Abstract:
In this research, robust and reliability-based multi-objective models are proposed to optimize sensor arrays under uncertainty. We have used the information theory for constructing two objectives functions based on selectivity and diversity criteria. A sensor system capable of detecting smokes and volatile organic compounds has been designed and used to demonstrate the proposed model. A statistical criterion, general resolution factor, is used to evaluate the optimization results in the feature spaces. The proposed method will be able to successfully identify a set of Pareto optimal solutions and optimized sensor arrays, providing improved input quality.
Faculty Advisor:
Dr. Kais Bin Zaman, Professor, Department of IPE, BUET.
Abstract:
Forecasting intermittent or sporadic demand is one of the most difficult tasks in supply chain management. There are many non-parametric and parametric methods like Croston’ method, SBA (Syntetos-Boylan Approximation), SBJ (Shale-Boylan-Johnston), TSB (Teunter Syntetos and Babai) etc. are usually used for forecasting sporadic demand but none of these is able to find patterns and extrapolate them for sporadic data. This is because of; patterns hardly appear in sporadic datasets. Implementation of machine learning algorithm can be useful in recognizing these irregular patterns in these datasets. k-Nearest neighbor (k-NN) has been applied for this purpose. Though traditional machine learning algorithms need large training data set, Nearest Neighbor approaches can identify patterns present in relatively small data set also. In this research, a framework will be proposed for applying k-Nearest Neighbor (k-NN) approaches with a constant vector length for forecasting sporadic demand and the results are compared to SVM, Croston, SBA, SBJ and TSB method. Therefore, this research incorporates the benefits of using artificial intelligence in forecasting intermittent demand and compares the relative performance with the methods existing in the literature.
Group members:
Dr. Syed Mithun Ali, Associate Professor, Department of IPE, BUET, Bangladesh
Dr. Golam Kabir , Assistant Professor, Industrial Systems Engineering, University of Regina, Canada
Dr. Sanjoy Paul, Senior Lecturer, University of Technology, Sydney, Australia
Nazmul Hasan, Lecturer, Department of IPE, BUET, Bangladesh
Abstract:
In this research, we propose a methodology of efficient k-means algorithm for both point data and interval data in clustering large datasets or big data. Big data if used properly can bring huge benefits to the scientific research, business, and social sciences. Big data analysis needs a combination of techniques for data mining with those of machine learning. The k-means algorithm is one of the most popular algorithms, which is capable of doing this. This report proposes a clustering algorithm for point data based on deterministic approach and for interval data based on probabilistic approach. In deterministic approach, at first data are sorted into some fixed intervals. The midpoints of these intervals are calculated. Lastly, these midpoints are clustered by k-means algorithm which results in less computational time. We presented the comparisons of our proposed method with the other well-known algorithms as well as the analysis of error has been done. In our probabilistic approach, the concept of applying minimizing expected sum of distance instead of minimizing actual sum of distance for clustering the interval data has been used. This also ensures a reduced computational time. Both of the two proposed approaches enable a quick implementation of k-means algorithm without sacrificing accuracy much.
Faculty Advisor:
Dr. Kais Bin Zaman, Professor, Department of IPE, BUET.
Abstract:
Nowadays, industries are dealing with various sporadic demands, but it is very difficult to forecast these demands with the traditional methods. The product with sporadic demands is the reason for irregular inventory with an increment of holding cost. Existing non-parametric methods are not sufficient for handling the irregularities present in these demands. An artificial intelligence algorithm, Support Vector Machine (SVM), using advanced methodology of linear regression shows significant success in forecasting sporadic demand. SVM has a very high dimensional featured space. A leaner decision surface has been built up in this featured space. The decision provides a high generalization ability of machine learning. This work with SVMs reduces mean absolute deviation, mean absolute percent error and mean square error, provides a fast and simple methodology to get the forecasting data, and indirectly reduces the inventory cost of a system. The experimental data was taken from a well-known pharmaceutical company where professional’s advice has been considered accordingly.
