Intelligent modeling and control of chemical processes for manufacturing composite materials

摘要

This dissertation aims at systematically establishing intelligent paradigms for modeling and controlling the bag-molding process for manufacturing composite materials by means of fuzzy logic, neural networks, and adaptive wavelet analysis.;In part I, a sensor system has been constructed by integrating a dual heat-flux sensor serving as a hard sensor and a recurrent neural network (RNN) serving as a soft sensor for monitoring the bag-molding process. The hard sensor determines the Damkohler number (Da) while the soft sensor predicts the degree of cure (DOC) in response to the Da evaluated by the hard sensor. A model based on an artificial neural network (ANN) has been constructed for predicting the resin contents of the final composite products. A model-predictive control system derived from the constructed model has been proposed for quality assurance of such products. An inverse ANN model has also been conceived to learn the dynamic behavior of the process.;In Part II, a novel system, wavelet-transform-neural network (WTNN), has been constructed. The WTNN has been demonstrated to be effective in identifying quantitatively the behavior of a complex nonlinear chemical process by estimating the exit age distribution of a non-ideal flow reactor. The WTNN has also been applied to the process trend analysis. A system is constructed by integrating the WTNN subsystem for identifying the trend of a process with a fuzzy-logic subsystem for controlling the process. The experimental results have unequivocally shown that this system is capable of effectively preventing thermal runaway. Moreover, the uniformity of cure of the composite part has been substantially improved.;The applications of wavelet transform has been studied in Part III. It has been found that the time-scale analysis provides a rigorous framework to determine features of a complex signal in both time and frequency domains. Part III has also demonstrated that the wavelet shrinkage algorithm is efficient in denoising a highly corrupted signal and preserving the critical features of the signal.;Part IV presents the significant conclusions. In addition, it recommends the possible extensions to this dissertation.