Estrategia de reemplazo automático de las herramientas gastadas en un proceso de torneado duro, basada en el análisis por series temporales de las señales de emisión acústica

  1. DIAZ MANTILLA, FRANKLIN OLIVER
Supervised by:
  1. Justino Fernández Díaz Director

Defence university: Universidad de Navarra

Fecha de defensa: 22 December 2004

Committee:
  1. José Germán Giménez Ortiz Chair
  2. Elisabeth Viles Diez Secretary
  3. Iñigo Iturriza Zubillaga Committee member
  4. Javier Canales Abaitua Committee member
  5. Jose Ignacio Verdeja González Committee member

Type: Thesis

Teseo: 317168 DIALNET

Abstract

Automatic Tool Replacement Strategy for Hard Turning based on the Time Series Analysis of the Acoustic Emission Signals of the process Franklin Oliver Diaz Mantilla Technological Campus of the University of Navarra. Mechanical Engineer Department (Spain), 2004 Keywords: Tool wear, PCBN, Discriminant Function, Neural Network. In this work an on-line automatic tool wear monitoring strategy has been developed for a Hard Turning process with PCBN tools. The strategy is based on Acoustic Emission (RMS) signals collected by a sensor attached to the tool shank and consists in detecting when the tool reaches an unacceptable wear level, so that the machine numerical control can order the worn tool replacement. The tool wear plays an important role in the outbreak of irregularities in any cutting process but in the case of Hard Turning wear has a particular outstanding importance, due to presence of very high passive forces that makes it difficult to achieve the very demanding geometric tolerances of the workpieces. In order to achieve that objective, the signal is fitted to a Time Series Model so that information provided by the signal can be condensed. It has been shown that the Time series Model is valid for representing signals coming from processes with different conditions (cutting speed, geometries, tools types and tool wear level). For the automatic classification of the tool wear in new/worn states, first a Linear Discriminant Function was developed employing the Time Series Model coefficients as discriminant variables but the classification results were not good enough. Then the standardized cutting time was added to the variables set and is was seen that now the Discriminant Function was appropriate for the classification although only in cases where the cutting speed, the composition and the geometry of tool were the same as the ones employed in the Discriminant Function development. In order to obtain a better classification strategy the Discriminant Function was replaced by a Neural Network and it was seen that the new strategy was able to classify tool wear with the following advantages with respect to the Discriminant Function: a) to complement the information given by the coefficients of the Time Series Model it is in this case not necessary, b) the Time Series coefficients do not need to be depurated before being fed to the Neural Network and c) the Neural Network can be applied to signals coming from types of tools different from the ones used in the network training.