An intelligent cutting tool conditioning monitoring system for milling operation

  • Pan Fu

    Student thesis: Doctoral Thesis

    Abstract

    A very important requirement of modern machining systems in an 'unmanned' factory is to change tools that have been subjected to wear or damage in time. The old tool change strategies are based on conservative estimates from the past tool wear data, hence tools can be replaced too early or too late. This can increase the production cost or endanger the quality of the products. An integrated tool condition monitoring system composed of multi-sensors, signal processing methodology and decision-making plans is a crucial requirement for automatic manufacturing processes. An intelligent tool condition monitoring system for milling operations will be introduced in this report. The system is composed of four kinks of sensors, signal transforming and sampling apparatus and a microcomputer. By using intelligent pattern recognition techniques, different sensor signals are combined and the tool wear states can be recognised reliably. The efficient and effective orthogonal experimental design procedure is applied to comprehensively verify the monitoring system in a limited number of test runs. 50 signal features are extracted from time and frequency domain and they are found to be related to the development of tool wear values. A fuzzy clustering feature filter has been developed to remove less tool wear relevant features under different cutting conditions. multi-sensor signals reflect tool condition comprehensively and the sensor fusion strategy is used to provide reliable recognition results. Combining fuzzy approaching degree and fuzzy closeness provides a unique and overall fuzzy similarity index, the two-dimensional fuzzy approaching degree. A new type of fuzzy system, the fuzzy driven neural network has been established. The network can assign signal features suitable weights to make the tool wear state recognition process more accurate and robust. The advanced B-spline neurofuzzy networks are also successfully applied in the tool condition monitoring process. This powerful modelling system is established by combining the qualitative fuzzy rule representation with the quantitative adaptive numeric processing process. The fuzzy driven neural network and the B-spline neurofuzzy network can then be combined to build a neurofuzzy hybrid pattern recognition system, which is more reliable and accurate. Armed with the well- developed pattern recognition methodology, the established intelligent tool condition monitoring system has the advantages of being suitable for a wide range of machining conditions, robust to noise and tolerant to faults. As can be seen in the thesis, several innovations have been made in the research process of this project. The fuzzy clustering feature filter can significantly improve the efficiency and reliability of the tool wear state recognition process. The two-dimensional fuzzy approaching degree comprehensively characterises the similarity between two fuzzy sets. The fuzzy driven neural network indirectly solves the weight assignment problem of the conventional fuzzy system. The established neurofuzzy hybrid pattern recognition system obviously improves the system's recognition resolution and reliability
    Date of Award2000
    Original languageEnglish
    Awarding Institution
    • Nottingham Trent University

    Cite this

    '