| 摘要: | 本論文探討如何運用人工智慧(AI)來輔助企業發展，當企業發展至一定規模後便會擴大生產量，因此在萬物齊漲的年代，當原物料及人事費用都成為企業發展的負擔時，這些成本無形之中也形成一種壓力。若又因為產品的品質不良發生時，因它造成的重工成本和損失都會嚴重損害公司的聲譽及形象。公司為了生存，除了壓低原物料成本，大量引進機器生產來代替人工但品質檢驗卻還是依靠人力，且常因產線調換或人員的流動和一些人員素質參差不齊等因素，從而引發人為的失誤。因此當產品被退貨時，公司必需再花費額外的回收成本及更多的人力去拆組產品並再重新生產及檢驗，相當費時耗力。倘若有一項技術可以代替人工目視檢測用模擬人類的視覺的方式，對產品的外觀品質進行檢驗，而且不會因為檢驗人員工作疲勞或人才流動等因素，導致人為的失誤進而影響公司出貨後的產品質量。故本論文主要探討的動機，便是期望可以提高產品檢驗的質量及節省人事成本的前提下，既可以滿足客戶的期望，更能維持公司的好形象。因此如何利用深度學習(DP)中的影像辨識技術，透過機器學習(ML)來訓練機器學習辨識產品外觀，利用機器檢視來替代人類視覺，對產品外觀進行檢驗把關，且因機器可以24小時運轉，除減少人員換班或因疲勞導致誤判等風險外更可提高檢驗效率。
以此目標做為研究的動機，故使用深度學習(DP)中的影像識別技術，利用YOLO演算法工具，標記產品之圖像，結合電腦的運算能力，以此來提升產品品質的檢驗一致性，經此實驗後，期望能利用這項研究的成果來取代人工檢驗方式，進而提升出貨的品質為目標。
;This paper discusses how to utilize Artificial Intelligence (AI) to assist in business development. As a company expands and increases production, the rising costs of raw materials and labor become burdensome. These costs inadvertently create pressure on the company. Poor product quality can lead to significant rework costs and damage the company′s reputation and image. To survive, companies must not only lower raw material costs and extensively use machinery to replace manual labor, but they also rely heavily on human resources for quality inspection. Frequent changes in production lines, personnel turnover, and inconsistent employee quality can lead to human errors. When products are returned, companies must incur additional costs for recovery and allocate more manpower to disassemble, reassemble, and re-inspect the products, which is time-consuming and labor-intensive.
If there was a technology that could replace manual visual inspection by simulating human vision to inspect the appearance quality of products, and it would not cause human errors due to factors such as fatigue of inspectors or talent turnover, which would then affect the company′s shipments? product quality. Therefore, the main motivation of this paper is to improve the quality of product inspection and save personnel costs, which can not only meet customer expectations, but also maintain a good image of the company.
Therefore, this thesis investigates how to use image recognition technology in deep learning (DL) and machine learning (ML) to train machines to recognize product appearances. This technology aims to replace human visual inspection with machine vision, ensuring product quality control. Since machines can operate 24/7, this reduces the risks of errors due to shift changes or inspector fatigue and improves inspection efficiency.
The goal of this research is to use deep learning (DL) and the YOLO algorithm to label product images and leverage computer processing power to enhance the consistency of product quality inspection. Through this experiment, we hope to replace manual inspection with this research′s findings, ultimately improving the quality of shipped products. |
