H 型鋼構件表面塗層的厚度高度影響H 型鋼構件的耐久性。 H 型鋼構件搭配不同 的噴塗材料可以抵抗各種影響H 型鋼構件的情況。噴塗的厚度是決定噴塗效果的重要因素， 過厚與過薄的噴塗都會影響H 型鋼構件的品質。然而，目前產業界檢測噴塗厚度大多為人 工檢測，方法不可靠且存在誤差，因此，本研究透過類神經網路模型判斷塗層厚度是否足 夠，進而建立一可應用於業界的自動厚度分類系統。噴塗結果的數值影像被預先分為三個 組別，每個組別表示一種噴塗狀況(未噴塗、噴塗中、噴塗完成)。噴圖影像被分為訓練組 與測試組，分別用於建立與驗證類神經網路模型。研究使用卷積神經網路預測表面噴塗的 狀況，並根據模型預測結果判斷表面噴塗之厚度是否符合規範，研究分類結果顯示模型分 類噴塗厚度準確率約為93%。研究所提出的方法可以降低檢驗成本，也大量減少了錯誤檢 驗的發生。透過與噴塗儀器相結合，可將結果應用於實際場域，提高型鋼表面噴塗工作的 效率。;The thickness of H-shaped Steel coating is important for the durability of H-shaped Steel. With different coating materials applied to the H-shaped steel, it will be much capable to resistance rust and other situations. The thickness of the coating is an essential factor that determine the effect of the coating, improper thickness will cause problems and wastes. However, the detection method nowadays is held by human. The result is unreliable and error existence. Hence, the research uses machine learning to classify the thickness of the coating. The digital images are categorized into three groups, each group present a coating situation (uncoated, coating, and fully coated). Each figure is then separated into training group and testing group, which are used to generate the neural networks and test the accuracy of the networks respectively. The results of the neural networks will show if the thickness of the coating meet the thickness regulation. The research accuracy for the three group thickness classification is about 93%. The proposed method can reduce the cost of inspectors also eliminate the occurrence of mistakenly detection. The result can be applied to practical use by combining the classifying system with the coating machine and thus improve the efficiency of coating process.
