本研究探討高頻振動測試環境下的電路板元件檢測，使用簡易設備（iPhone 14 Pro 與 Lenovo P1 Gen1 筆記型電腦），應用YOLOv8 物件偵測模型監測電路板於5 Hz ~ 2000Hz隨機振動測試過程中的元件狀態。 在進行YOLOv8模型訓練之前，本研究使用Albumentations開源影像擴增函式庫進行旋轉與HSV色調變換，直接擴充訓練影像與標注資料，用以提升模型對不同視角與光線條件的泛化能力。影像經擴增後訓練出的模型，在F1-score、Precision、Recall、mAP等指標皆有顯著提升，並於高頻振動環境下展現更穩定的偵測表現。為呈現模型實際運作效果，本論文彙整測試過程中每幀影像的元件檢出結果，依時間軸、拍攝距離與元件類別等條件繪製圖表，視覺化顯示模型偵測成果與表現變化趨勢。 此外，本研究結合YOLOv8訓練模型與OpenCV函式庫，開發即時偵測系統，透過計算相鄰幀之Bounding Box幾何中心距離，實現於振動測試影片中進行零件追蹤，並成功應用於測試過程中之即時監控。 本研究驗證了 YOLOv8 在高頻振動測試場域的應用潛力，適合於產品設計與開發初期階段，提供有效的電路板零件偵測方案，提升了品質驗證測試的能力。;This study explores circuit board component detection under high-frequency vibration using simple equipment, including an iPhone 14 Pro and a Lenovo P1 Gen1 laptop. The YOLOv8 object detection model monitors component status during random vibrations from 5 Hz to 2000 Hz. Before training, the Albumentations library is used to apply image rotation and HSV adjustments. The augmented images and labels are directly added to the dataset to improve model generalization under varying angles and lighting. The resulting model shows notable improvements in F1-score, Precision, Recall, and mAP, and offers more stable performance under vibration. To present the model’s behavior, this study visualizes per-frame detection results using charts based on time, distance, and component type, highlighting performance trends. A real-time detection system is also developed by combining the trained YOLOv8 model with OpenCV. By measuring distances between bounding box centers across frames, the system tracks components and is applied to real-time monitoring during vibration tests. The results confirm YOLOv8’s potential in vibration testing and its suitability for early-stage product development. The approach provides an effective solution for component detection and enhances quality verification.