Abstract:
In today’s competitive global market, companies must perform strategic changes to increase productivity and choose efficient ways to select the best partner. Selecting the appropriate partners is a prerequisite for gaining competitive advantages. This study focuses on facilitating the partner selection process for multi-echelon supply chain network through the application of artificial bee colony algorithm. The artificial bee colony algorithm is applied in order to find the most efficient partners to fulfill the customer demand and maximize the profit. The artificial bee colony optimization algorithm is inspired by bees' behavior in nature. The basic idea behind the ABC is to create the multi-agent system (colony of artificial bees) capable to solve difficult combinatorial optimization problems successfully. The results obtained from the proposed methodology presented in this paper outperforms the results from many traditional models in the existing literature.
Abstract:
Forecasting sporadic demand can be considered as one of the biggest challenges in supply chain management. It is very difficult to forecast the sporadic demand because of the irregularities present in this type of dataset. This ultimately reduces the overall supply chain performance. Traditional methods used to forecast sporadic demand include Simple Exponential Smoothing (SES), Croston’s method, Multiple Linear Regression (MLR), and other parametric and non-parametric methods. However, none of these methods considers the factors behind the irregularities present in sporadic datasets. Principal component analysis (PCA), an artificial intelligence algorithm, can analyze a dataset of two or more variables and observations are explained by distinct inter-correspond variables. In this paper, a framework for forecasting sporadic demand considering multiple factors of the irregularities is presented using Principal component analysis (PCA). Furthermore, a numerical illustration is provided using automotive spare parts data to demonstrate the effectiveness of the proposed model. The main ambition of the proposed PCA model is to extract relevant information from the provided dataset and provide predictive models. The proposed forecasting model with PCA is able to reduce forecasting error and forecast the sporadic demand with a higher degree of accuracy compared to other traditional methods.
Abstract:
Vehicle routing problem (VRP) is one of the most widely researched topics in the fields of transportation, distribution and logistics, mostly because of its capabilities for potential cost savings and improved service performance leading to better customer satisfaction. Nowadays, the rising concerns about global warming have forced companies to reduce their carbon emissions. In this paper, a framework for multi-objective multi-depot heterogeneous fleet green vehicle routing problem (MDHFGVRP) has been developed. The model maximizes revenue and minimizes costs, time and carbon emissions considering heterogeneous fleet. The heterogeneous fleet consists of different types of vehicles available to each depot. An efficient ant colony optimization algorithm (EACO), a population-based metaheuristic, has been applied to solve the problem. The EACO model is inspired by ant’s behaviors in nature. The proposed EACO model uses a novel approach of applying k-NN Classification with traditional ant colony optimization (ACO), which ensures more efficient solutions with better accuracy. The results obtained through the proposed EACO shows better performance than the traditional methods existing in the literature and provides improved solution quality. Therefore, improved responsiveness and simplicity are achieved through the application of EACO algorithm for solving the MDHFGVRP problem.
Abstract:
Now a day’s Mixed-Model Assembly Line Balancing is becoming more and more popular for mass production system. In mass production, huge costs, time, equipment and manpower are involved. Therefore, to reduce those cost, time, equipment and manpower, Mixed-Model Assembly Line Balancing can be very useful. In Single Model Assembly Line Balancing, only one type of product or model can be produced. But in Mixed- Model Assembly Line Balancing, multiple objects or multiple products can be produced at the same time.Therefore, mixed model assembly line balancing can provide better results in optimizing these resources. Different type of methods can be used to solve Mixed-Model Assembly Line Balancing problem. However, we have used a framework of Mixed Integer Linear Programming to solve a mixed model assembly line problem for garments industry. The proposed model using the Mixed Integer Linear programming outperforms other traditional model for optimizing the set of resources. Moreover, in this paper cost, space and cycle time have been reduced simultaneously. And, a real life illustration has been shown for better understanding for the practitioner.
Abstract:
In Bangladesh, there are a huge number of ordinary wheelchair manufacturer. But there is no local manufacturer of stair climbing wheelchair. In this project we have designed and developed stair climbing wheelchair combining our knowledge of ergonomics and mechanics for assisting the physically challenged persons.