Dynamic modeling of manufacturing process error patterns using distributed adaptive systems

Hung-Kang Jan, Purdue University

Abstract

Implementing "manufacturing for quality" in real time while retaining cost efficiency in an automatic machining processes has been a very challenging task. The difficulties arise from the fact that the development of errors in a computer controlled machine system is a time-varying nonlinear process and there is no effective method to gauge the machine spatial errors while the product is machined. This research develop a hybrid methodology for dynamic modelling of low frequency domain spatial errors of CNC machine systems. The objective is to perform real-time, in-process quality control for computer controlled precision machining processes. The problem is approached by predicting the machine system errors at bounded arbitrary coordinates and operation settings, using a hybrid model consisting of both neural network and analytical approaches. The neural network is used to track the time-varying thermal errors, while kinematic and blending functions are used to construct 3D spatial error maps. The hybrid approach is able to modify the model parameters adaptively as the machine conditions change. A dedicated microcomputer-based multiple channels temperature and displacement measurement system is developed to (1) collect data in the off-line modeling phase for training the neural network and for constructing the spatial errors model and (2) provide real-time, in-process system conditions data at arbitrary time intervals in the application phase. The output of the trained neural network is used for machine performance evaluation, error tracking and controller compensation. Results of experimentation demonstrate that a neural network based hybrid model can be very effective in predicting machine time-varying errors with predictive accuracy close to the machine system's random error. In application, this approach does not interfere with the machine processes at all and remains robust in an electrical noise corrupted environment. The numerical results of the model represent current machine accuracy and can be reported at any time to a human supervisor for judging whether the machine performance and product quality are still acceptable or not. Integrated with the CNC machine controller, the outputs of the hybrid model can compensate for machine inaccuracy automatically and thus improve product quality in real time.

Degree

Ph.D.

Advisors

Liu, Purdue University.

Subject Area

Industrial engineering|Mechanical engineering|Systems design

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS