Indirect multisignal monitoring and diagnosis of drill wear: Dissertation

A machine tool utilisation rate can be improved by an advanced condition monitoring system using modern sensor and signal processing techniques. A drilling test and analysis program for indirect tool wear measurement forms the basis of this thesis. For monitoring the drill wear a number of monitoring methods such as vibration, acoustic emission, sound, spindle power and axial force were tested. The signals were analysed in the time domain using statistical methods such as root mean square (rms) value and maximum. The signals were further analysed using Fast Fourier Transform (FFT) to determine their frequency contents. The effectiveness of the best sensors and analysis methods for predicting the remaining lifetime of a tool in use has been defined. The results show that vibration, sound and acoustic emission measurements are more reliable for tool wear monitoring than the most commonly used measurements of power consumption, current and force. The relationships between analysed signals and tool wear form a basis for the diagnosis system. Higher order polynomial regression functions with a limited number of terms have been developed and used to mimic drill wear development and monitoring parameters that follow this trend. Regression analysis solves the problem of how to save measuring data for a number of tools so as to follow the trend of the measuring signal; it also makes it possible to give a prognosis of the remaining lifetime of the drill. A simplified dynamic model has been developed to gain a better understanding of why certain monitoring methods work better than others. The simulation model also serves the testing of the developed automatic diagnostic method, which is based on the use of simplified fuzzy logic. The simplified fuzzy approach makes it possible to combine a number of measuring parameters and thus improves the reliability of diagnosis. In order to facilitate the handling of varying drilling conditions and work piece materials, the use of neural networks has been introduced in the developed approach. The scientific contribution of the thesis can be summarised as the development of an automatically adaptive diagnostic tool for drill wear detection. The new approach is based on the use of simplified fuzzy logic and higher order polynomial regression analysis, and it relies on monitoring methods that have been tested in this thesis. The diagnosis program does not require a lot of memory or processing power and consequently is capable of handling a great number of tools in a machining centre.