Explainable AI Assisted Evolutionary Search of Engineering Designs

Evolutionary algorithms are widely utilized to dis-cover optimal and feasible solutions for NP-hard problems. However, the factors influencing a solution's performance, such as objective values, feasibility, or infeasibility, often lack ex-planation. This paper presents a novel Explainable AI (XAI)-assisted evolutionary search approach that utilizes evolutionary data to learn problem characteristics and guide the search process, thereby enhancing the quality of evolved solutions. The proposed method employs local and global explainable techniques efficiently and precisely to optimize parameters. The paper also introduces a framework that evolves engineering designs using the proposed XAI-assisted evolutionary approach. A practical application involving the optimization of a 2D car chassis demonstrates the effectiveness of the proposed algorithm. Experimental results show that the proposed XAI -assisted evolutionary algorithm improves the quality of solutions by between 8.15% to 39.95% over four iterations compared to traditional methods.