Multi-Objective Optimization & Design Engineering
RESEARCH THEME
Purpose of Design Engineering is to make system that make good products. Used to be, they were competitions to raise functions. However, we need to find users wants and make good concepts. Our goals are making methods to realize them.
Design Engineering
Multi-Objective Optimization
Design Thinking
Diagnosis
Multi-Objective PSO driven by DEA
“Multi-Objective PSO driven by DEA” is a method that combines Multi-Objective Particle Swarm Optimization (PSO) with Data Envelopment Analysis (DEA).
In this approach, the PSO algorithm is used to address problems with multiple optimization objectives, while DEA is employed to assess and enhance the solution space. PSO algorithm simulates the foraging behavior of birds to search for the optimal solution, while DEA is used to evaluate and optimize the efficiency and effectiveness of solutions. By combining these two methods, it becomes possible to effectively address problems with multiple optimization objectives and identify the Pareto optimal solution set.
Development of Surrogate Optimization Driven by PCA-RBFN
The aim of this study is to improve the efficiency of finding optimal solutions for large-scale problems. We conducted experiments using Golinski’s Speed Reducer (GSR) benchmark problem, which has 7 design variables.We used Principal Component Analysis (PCA) to reduce the dimensionality of the problem to 3 design variables and employed them to train a Radial Basis Function Network (RBFN) for function approximation.Subsequently, we utilized Particle Swarm Optimization (PSO) to search for optimal values. The PSO-discovered values were added to the set of centroids of the RBFN. We retrained the RBFN to adjust the approximation function and repeated this process until a satisfactory value was found.
In this experiment, we used 100 sets of data. Each dataset included design variables, corresponding function values, and constraints on the design variables. Due to the loss of information caused by PCA, the error of the approximation function became large (approximately 106) when using the data to train RBFN.However, when applying this function to PSO, we were able to find the design variables of the optimumvalue(2994.35), even though the predicted value was around (2653.54).