Multivariate tool condition monitoring in a metal cutting operation using neural networks
Cast your vote
You can rate an item by clicking the amount of stars they wish to award to this item.
When enough users have cast their vote on this item, the average rating will also be shown.
Your vote was cast
Thank you for your feedback
Thank you for your feedback
AuthorsDimla, Dimla E., Snr
MetadataShow full item record
AbstractThis thesis gives an account of an investigative study to develop a modular Tool Condition Monitoring System (TCMS) for metal cutting operations (turning in particular) through the application of multivariate process parameters and perceptron neural networks. The initial phase of the work was concerned with a literature survey that was conducted to investigate the extent of applicability of neural networks to TCMS to-date. The survey showed that in spite of well over a decade of research in this direction, a truly universally applicable TCMS has not yet been developed. A test bed to generate cutting data using a simple centred lathe onto which were attached the necessary sensors (a triaxial accelerometer and a triaxial dynamometer) was set-up. The instrumentation for the test bed comprised a personal computer fitted with a data acquisition card and connected to various peripheral signal conditioning instruments. After validating the test rig against an existing result set, a number of limited cutting test cuts were carried out to assess the impact of cutting conditions on signal behaviour for both worn and sharp plane-faced tool inserts. The acquired cutting data was used to assess the ability of a neural network in tool wear diagnosis. The application of a linear neural network model in the form of a Single Layer Perceptron (SLP) could not satisfactorily classify the cutting data into the two tool state classes desired. The non-linear nature of the cutting data led to the application of Multi-Layer Perceptron (MLP) networks that were better at handling such non-linearity. However, when the MLP neural network was trained and tested on cutting data, the nature of the results lead to the conclusion that changes in chip form during machining resulted in false alarms. The propensity for such false alarms was reduced by using tool inserts with controlled tool-chip contact (i.e. chip breaker geometry). The success of an initial system utilising flank wear only as the tool wear indicator to effect classification was extended. Nose wear and flank wear were used together but limiting the desired tool states to worn and sharp. This was further extended to include part worn, chipped and/or fractured tools. Finally, the developed model's range of application in a turning process was performed and its ability to recognise and distinguish tool wear from changes in the cutting conditions assessed. Though the developed modular TCMS model was capable of reasonable tool wear classification, it was incapable of detecting changes in the cutting conditions (particularly a reduction or an increase in the area of cut). A different classifier was developed to be used in parallel with the TCMS model in order for the system to be applicable over a broad range of cutting conditions.
PublisherUniversity of Wolverhampton
TypeThesis or dissertation
DescriptionA thesis submitted in partial fulfilment of the requirements of the University of Wolverhampton for the degree of Doctor of Philosophy
Except where otherwise noted, this item's license is described as https://creativecommons.org/licenses/by-nc-nd/4.0/