Thesis defense of Aitor Duo
Thesis defense of Aitor Duo
Thesis defense of Aitor Duo
Title of the thesis: "Sensor and CNC internal signal evaluation to detect tool and workpiece malfunctions in the drilling process". Obtained the SOBRESALIENTE CUM LAUDE qualification and has received the ‘Doctor Internacional’ mention.
- Title of the thesis: "Sensor and CNC internal signal evaluation to detect tool and workpiece malfunctions in the drilling process".
- Court:
- President: Garret Edward O Donell (Trinity College)
- Vocal: Dra. Maite Beamurgia Bengoa (Fagor AOTEK S. Coop.)
- Vocal: Dr. German Terrazas Angulo (University of Cambridge)
- Vocal: Dr. Joaquín Barreiro García (Universidad de León)
- Secretary: Dr. Mikel Cuesta Zabaljauregui (Mondragon Unibertsitatea)
Abstract
To boost competitiveness and meet changing customer demands, the manufacturing sector is taking advantage of Information and Communication Technologies (ICT). Machining is no exception, and machining processes are moving toward a more intelligent and connected network to become part of an industrial digital ecosystem.
Despite the advances made to date, there are still considerable opportunities for improvement because of the complexity of machining processes. In this context, extracting and analysing data from machining operations can provide valuable information to predict undesirable aspects, and take actions to reduce or prevent them.
The machining process taken as the focus of the present work is drilling. Drilling is one of the most commonly used and critical machining operations on many machined components. It is carried out in the last stages of product manufacture, where a mistake can result in a defective part. In this thesis, a comparison and selection of the sensors with the best prediction capacity of tool condition and surface roughness is carried out for the development of data driven models that predict the mentioned parameters of the drilling process.
Various sensors were installed on a drilling machine, as well as internal machine signals, to take series of physical measurements of the tool condition and the machined component. The resulting data determines relationships for the creation of predictive models to identify errors that may have occurred in the drilling operation based on acquired signals.
Through statistical analysis (t-test) of the results obtained from the data-driven models, insight was gained into the predictive capability of each sensor. The most viable ones for tool condition monitoring systems were then selected.
The features of the signals that best adapt to specific surface finish were established. Based on a series of random measurements of the machined surface roughness, a methodology was developed to map the signal features that best suit the roughness distribution of the machined holes. By using hierarchical clustering and principal component analysis of the mapped signal features, clusters are created to identify holes with damaged surfaces.
The adaptability of machining process monitoring systems to various input parameters is a fundamental challenge for the automatic reconfiguration of such systems. For this reason the dimensions of the features obtained were reduced to two dimensions using principal component analysis as t-distributed stochastic neighbour embedding to be able to better identify the input parameters.