Zorig Melong- A Technical Journal of Science, Engineering and Technology

COST ESTIMATION MODEL FOR HIGHWAY PROJECTS USING ANN

2023-06-19T06:03:30+00:00

Construction and maintenance works are carried out by contractors for a period of few months to few years. Those contracts show cost variations between the actual and estimated costs. For example, cost estimation of 3 out of 5 projects from 2015- 19 in Trashigang Dzongkhag have not only been underestimated but the actual costs are on average 27% higher than the estimated costs. Moreover, researches reveal that there are issues in cost estimation at the conceptual stage affecting the project progress and cost (Flyvberg, 2002). In this study neural network has been used to predict the actual cost based on the initial estimation and duration. During the study, it was discovered that the neural network demonstrated its usefulness as a tool for cost estimation and could be effectively employed by decision makers. However, it is important to note that the neural network is not designed to replace the entire cost estimation method. Instead, it should be regarded as one of the tools to be utilized alongside other methods when estimating costs.

Developing Resources for Dzongkha Handwritten Digit Recognition

2023-08-22T08:00:51+00:00

Machine learning engineers usually have access to large corpus of data related to their field of study. Unfortunately for researchers studying topics specific to Bhutan, face issue of acute shortage of data or complete lack of it. This paper showcases a part of our project which aims to create a machine learning model to recognize dzongkha handwritten digits. The project is funded by Annual university Research Grant 2022. There are no dzongkha handwritten datasets that we can use for our study hence as a part of our project we had to first develop the dataset to train our model. We were successful in creating the dataset which now consists of 25000 images with 2500 different images per class. The dataset will be used in future to train and create the recognition model as part of our project. The dataset will be then uploaded to GitHub and will be available to future researchers.

Rainfall Prediction using Decision Tree: A Case Study of CST, Phuntsholing

2023-06-19T07:34:31+00:00

In this study we perform hourly rainfall prediction. Climatic data is chaotic in nature and performing regression analysis for short time periods, using limited data recorded by the weather station does not yield good results. Hence, in this study we consider rainfall prediction as a binary classification problem and classify rainfall events into two classes: rainy (positive class) or non-rainy (negative class). Using the independent climatic parameters of the current hour the rainfall status of the next hour is predicted. The dataset used was collected from CST weather station and contains records of 8 weather parameters recorded hourly. We want to study the usability of this data collected by CST weather station for predictive tasks. Since, there is no baseline prediction result on this dataset, we used logistic regression as the baseline model. The accuracy score of logistic regression was 73%. Decision tree which is the focus of this study to perform binary rainfall classification is a popular supervised machine learning algorithm, which forms a flowchart like structure where each internal node represents a feature. The optimization of parameters was conducted through grid search and we used k-fold validation with k value of five and we achieved an accuracy score of 79 percentage.

THE FOREST FIRE HAZARD MAPPING IN BHUTAN

2023-06-19T06:15:12+00:00

The study intends to assimilate data for assessing fire hazards for the forest areas in Bhutan. Over the past few decades, there has been an increase in fire events and forest fires have made it urgent to research this issue. Bhutan documented 1,403 occurrences of forest fires in less than ten years (Kuensel, 11th Jan. 2020). Many fossil fuels are burned during forest fires, and a significant amount of carbon dioxide is produced, which adds to the most prevalent greenhouse gas and changes for the regional climate. Moreover, studies have indicated that one of the influential elements contributing to forest fires is climate change. The goal of this research is to consolidate all datasets from multiple sources onto a single GIS platform for the creation of maps that take climatic conditions into account while calculating forest fire hazard. To lessen the risk of forest fires, recommendations and adaptation methods are offered in this paper. In particular, the Department of Forest and Park Services in Bhutan can benefit from the findings of this research for the development, planning, and implementation works.

The Effect of Tilt Angle Variation on the Performance of the Flat Collector Solar Thermal System (Passive Type)

2023-06-19T05:09:16+00:00

The study demonstrates the importance of proper design and installation of a passive solar thermal system and the need to consider the appropriate collector tilt angle for optimal thermal performance. The experimental setup consisted of a 200-liter passive system installed alongside with an existing 500-liter active thermal system, enabling a solar hot water capacity of 700 liters. The temperature of the flat-type collector outlet and inlet, hot tank outlet, and ambient temper-ature were recorded using temperature sensor data loggers, while solar irradiance was measured using a pyranometer placed on the collector surface. Four system metrics, namely collector effi-ciency, thermal efficiency, heat loss, and mass flow at different tilt angles, were studied. The thermal performance of the system improves with a steady change in solar irradiation when the collector is placed between the tilt angle of 25° and 30°. However, when the tilt angle is increased above 30 degrees, the system efficiency decreases. The thermal efficiency of 60-70% can be obtained at a collector angle of 20° to 30° at the daily solar irradiance of 300 to 900 W/m2. Furthermore, the experiment emphasizes the importance of timely draining out of hot water from the system. This practice helps maintain the system's efficiency and sustainability.

Vehicle Detection in Bhutan Using Convolutional Neural Network

2023-08-22T07:47:57+00:00

Manual vehicle entry at different checkpoints in Bhutan by police personnel creates traffic congestion and delay. Drivers wait in a queue to register vehicles by providing details such as vehicle type and number. There is no automatic system to detect vehicles. The purpose of the study was to create a machine-learning model to detect different types of vehicles in Bhutan. For this study, a total of 20 popular light vehicle classes were identified. The images and videos were captured. The number of frames was extracted from the videos and different types of data augmentation approaches were then adopted to create variations in the curated dataset for greater model generalization. Then different algorithms were evaluated on this dataset. However, the Convolutional Neural Network outperformed all other algorithms. The training and testing accuracy obtained was 99.85% and 99.62% respectively. The model was then deployed using the Flask web framework.

Fruits and Vegetables Recognition System in Dzongkha Using Visual Geometry Group Network

2023-06-19T06:22:22+00:00

English speaking among the youth is gaining popularity in Bhutan. The Ministry of Education and Dzongkha Development Commission have been trying to promote the national language through different strategies. Furthermore, developed interesting learning materials in Dzongkha. However, youth find it difficult even to name common fruits and vegetables in Dzongkha, and this is an increasing concern. The purpose of this study was to develop an automatic fruit recognition system in Dzongkha using machine learning techniques. Ten classes of fruits and 17 classes of veg- etables that are found in Bhutan are considered for the study. The fruits and vegetable datasets were downloaded from Kaggle and Websites. Furthermore, images were augmented and rotated every 15 degrees. The dataset comprised 405000 images. The model was trained and deployed using customized VGGNet and Gradio. The training and vali- dation accuracy of the proposed model was 97.76% and 97.80% respectively.