| This thesis presents the work on developing an intelligent on-line monitoring system for sheet metal stamping processes. It focuses on solving three major questions: (1) how to effectively measure and select the features from the appropriate sensor signals; (2) how to define an effective and efficient learning algorithm to acquire the knowledge from the limited training samples; and (3) how to generate an accurate and simple criterion of the decision making for on-line application.; Next, four learning techniques are employed to obtain the knowledge from the samples, resulting in developing different decision making methods, of which the nature is pattern classification and computational learning. Among them, tracking thresholds checking (TTC) techniques is the advance of the traditional limitation checking method, which provides a pair of envelopes to cover the wavelet-based principle component of the strain signal under normal condition.; Then, through studying the noise density when using support vector regression technique, a new approach to measure similarity of the signals is proposed. It helps us to construct a mathematical model to describe the strain signals and propose, which is able to compare the dynamic and real-time data.; Throughout, the techniques proposed herein are tested on the stamping operations data, collected through certain sensors in our lab and real-world application. The success rate in HMM technique reaches 96.9%, and the results of SVM and SVR are 99.44% and 99%, respectively. In comparison with the traditional techniques (i.e., ANN's success rate is around 94%), the results demonstrate that these proposed techniques possess their own characteristic being good at different purposes.; Finally, a new commercialized system is developed including software and hardware according to the results of the above work. To meet the coming requirements from the markets, the trials on expending the system to Internet-based applications are also carried out with some good results. The system has been tested and evaluated in many local companies, and the technical transform to the end users is carried out for their application. (Abstract shortened by UMI.)... |
