Robust control of surface roughness in a turning operation

Tool wear measure is very important in turning operations due to the adverse effects of tool wear on machined parts. In this thesis, a novel way of gauging tool wear is achieved with relation to the surface roughness, using micro-optics and image analysis algorithms, which provides better accuracy and reliability in the measurement of the tool wear. Tool flank and nose wear areas and the center of gravity are formulated into a tool wear index (TWI), which shows a clear relationship with the surface roughness. Compared to the conventional measure of tool wear (i.e., flank wear land width), TWI measures tool wear more comprehensively and directly, hence presents the potential for a better control of surface roughness.;Using TWI, a surface roughness control model is developed. Additionally, a tool life model is developed in an effort to extend the tool usage and minimize the in-process tool failure, which measures the material loss from the tool face and delimits the tool usage once the material loss reaches the predetermined value. An optimal control strategy is derived based on the tool life and surface roughness model. Rather than using a single feed rate and discarding the tool, the control strategy permits adjustment of machining parameters (feed rate) to maximize tool usage, while satisfying the surface roughness within the specifications and minimizing the tool failure, in disparate, successive finishing operations. An example simulation shows that the strategy can reduce production time and production cost in a batch production mode.;In order to address the limitations of empirical equations for the control of surface roughness, a fuzzy adaptive modeling of surface roughness is developed. In today's rapidly changing manufacturing environment, slight variations to the process may be inevitable and frequent. The proposed modeling method incorporates the variations into a fuzzy rule base and approximates the gradients of the surface roughness curve after process variations are introduced. The comparison of three sets of experimental data shows good agreement with the proposed modeling method, which suggests that the proposed method could be implemented into a CNC controller where process variations are expected.