A Multi-objective Optimization Approach for Feature Selection in Gentelligent Systems

TL;DR

The paper tackles feature selection for fault detection in gentelligent smart manufacturing by formulating a bi-objective optimization that minimizes feature count and cost. It introduces a dominance-based MOEA that identifies all Pareto-optimal feature subsets in a single run and uses a DNN to estimate subset costs, guiding selection. The approach combines non-dominated sorting with crowding distance to maintain diversity and leverages a neural cost predictor to evaluate candidate feature sets. Empirical results on SECOM and Tennessee Eastman Process datasets show superior accuracy with fewer features compared to several evolutionary algorithms, highlighting practical benefits for Industry 4.0 applications and predictive maintenance.

Abstract

The integration of advanced technologies, such as Artificial Intelligence (AI), into manufacturing processes is attracting significant attention, paving the way for the development of intelligent systems that enhance efficiency and automation. This paper uses the term "Gentelligent system" to refer to systems that incorporate inherent component information (akin to genes in bioinformatics-where manufacturing operations are likened to chromosomes in this study) and automated mechanisms. By implementing reliable fault detection methods, manufacturers can achieve several benefits, including improved product quality, increased yield, and reduced production costs. To support these objectives, we propose a hybrid framework with a dominance-based multi-objective evolutionary algorithm. This mechanism enables simultaneous optimization of feature selection and classification performance by exploring Pareto-optimal solutions in a single run. This solution helps monitor various manufacturing operations, addressing a range of conflicting objectives that need to be minimized together. Manufacturers can leverage such predictive methods and better adapt to emerging trends. To strengthen the validation of our model, we incorporate two real-world datasets from different industrial domains. The results on both datasets demonstrate the generalizability and effectiveness of our approach.

Figures (8)

• Figure 1: Different phases of manufacturing data life-cycle management.
• Figure 2: Different phases of the implemented model including the proposed feature selection methods.
• Figure 3: Representation of solutions assigned to different fronts.
• Figure 4: The neural network architecture considered to compute feature selection costs.
• Figure 5: Representation of solutions in the second phase