Search

This paper presents the application of an artificial neural network (ANN) to develop a model able to predict the holdup of a two-phase flow composed of water and mineral oil in a horizontal pipe. The surface velocities of each fluid and the pressure differential in the pipe are used as input parameters of the multi-layer artificial neural network with backpropagation, while the volumetric fraction of the fluids is used as an output parameter in the training. The 56 experimental data were obtained in the laboratory LabPetro-CEPETRO-UNICAMP. The results of the predictive model with the best performance show a mean absolute error (AAPE) of 3.01 % and a coefficient of determination R² of 0.9964, using 15 neurons in the hidden layer of the network and the TanSig transfer function. [ABSTRACT FROM AUTHOR]

Companies engaged in the manufacture, marketing and maintenance of cylinders for liquefied petroleum gas in Colombia, stamped steel plates and welded to the product a unique serial code to be identified within the cylinder of the country park. Currently, the identification process is manual and checked approximately 7000 cylinders per day in a single factory. The main objective of this paper is to present a vision system that uses artificial neural networks to recognize the code. This system consists physically of a portable device that controls light environment and scene for the acquisition of images. Another component of the system is to adjust the image. The adjustment is based on median filtering, binarization, label, and segmentation, this processing allows more meaningful information and image discrimination. Finally, the intelligent component identification is performed with Hopfield neural networks and an algorithm that checks the development of image recognition. The effectiveness of the system was reported with experimental results obtained on the basis of error with a significant number of samples. [ABSTRACT FROM AUTHOR]

Artificial neural networks are an important technique in nonlinear time series forecasting. However, training of neural networks is a difficult task, because of the presence of many local optimal points and the irregularity of the error surface. In this context, it is very easy to obtain under-fitted or over-fitted forecasting models without forecasting power. Thus, researchers and practitioner need to have criteria for detecting this class of problems. In this paper, we demonstrate that the use of well known methodologies in linear time series forecasting, such as the Box-Jenkins methodology or exponential smoothing models, are valuable tools for detecting bad specified neural network models. [ABSTRACT FROM AUTHOR]

This paper described the problem of path-planning in mobile robots in a partially structured environment. The work was divided in three parts: design of a mechatronic system, denominated Robot IO, study of advantages and disadvantages of three path-planning algorithms: neural nets, genetic algorithms and fuzzy logic, and implementation of algorithms that allow to identify the workspace where IO moves. [ABSTRACT FROM AUTHOR]