| Automation of manufacturing processes requires reliable sensing and control of critical parameters. In this research a novel approach using a single semiconductor strain gage as a multivariable sensor is presented to detect tool wear levels and levels of self-induced vibrations (chatter). Machining experiments on a bench-top lathe for different depths of cut, spindle speed, feed rates and tool wear levels on carbon steel specimens generated an extensive database of strain gage signals. The signals were filtered into three frequency bands to obtain indicators related to temperature, vibration, and stress waves. At low frequencies (<10 Hz), the signal is related to temperature changes. Medium frequencies (5 Hz--15 kHz) contain information related to forces generated by machining and mechanical vibrations. With proper gage sizing, high frequencies (up to 200 kHz) detect stress waves or acoustic emissions generated at the machining source. The data shows that tool wear levels can be detected in the medium frequency range near 12 kHz by evaluating the maximum amplitudes and corresponding frequencies of the signals. The detection is consistent for all the combinations of machining parameters used. Moreover, the detection is consistent when both aluminum and brass specimens were machined. The force-related medium frequency signals also exhibited large amplitudes when self-induced vibrations (chatter) were present. The variances of these signals were employed as a variable related to the stability of the machining. The results of several machining experiments using step and ramp changes in the depth of cut to generate self-induced vibrations confirmed the reliability of this indicator. Based upon these observations, designs are presented for a tool wear monitoring system using a three-layer feedforward neural network and for the control of self-induced vibration (chatter) based on a Mamdani fuzzy inference controller. Subsequent experiments confirm the success of these designs. |
