To improve the geometric accuracy of CNC machined parts, dynamic machining errors due to on-line disturbances (tool deflection, tool wear, heat deformation, etc.) should be accounted for in some manner. Unless these on-line disturbances are properly handled, it is obvious that a high degree of geometric accuracy is difficult to achieve. Many attempts have been made to compensate for these on-line disturbances such as the development of engineering models; however, the models are not adequate enough for reflecting the real phenomenon and are dependent on continuous process monitoring using a variety of sensors. Closed-loop process control is a scheme for manufacturing parts and compensating for on-line disturbances and machine tool inaccuracies using error feedback. The goal has been to develop a system that automatically provides dimensional error feedback to the process machine. Closed-loop process control can be achieved before, during (in-process) or after the machining cycle. In-process control is achieved by measuring the part prior to finishing cuts while the part is fixtured to the machine tool. Although the theory behind an automated closed-loop, in-process control system would significantly reduce manufacturing costs, at the present time, machining errors typically are compensated through manual error feedback. This thesis presents a systematic approach for automatically compensating for dynamic machining errors based on a new closed-looped machining scheme. The new scheme incorporates these errors, found through in-process inspection, into a modified CAD model or "Variational Part Model". As a result, the Variational Part Model inherently contains the online disturbances associated with machining. It is important to note that this new scheme assumes the machine tool's static error (ball screw error, joint misalignment, perpendicularity error, etc.) has been addressed by some other compensating method and this scheme only addresses the dynamic machining error. To create the Variational Part Model, the machined part is measured on the machine and compared to the CAD model's theoretical data. The data is then used in conjunction with modeling functions contained in NX's Application Programming Interface (API) to interact with the CAD model and modify its feature geometry. The validity and effectiveness of the methods are presented as well as results from experimental testing. This thesis also presents the methods necessary for automatic CAM process updating to ultimately close the loop between machining and inspection.

A method for precisely synchronizing an external serial data stream to the pulse-per-second (PPS) output signal from a global navigation satellite-based system (GNSS) receiver was investigated. A signal timing device was designed that used a digital signal processor (DSP) with serial inputs and input captures to generate time stamps for asynchronous serial data based on an 58593.75 Hz internal timer. All temporal measurements were made directly in hardware to eliminate software latency. The resolution of the system was 17.1 µs, which translated to less than one millimeter of horizontal position error at travel speeds typical of most agricultural operations. The dynamic error of a TTS was determined using a rotary test fixture. Tests were performed at angular velocities ranging from 0 to 3.72 rad/s and a radius of 0.635 m. Average latency from the TTS was shown to be consistently near 0.252 s for all angular velocities and less variable when using a reflector based machine target versus a prism target. Sight distance from the target to the TTS was shown to have very little effect on accuracy between 4 and 30 m. The TTS was determined to be a limited as a position reference for dynamic GNSS and vehicle auto-guidance testing based on angular velocity. The dynamic error of a GNSS receiver was determined using the rotary test fixture and modeled as discrete probability density functions for varying angular velocities and filter levels. GNSS position and fixture data were recorded for angular velocities of 0.824, 1.423, 2.018, 2.618, and 3.222 rad/s at a 1 m radius. Filter levels were adjusted to four available settings including; no filter, normal filter, high filter, and max filter. Each data set contained 4 hours of continuous operation and was replicated three times. Results showed that higher angular velocities increased the variability of the distribution of error while not having a significant effect on average error. The distribution of error tended to change from normal distributions at lower angular velocities to uniform distributions at higher angular velocities. Internal filtering was shown to consistently increase dynamic error for all angular velocities.

Kyoto University (京都大学) / 0048 / 新制・課程博士 / 博士(工学) / 甲第16841号 / 工博第3562号 / 新制||工||1538(附属図書館) / 29516 / 京都大学大学院工学研究科マイクロエンジニアリング専攻 / (主査)教授 松原 厚, 教授 松久 寛, 教授 西脇 眞二 / 学位規則第4条第1項該当