Aplicação de redes neurais artificiais no monitoramento da operação de dressagem
Carregando...
Arquivos
Data
2012-07-06
Autores
Grizzo, Daniela Fernanda [UNESP]
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O processo de retificação confere à peça o acabamento final, minimizando as irregularidades superficiais através de interações entre os grãos abrasivos de uma ferramenta (rebolo) e peça retificada. O desgaste do rebolo devido ao atrito entre o rebolo e a peça retificada torna a ferramenta inadequada para nova utilização, sendo necessária a realização do processo de dressagem do rebolo para remoção e ou avivamento dos grãos gastos de sua superfície de corte, de forma e deixá-lo em condições para novo uso. O presente trabalho tem como objetivo classificar a condição do rebolo durante a operação de dressagem utilizando o sinal de emissão acústica (EA) e estatísticas derivadas desse sinal, por meio de redes neurais artificiais. Nos experimentos realizados usou-se um rebolo de óxido de alumínio instalado em uma retificadora plana, um sistema de aquisição de sinais e um dressador de ponta única de diamante. O processamento digital de sinais foi obtido através do software MATLAB. Os ensaios foram realizados com diferentes graus de recobrimento e profundidade de dressagem. A partir dos dados obtidos de EA puro, calculou-se o valor médio quadrático (RMS), bem como mais duas estatísticas, as quais já foram empregadas com sucesso em trabalhos de detecção de queima, no processo de retificação. Essas estatísticas também se mostraram bons indicadores para o monitoramento da operação de dressagem. Uma rede neural perceptron multicamadas (MLP) foi utilizada com o algoritmo de aprendizado Levenberg-Marquardt, cujas entradas foram as duas estatísticas mencionadas e o valor RMS de EA. Os resultados mostram que o método empregado foi capaz de classificar as condições do rebolo no processo de dressagem, identificando o rebolo como afiado (com capacidade de corte) e rebolo se afiação (com perda de capacidade de corte), viabilizando a redução do tempo e custo dessa operação e minimizando a remoção excessiva
The grinding process gives the piece a final finish by minimizing surface irregularities through interactions between the abrasive grains of a tool (wheel) and the part to be ground. The wear of the grinding wheel due to excessive friction between the grinding wheel and ground workpiece makes the tool unsuitable for further use; it is imperative the accomplishment of the process of dressing the grinding wheel to remove or resharpen the worn grains of its surface in order to make if suitable for use again. The present study aims to classify the condition of the grinding wheel during operation using acoustic emission (AE) signal and statistics derived from this sinal through artificial neural networks. In the experiments an aluminum oxide grinding wheel installed to a surface grinding machine was used along with a data acquisition system and a single point diamond dresser. The digital processing of these data was obtained using the MATLAB software. Tests were performed with different overlap ratio and depth of cut. The root mean square value of the AE signal as well as two other statistics were obtained from the raw acoustic emission signal, which have been successfully used in grinding burn detection. These statistics were also good indicators for monitoring the dressing operation. A multilayer perceptron neural network (MLP) was used with the learning algorithm Levenberg-Marquardt, whose inputs were the statistics previously mentioned and dressing conditions. The results show that the method used was able to classify the conditions of the grinding wheel in the process of dressing, identifying the wheel as sharp (cutting capacity) and dull (with loss of cutting capacity), enabling the reduction of time and cost of operation and minimizing the excessive removal of the wheel abrasive material
The grinding process gives the piece a final finish by minimizing surface irregularities through interactions between the abrasive grains of a tool (wheel) and the part to be ground. The wear of the grinding wheel due to excessive friction between the grinding wheel and ground workpiece makes the tool unsuitable for further use; it is imperative the accomplishment of the process of dressing the grinding wheel to remove or resharpen the worn grains of its surface in order to make if suitable for use again. The present study aims to classify the condition of the grinding wheel during operation using acoustic emission (AE) signal and statistics derived from this sinal through artificial neural networks. In the experiments an aluminum oxide grinding wheel installed to a surface grinding machine was used along with a data acquisition system and a single point diamond dresser. The digital processing of these data was obtained using the MATLAB software. Tests were performed with different overlap ratio and depth of cut. The root mean square value of the AE signal as well as two other statistics were obtained from the raw acoustic emission signal, which have been successfully used in grinding burn detection. These statistics were also good indicators for monitoring the dressing operation. A multilayer perceptron neural network (MLP) was used with the learning algorithm Levenberg-Marquardt, whose inputs were the statistics previously mentioned and dressing conditions. The results show that the method used was able to classify the conditions of the grinding wheel in the process of dressing, identifying the wheel as sharp (cutting capacity) and dull (with loss of cutting capacity), enabling the reduction of time and cost of operation and minimizing the excessive removal of the wheel abrasive material
Descrição
Palavras-chave
Retificação e polimento, Acabamento, Rebolos, Redes neurais (Computação), Emissão acústica, Grinding and polishing, Finishes and finishing, Grinding wheels, Neural networks (Computer science), Acoustic emission
Como citar
GRIZZO, Daniela Fernanda. Aplicação de redes neurais artificiais no monitoramento da operação de dressagem. 2012. 122 f. Dissertação (mestrado) - Universidade Estadual Paulista, Faculdade de Engenharia, 2012.