Showing papers by "Sidharta Gautama published in 2021"

Operator's Experience-Level Classification Based on Micro-Assembly Steps for Smart Factories

Abstract: Tracking assembly lines in manufacturing to provide assistance is one of the essential requirement in Smart Industry. Nevertheless, these given assistance and guidelines should be offered to operators when needed. Otherwise, it can be deemed patronising in some cases, e.g., experienced operators may require less assistance than junior operators. Therefore, to provide tailored guidance and assistance in assembly lines, the operators' experience-level should be classified at different levels. In this paper, we introduce three scenarios to achieve the classification of operators expert levels in a real case study (micro-step time-series data from a factory assembly line). We implement a Convolutional Neural Network model for time-series classification, using 5 convolutional layers, max-pooling layers and 5 dense layers with dropout to avoid overfitting. We compare the results of our approach with the ground truth and also with other classifiers as K-nearest neighbours, Random Forest and Naive Bayes classifier. Results show an accuracy of 77 to 98% and 71 to 88% for two of considered scenarios.

A Methodology for Evidence-Based Data-Driven Decision Support in Policymaking

Abstract: Policymakers have the crucial task to develop innovative solutions to increasingly complex policy problems - sometimes referred to as ‘wicked’ problems - that lack the sense of clarity that most of the problems in science or engineering have, where a problem statement can be clearly defined. A DSS appropriate for handling ‘wicked’ problems in policymaking should help decision-makers cope with the problem's complexity, facilitate the assessing of multiple alternatives, and favor a discussion towards a common agenda. Such DSS could promote institutional efficiency and strengthen institutional integrity. Considering such requirements, we develop a methodology for the systematic exploration of the problem's solution space using expert knowledge. The application of the methodology in a specific use case suggests the methodology could be applied in a DSS for the identification of patterns and trends in policy-relevant data, the identification of possible policy configurations, the drafting of alternative scenarios based on the possible configurations, and institutional efficiency in terms of time and resources needed to formulate a solution to the policy problem.

Aggregate Planning for Multi-product Assembly Lines with Reconfigurable Cells

Abstract: This paper deals with aggregate planning of Reconfigurable Assembly Lines (RAL). The assembly line considered in this paper consists of hexagonal cells. These have multiple slots where processing modules can be inserted to perform certain operations. In addition, each cell has a single central slot where a central module can be inserted for inter-cellular and intra-cellular transportation of parts. Multiple products with different assembly sequences must be handled over multiple planning periods. An Integer Quadratic Programming (IQP) model is proposed to solve the following problems simultaneously: (i) assigning processing modules and a central module to the cells; (ii) installation of the cells and conveyors between the cells; and (iii) routing products, ensuring that availability of the resources is not exceeded. The assembly line should be reconfigured over time to adapt to possible product functionality and demand changes at minimum reconfiguration, operational and material handling costs while ensuring the demand is met within each period. The IQP model is implemented and solved for an illustrative problem and its extensions using Gurobi.

Genetic algorithm for solving coverage issues in undersea mining

Abstract: To be cost-effective, robot-based undersea mining must comply several operational constraints. Among the main constraints are the time and energy required to extract the mineral from the seabed. It is also important to reduce the wear of the joints that connect the ship on the surface with the robot crawler that does the mining on the seabed, since this not only reduces operating costs, but also lengthens the useful life of these parts which increases system security. For this reason, the least amount of twisting in these pieces is preferable, so it is advisable to reduce the number of turns or changes of direction in the trajectory of the robot that extracts the mineral. In this article, we present an algorithm to optimize Coverage Path Planning using Genetic Algorithm to produce paths with longer segments, which can be used in underwater mining and reduce the effects the mentioned turning problem. The resulting paths have on average 55% less changes of directions in the trajectory than a GA with standard cost function. In addition, in tests made by placing small obstacles in a random way, 76% of useful paths were obtained and up to 59% of useful path when the obstacles were grouped into a single larger obstacle.

Analysis of Minimal Clearance and Algorithm Selection Effect on Path Planning for Autonomous Systems

Abstract: There are many path planning algorithms in the literature, with different classifications, domains of use, efficiency to find the shortest path or to make a complete coverage of the area to be studied. In the literature, we can also find evaluations of all these algorithms in terms of their performance in the search for the shortest path, execution time and comparisons between them. In this work, twelve algorithms from the literature were studied to analyze their sensibility to the number of obstacles and the clearance value between them. Data analytics methods were used to make a qualitative study of the sensibility of these algorithms to the constraints studied. For investigation of the problem, two metrics were used, the length of the generated path and the number of iterations used to find the solution. The number of iterations here refers to the number of nodes evaluated by the algorithm when searching for the target node. The results are synthesized in two tables that show the sensibility of the algorithms to the change in the constraints studied and the immunity of others, and the correlation among the algorithms, the constraints and the metrics.