fuel compared to other available options, for example, liquefied petroleum gas, compressed natural gas, etc. In the presented study, investigation was conducted using hydrogen gas in dual-fuel method in single-cylinder CI engine. Hydrogen fuel was inoculated in intake manifold for different injection duration while injecting diesel into combustion chamber directly. The performance of engine is compared with baseline diesel performance at varying injection duration. Experimental observations demonstrated the performance improvement and exhaust emissions using hydrogen enrichment technique. The brake thermal efficiency observed to be improved by 3.17%, and brake specific energy consumption reduces by 10.81% at fully loaded condition for hydrogen gas injection duration of 6 ms as that of baseline diesel performance. Improvement is seen in performance parameters as well as in emissions also, hydrocarbon reduces by 68.18%, and carbon dioxide reduces by 43.33% at full load condition with same injection duration of 6 ms. It was experiential that due to homogeneous mixing of hydrogen with air leads to complete combustion of fuels with lesser emissions. The current study proved that hydrogen enrichment is a potential technology which could be used in compression ignition engines to improve performance and lessening emissions without any major modification in hardware of basic diesel engine.

Investigations of Wear Behavior of Journal Bearing Materials

Abstract: Behavior of wear in tribological condition in metal-based brass and graphite material at dry and wet sliding conditions. Wear test of these materials were performed on wear test machine. The test of wear in mass loss was performed under different loads 10, 20, 30 N with interval and at different 1000, 1200, 1400 rpm speed, etc. Wear characteristics of materials are investigated through load, speed, and time. The main objective is to investigate the wear rate of journal bearing materials at different loads and different speeds.

Online System for Identifying Need of Machine Maintenance by Mining Data Streams and Handling Concept Drifts

Abstract: Data streams represent an ongoing stream of data, in many forms, coming from different sources. In real time data often comes in streams and is changing over time. Concept drift in supervised learning means that the data is going through a change. While solving predictive maintenance tasks on the streaming data, traditional models, trained on historical data, may become invalid, when such change occurs. Hence, the learning models need to adapt to changes very quick and accurately. Adaptive ensemble models are used for classification on data streams. In this paper, we implemented the modifications of the adaptive bagging methods, which uses internal class-weighting schemes for the model adaptation. Implemented models were evaluated on manually created data streams with ensemble methods and analyzed performance evaluation of different classifiers. This performance is greatly differed than the traditional model and hence handles the drift in much more effective way.

Comparative Study of Application of Artificial Neural Networks for Predicting Engineering Properties of Soil: A Review

Abstract: The primary aim of the synthetic neural network approach was to unravel the issues similarly that a person’s brain would. The artificial neural network system was extensively applied in geotechnical engineering. Geotechnical engineering properties of soil hold the solidity of engineering structures. The engineering properties of soils are much worried about the distortion and strength of bodies of soil. Engineering properties of soil which measure the engineering behavior of soils. This review paper presents a quick overview of artificial neural network (ANN) applications of engineering properties of soil, viz. optimum moisture content, maximum dry density, permeability, shear strength parameters, and unconfined compressive strength. The review suggests that ANN with different models can predict the engineering properties of soil accurately. The survey recommends that the ANNs had been exceptionally valuable in effectively interpreting inadequate input information. This study shall help the researchers those working in the area of applications of ANN on soil behavior.

Application of Data Mining Technique for Retail Industry

Abstract: This research work presents an implementation of a hybrid model for consumer segmentation in online shopping business support and decision-making by employing an unsupervised learning algorithm and Fuzzy C-means approach. In this research, the customers are sub-divided into distinct segments by using the cluster analysis method. It will gather the information about component scores, cluster model and similarities among the customers and suggests a way for grouping them. In specific, the proposed model has used K-means clustering, which is a type of unsupervised learning method. The procedure aims to classify a given dataset through a certain number of clusters (assuming k clusters) fixed a priori.

A Review of Mesh Generation in ANUGA

Abstract: ANUGA is a freely available inundation software developed by the Australian National University (ANU) and Geoscience Australia (GA). It is a tool used for 2D hydrodynamic modeling of realistic flow problems such as tsunamis, floods, storm surges, or dam breaks and can be used to simulate their effects on the environment. It is based on a finite volume method used for solving the shallow water wave equation. It makes use of a mesh of triangular cells to represent the area of study. The mesh is generated following the properties of Delaunay triangulation. This paper discusses the methods of mesh generation used by ANUGA, the importance of an accurate and optimal mesh for a good prediction and proposes a new method of mesh generation.

Circularly Polarized 1 × 4 Antenna Array with Improved Isolation for Massive MIMO Base Station

Abstract: This paper proposes a circularly polarized 1 × 4 antenna array with improved Isolation for Massive MIMO Base Station application. Massive MIMO playing an important role in the design and implementation of 5G. An antenna array is designed using a Hexagonal microstrip antenna. The proposed antenna has eight ports in the design process. Circular polarization is incorporated using a dual coaxial probe feed technique with equal amplitude and 90-degree phase shift. An antenna array is simulated at a spacing of 0.50λ, 0.55λ, and 0.60λ. Improved Isolation is achieved at a spacing of 0.55λ. The proposed antenna is simulated using HFSS13.0v at 3.7-GHz frequency and fabricated on a Rogers RT/duroid 5880. Designed antenna have an impedance bandwidth of 160 MHz (at S11 = −10 dB), gain of 4.97 dB per port, and axial ratio of 0.27 (<3 dB). The inter-element spacing of 1 × 4 antenna arrays is analyzed using HFSS so that Isolation will be greater than 20 dB. Measured and simulated results are found in good agreement.

A Survey on Diabetic Retinopathy Detection Using Deep Learning

Abstract: Diabetic retinopathy happens when there are high blood pressure and high sugar level in the body that damages the blood vessels and veins in retina. These arteries can become swollen and leaky, or they may close, block the flow of blood. Sometimes new, unusual blood arteries grow in the retina part. These unconditional changes can steal your eyesight. Manual examination and analysis of fundus images to detect morphological changes in the eyes are very sluggish and tedious. In the current scenario, deep learning has been set up as the most popular approach with superior performance in various areas and over traditional machine learning methods, especially in image analysis and treatment. In this paper, we adhere to traditional strategies mainly containing input Data acquisition, pre-data processing, segmentation and data preparation, feature measurement, feature extraction, model creation, model training, model testing on testing data, and outcome and analysis of the model. We have reviewed various algorithms and their challenges that help in the diagnosis of methods used in the detection of diabetic retinopathy.