|
The study aims at introducing a novel hybrid optimization algorithm (HOA), incorporating three different types of optimization methods, namely, genetic algorithm (GA), artificial neural network (ANN) and mathematical programming (MP), for real complex engineering design problems. The underlying idea of the proposed HOA is to take advantage of the superior features of these three different optimization algorithms while easing their drawbacks, such as, the lack of an effective termination criterion in GA. In the proposed HOA, the GA is responsible for not only evolving the population toward better fitness value but also, based on the newly-evolved populations or feasible design points at each GA generation, for continuously updating the proposed ANN mathematical model for better approximation of the objective and constraint functions. The ANN technique here is used to construct the approximate macro mathematical model or neural network model of the desired objective and constraint function. In the ANN evolution using backpropagation neural network (BPN) algorithm, the feasible design points obtained from each GA generation are considered as example pairs for training and testing the ANN model. The training would continue until the root mean square (RMS) error between the network&;#39;s output and the target value over all the example pairs is minimized. For each or every few GA generations, the newly-updated neural network models, representing the approximate objective and constraint functions, are further used to construct the optimization sub-problem. The solution of the optimization sub-problem is sought through a mathematical programming model using generalized reduced gradient (GRG) algorithm. As the optimization proceeds, a sequence of approximate solutions associated with the continuously-updated ANN models is derived. The iterative process continues until the convergence of the approximate solutions is attained. To deal with the multi-criteria and constrained optimization problems, composite objective formulation (also called weighting method) and exterior penalty method (EPM) are employed in the present HOA, respectively. Besides, several different hybrid design procedures and mutli-criteria design models are also proposed.
To determine the effectiveness of the proposed algorithm, several nonlinear programming test problems are used, in which the calculated results are compared with those of a GA and an MP algorithm, and also with the literature data. At last, the applicability of the proposed HOA is demonstrated through design optimization of a real complex engineering design problem, i.e., the design optimization of the process-induced thermal-mechanical behaviors of an anisotropic conductive film (ACF)-based ultra-thin chip on film (UTCOF) interconnect technology during bonding process. This is a multi-criteria optimization problem, in which the design objective is to seek minimization of the process-induced warpage of the silicon chip and the peeling stress at the ACF/chip interface and maximization of the contact stress at the ACF joints. Results show that the proposed HOA can be applicable for not only the ill-posed but also constrained and multi-criteria optimization problems. Furthermore, the developed algorithm can provide good optimal solutions with much less computational effort, as compared to the GA and MP method, where a larger scale of design problems would yield a more significant improvement in the computational efficiency.
|