Sistem Inspeksi Visual Berbasis Kecerdasan Buatan untuk Kualitas Manufaktur: Tinjauan Literatur
Year: 2025
Type: article
Study design: Systematic Review
Abstract: Visual inspection is a critical quality control process in modern manufacturing industries. Traditional manual inspection methods face significant limitations in terms of accuracy, consistency, and throughput. Recent rapid developments in artificial intelligence (AI), particularly in computer vision and deep learning, have opened new opportunities for automated visual inspection systems with superior accuracy levels. This research presents a systematic literature review analyzing 52 peer-reviewed studies published between January 2019 and October 2024. The research methodology follows PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines with systematic searches through IEEE Xplore, ScienceDirect, Springer, ACM Digital Library, and Google Scholar databases. Inclusion criteria encompass empirical studies implementing AI/ML techniques for visual inspection in manufacturing contexts with quantitative results. Analysis results demonstrate the dominance of deep learning methods (87% of total studies) with Convolutional Neural Networks (CNN) architecture as the most popular approach (64%), followed by YOLO series (27%) and R-CNN variants (21%). Transfer learning from pre-trained models was utilized in 71% of studies to address dataset limitations. Reported accuracy rates range from 92.3% to 99.8% with a median of 96.7%. Applications are distributed across various sectors: electronics (31%), automotive (25%), food processing (17%), pharmaceutical (13%), and general manufacturing (14%). Main challenges identified include limited labeled dataset availability (67% of studies), high computational requirements (54% of studies), generalization across production conditions (51% of studies), and class imbalance (46% of studies). This research contributes a comprehensive framework for appropriate AI technique selection, identification of implementation best practices, and a future research roadmap including few-shot learning development, explainable AI, and edge computing integration.
Source: JURNAL MASINA NIPAKE
Institution: Politeknik Negeri Manado
Language: Indonesian
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7220155533