Implementación de un sistema de visión artificial en líneas de producción de atún, para la detección de fallas de codificado en la empresa Puertomar S.A, Provincia de Manabí.

Abstract

In this project it was proposed: the design of an artificial vision system capable of determining the coding faults in the tuna cans in real time based on the digital image processing using filters and algorithms through an artificial neural network model convolutional, implemented through the Python programming language and the use of the free software library called TensorFlow, which together allow us to identify if the can of tuna is encoded in a correct way, or on the contrary if it presents any type of problem. It is necessary to carry out fault diagnosis in real time, reaching not only its detection and isolation but also the estimation of its magnitude, with the aim of interrupting the propagation of faults in a given system. In the case of having to detect failures in complex processes, where there is parametric and structural uncertainty such as the case study that it occupies (tuna production line), there are no mathematical models of it, so artificial intelligence techniques are required. As a result of this project, a convolutional neural network with an efficiency of 84% was trained using TensorFlow libraries and with the help of the Convolutional model (Neocognitron Mobilenet_v1_1.0_224), for which a data set of 417 images was used, and was trained by 4000 iteration steps, with the implementation of the artificial mink system the company improved its efficiency by 84% within the tuna cans coding line, thus generating a substantial profit in its production lines. Finally, it is recommended to train the model with a larger data set in order to improve the efficiency of the neural network, and achieve an effectiveness greater than 95%.