| To produce quality products for the purpose of coping with the global manufacturing competition, it is essential to automate the design practice, diagnose faults, and control the process. In this study, a generic scheme to establish the norms for automation of design, diagnostics, and control by employing fuzzy logic and neural networks for continuous processes is proposed. First, design of a grinding process is accomplished by initial determination of a set of optimal design variables in order to achieve a set of desired process variables. Next, process diagnostics is identified as a pattern matching task so that it can be processed by neural networks/fuzzy systems to match sensor readings of system parameters to a pre-defined abnormal scenario. The research results are applied to the real life example of the TMI-2 nuclear reactor. Then, a hierarchical control structure is proposed for a turning process. By utilizing the adaptation/learning ability of both fuzzy logic and neural nets, the control algorithm automatically tunes PID (Proportional-Integral-Derivative) gains for the SISO (Single-Input-Single-Out) turning system. That is, the control action is carried out through the manipulation of the feedrate in order to maintain the cutting force at a reference point. Moreover, a generic scheme has been proposed to serve as the first step towards full integration of neuro-fuzzy systems; not only fuzzy inference has been applied to speed up the learning process of neural nets, but also the advantage of the learning ability of neural nets has been taken to fine tune the fuzzy decision table. Performance evaluation and comparison between neural networks and fuzzy logic theories are analyzed so that conclusions can be drawn about pros and cons between neural and fuzzy systems. In short, the main contribution of this study is to explore the potential usefulness and fitness of fuzzy logic and neural networks in the domain of manufacturing. In addition, the analytical tools developed in this study are generalizable for construction of other larger scale, complex systems. The applicability and validity of this study will be demonstrated via the simulation of designing a surface grinding model, of diagnosing the TMI-2 nuclear reactor, and of controlling an experimental turning model, respectively. |
