City University of Hong Kong

CityU Institutional Repository >
3_CityU Electronic Theses and Dissertations >
ETD - Dept. of Manufacturing Engineering and Engineering Management  >
MEEM - Doctor of Philosophy  >

Please use this identifier to cite or link to this item:

Title: Hybrid spectral/intelligent modeling, control and diagnosis for nonlinear distributed curing processes
Other Titles: Ji yu pu fang fa de fei xian xing fen bu can shu gu hua guo cheng zhi neng jian mo, kong zhi ji zhen duan
基於譜方法的非線性分佈參數固化過程智能建模, 控制及診斷
Authors: Deng, Hua (鄧華)
Department: Dept. of Manufacturing Engineering and Engineering Management
Degree: Doctor of Philosophy
Issue Date: 2005
Publisher: City University of Hong Kong
Subjects: Microelectronic packaging
Notes: CityU Call Number: TK7874.D45 2005
Includes bibliographical references (leaves 169-181)
Thesis (Ph.D.)--City University of Hong Kong, 2005
ix, 183 leaves : ill. ; 30 cm.
Type: Thesis
Abstract: To produce high quality semiconductor devices, the more and more complex processes are used in the semiconductor back-end packaging industry. The cure process is a typical example that provides a required temperature distribution. After die-bonding, the adhesive bonded dies with leadframes are moved into a curing oven to cure at a prespecified temperature. It is critical to maintain the temperature at every place of the object to be cured within the specification because of cure quality and throughput. However, one serious problem in the curing process is that it is difficult to maintain consistent performance by traditional low-level feedback controllers (e.g. PID controllers). The curing process belongs to distributed parameter systems (DPSs) described by partial differential equations (PDEs). Control of this thermal process is challenging in semiconductor-packaging environment due to following reasons: 1. The process is not exactly known and has time-space coupling, nonlinear and multi-input/multi-output (MIMO) features. 2. Lacking on-line measurement of the temperature distribution and only finite sensors available result in extra difficulty in maintaining a required temperature distribution. 3. There is no systematic approach to design cure schedules for an optimal production with high quality bonds between a die and a leadframe. A cure schedule is usually determined in a trial and error process in practice even though it is costly and may not guarantee the reliability of the adhesive die attach. 4. Lacking on-line sensing techniques for cure content that is critical to the strength of the bond produces difficulty in monitoring the curing process. In view of the above problems and little research work available for the distributed curing process, the objective of this thesis is to provide a fundamental, practical and somewhat complete investigation in modelling, control, fault diagnoses and cure optimization for the process. The investigation aims to help people understand and solve some of the fundamental problems encountered during curing. The developed methodologies are expected to be engineering-preferred and be applicable to a wide range of industrial processes including curing. The developed spectral approximation based intelligent modelling is the core of the thesis. To characterize the nonlinear distributed curing process, the first principle model of the process described by parabolic PDEs is derived by physical laws with unknown nonlinearities and uncertainties. With the help of spectral methods and partial knowledge of the process, the dynamics of the curing process derived from physical laws can be approximated by a model of low-order nonlinear ordinary differential equations (ODEs) with a few uncertain parameters and unknown nonlinearities. A hybrid general regression neural network (NN) is then trained to be a nonlinear model of the process in state-space formulation, which is suitable for further system analyses and control design. Generally, it is difficult to control an unknown nonlinear MIMO process described by PDEs such as the distributed curing process. It is worth noting that an input-output model is usually insufficient for control of a distributed parameter system because finite measurements can not provide sufficient space information. For instance, in the curing process, a state space model is required to estimate the temperature distribution in the oven chamber. Using the hybrid spectral and neural model in state space form built for the curing process, an NN approximate decoupling control methodology is then developed. Based on an innovative approximation of NN nonlinear models, the neural control law can be derived directly with straightforward implementation for the curing process. To guarantee cure quality and productivity for the curing process, it is necessary to deal with uncertainties during curing control. An approximate internal model based neural control strategy is proposed for the curing process subject to model mismatch and disturbances. The proposed neural control strategy combines the advantages of both inverse control and internal model control (IMC) and has clear advantages than the existing neural IMC methods. Since faults may change the availability of the actuators and sensors or the process dynamics severely, fault diagnosis is very important for maintaining good quality of curing devices. A fault detection and isolation (FDI) strategy that combines NN models and FDI observers is proposed for the FDI of the curing process. By linearizing the NN model at some working points, FDI observers are then designed and used to detect and isolate faults. To reduce unnecessary computation, the existence conditions of FDI observers that can be checked at the beginning of the observer design are proposed and an effective design algorithm for FDI observers is also developed. Frequently, it is not enough to have only the location of detected faults to make appropriate decisions in practice. A new NN fault reconstruction scheme is proposed for the nonlinear curing process to identify the sizes of detected faults based on the extended Kalman Observer (EKO) and the NN model of the process. The thesis also provides a novel model-based integration of decoupling control and supervision for the curing process to maintain both reliability and throughput. The optimal cure schedule can be systematically determined from the model of the process and the cure kinetics of the adhesives used. The method is straightforward and effective, and can be easily applied to the curing supervision. Such a system-wide integration of control and supervision can be utilized to replace the traditionally used unreliable trial and error process, and will provide an optimal production that is able to adapt to varying operating conditions.
Online Catalog Link:
Appears in Collections:MEEM - Doctor of Philosophy

Files in This Item:

File Description SizeFormat
fulltext.html159 BHTMLView/Open
abstract.html159 BHTMLView/Open

Items in CityU IR are protected by copyright, with all rights reserved, unless otherwise indicated.


Valid XHTML 1.0!
DSpace Software © 2013 CityU Library - Send feedback to Library Systems
Privacy Policy · Copyright · Disclaimer