臺灣博碩士論文加值系統

不鏽鋼泵浦因耐腐蝕和耐久環境下提供高效能流體運輸，而在排水系統、灌溉與其他農業應用設備以及高層建築設備領域被廣泛應用。然而焊接缺陷的發生經常由材料特性、焊接人員的焊接品質差異及製程變數的不穩定性因素進而影響不鏽鋼泵浦產品品質。由於傳統上在焊接缺陷都是依賴焊接人員進行目視檢測，容易受人為因素而誤判。本研究採用深度學習技術，開發一套基於卷積神經網路的氬焊焊道瑕疵自動檢測系統，並採用VGG、Xception、ResNet-152及DenseNet等四種模型進行訓練，可有效檢測良品及五種不同的焊接缺陷。
學習資料集包含5448張299×299灰階焊接焊道圖像，劃分為80%的訓練資料和20%的驗證資料，並經過100個Epochs進行模型的訓練。實驗的結果顯示，VGG、Xception、ResNet-152和DenseNet四種模型均可準確的分類出焊接焊道類別，其中DenseNet模型，在訓練準確度最高達97.5%，在驗證準確度則達96.8%。在測試資料集包含544張299×299灰階焊接焊道圖像，模型經測試後，準確度達到了96.7%。研究結果顯示了其訓練模型具備廣泛的檢測能力，為未來實際生產中減少人工檢測負擔，同時提升不鏽鋼泵浦產品品質水準。

Stainless steel pumps are extensively utilized in drainage systems, irrigation and other agricultural applications, as well as in high-rise building equipment due to their high-efficiency fluid transportation in corrosive and durable environments. However, welding defects often occur due to variations in material properties, differences in the welding quality of personnel, and instability in process variables, which subsequently affect product quality. Traditionally, the identification of welding defects has relied on visual inspections by welders, which are susceptible to human error. This study employs deep learning technology to develop an automatic defect detection system for argon welding seams using convolutional neural networks. The system was trained using four models: VGG, Xception, ResNet-152, and DenseNet, which effectively detect good products and five different types of welding defects.
The dataset used includes 5,448 grayscale images of welding seams, each sized at 299x299 pixels, divided into 80% training data and 20% validation data, and underwent training over 100Epochs. The experimental results demonstrate that the four models, VGG, Xception, ResNet-152, and DenseNet, can accurately classify welding seam categories, with the DenseNet model achieving the highest training accuracy of 97.5% and validation accuracy of 96.8%. The models were also tested on a dataset containing 544 grayscale welding seam images, achieving an accuracy of 96.7%. The results highlight the broad detection capabilities of the trained models, reducing the burden of manual inspections in actual production and enhancing the quality of stainless steel pump products.

摘要 ............................................................................................................................................. i
Abstract ....................................................................................................................................... ii
誌謝 ...........................................................................................................................................iii
目錄 ........................................................................................................................................... iv
表目錄 ........................................................................................................................................ v
圖目錄 ....................................................................................................................................... vi
第一章 緒論 .............................................................................................................................. 1
1.1 研究動機 ..................................................................................................................... 1
1.2 文獻回顧 ..................................................................................................................... 2
1.3 研究方法 ..................................................................................................................... 4
1.4 論文架構 ..................................................................................................................... 6
第二章 導輪焊道缺陷資料收集 .............................................................................................. 7
2.1 導輪焊道缺陷檢測環境 ............................................................................................. 7
2.2 導輪焊道瑕疵資料庫建立 ....................................................................................... 10
2.3 導輪焊道缺陷檢測之影像預處理 ........................................................................... 16
2.4 泵浦導輪焊道系統架構 ........................................................................................... 20
第三章 導輪焊道瑕疵模型訓練 ............................................................................................ 22
3.1 VGG模型優化導輪焊道分析 .................................................................................... 22
3.2 Xception模型優化導輪焊道分析 ............................................................................. 25
3.3 ResNet-152模型優化導輪焊道分析......................................................................... 28
3.4 DenseNet模型優化導輪焊道分析 ........................................................................... 31
第四章 研究結果與分析 ........................................................................................................ 35
4.1 VGG導輪焊道瑕疵分類模型訓練結果 .................................................................... 35
4.2 Xception導輪焊道瑕疵分類模型測試結果 ............................................................. 36
4.3 ResNet-152導輪焊道瑕疵分類模型測試結果......................................................... 37
4.4 DenseNet導輪焊道瑕疵分類模型測試結果 ........................................................... 38
4.5 導輪焊道缺陷檢測模型的綜合評估與效能比較 ................................................... 39
第五章 結論 ............................................................................................................................ 40
參考文獻 .................................................................................................................................. 41
Extended Abstract ..................................................................................................................... 43

[1]林玄良，2006，”銲接製程穩健設計最佳化之研究”，國立交通大學機械工程系所博
士論文。
[2]鄭家傑，2022，”氬銲製程應用在不鏽鋼薄短管件強化銲接品質之研究”，國立高雄
科技大學機械工程系碩士論文。
[3] Brosnan, T., & Sun, D.-W. , 2004, "Improving quality inspection of food products by
computer vision", Journal of Food Engineering, 61, 1, pp. 3-16.
[4] Cha, Y. J., Choi, W., & Büyüköztürk, O., 2017, " Deep Learning-Based Crack Damage
Detection Using Convolutional Neural Networks", Computer-Aided Civil and
Infrastructure Engineering, 32, 5, pp. 361-378.
[5] Krizhevsky, A., Sutskever, I., & Hinton, G. E., 2017, " ImageNet Classification with Deep
Convolutional Neural Networks", Communications of the ACM, 60, 6, pp. 84-90.
[6] Shaheed, K., Mao, A. H., Qureshi, I., Kumar, M., Hussain, S., Ullah, I., Zhang, X. M., 2022. "DS-CNN: A pre-trained Xception model based on depth-wise separable convolutional neural network for finger vein recognition. ", Expert Systems with Applications, 191, pp. 18.
[7] Sitaula, C., & Hossain, M. B., 2021, "Attention-based VGG-16 model for COVID-19 chest
X-ray image classification", Applied Intelligence, 51, 5, pp. 2850-2863.
[8] Tong, W., Chen, W. T., Han, W., Li, X. J., & Wang, L. Z., 2020, "Channel-Attention-Based
DenseNet Network for Remote Sensing Image Scene Classification ", IEEE Journal
of Selected Topics in Applied Earth Observations and Remote Sensing, 13, pp. 4121- 4132.
[9] Yang, J., Li, S. B., Wang, Z., Dong, H., Wang, J., & Tang, S. H., 2020, "Using Deep
Learning to Detect Defects in Manufacturing: A Comprehensive Survey and Current
Challenges", Materials, 13, 24, pp. 23.
[10]李柏桀，2018，”基於機械視覺之殘差網路深度學習像膠墊片表面瑕疵檢測”，國
立臺北科技大學機械工程系機電整合碩士論文。
[11] Simonyan, K., & Zisserman, A., 2015, "Very Deep Convolutional Networks for Large-
Scale Image Recognition", International Conference on Learning Representations (ICLR), pp. 1-14.
[12] Chollet, F., 2017, "Xception: Deep Learning with Depthwise Separable Convolutions"
Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1800-
1807.
[13] He, K., Zhang, X., Ren, S., & Sun, J., 2016, "Deep Residual Learning for Image
Recognition", Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770-778.
[14] Huang, G., Liu, Z., & Weinberger, K.Q., 2017, "Densely Connected Convolutional
Networks", 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2261-2269.