本論文以渦電流檢測技術為核心，提出一套應用於印刷電路板（Printed Circuit Board, PCB）製程中的非破壞性金屬厚度檢測系統，專注在銅箔厚度的高精度辨識，對於確保產品品質與提升製程穩定性具有重要意義，利用線圈感測器對不同厚度的銅箔進行掃頻檢測。由於厚度的不同會影響待測物上渦電流分布，進而改變感測線圈的阻抗與相位，因此本系統透過 1 MHz 以下的頻率掃描，提取待測物之阻抗與相位為主要辨識特徵，並於多個頻率下的資料中選取最具鑑別力之最佳頻率點，提升特徵穩定性與量測精度。在分類模型方面，本系統結合隨機森林（RF）、高斯混合模型（GMM）以及支持向量分類器（SVC），透過多模型融合策略提升分類穩定性。本研究利用有限元素法（FEM）進行感測線圈模擬，評估線圈參數（如匝數、線徑、間距）與樣本厚度對感應特性之變化趨勢，並探討其對柔性銅箔基板之電磁耦合影響。實驗結果顯示，本系統於三種機器學習模型下，整體分類正確率最高可達95%以上，顯示所提方法具備穩定且精確的量測能力，足以應用於實際生產線上之品質監控與製程優化。

The present paper sets out a non-destructive metal thickness detection system for printed circuit board (PCB) manufacturing processes based on eddy current technology, with a particular focus on high-precision copper foil thickness identification. The system is of significance in ensuring product quality and enhancing process stability through the utilization of coil sensors to perform frequency scanning on copper foils of varying thicknesses. It is evident that thickness variations have a significant impact on the distribution of eddy currents within the test specimen. This in turn results in alterations to the impedance and phase of the sensing coil. Consequently, the system utilizes frequency scanning below 1 MHz as a primary identification technique to extract impedance and phase. The system selects optimal frequency points with the greatest discriminatory power from multi-frequency data to enhance feature stability and measurement precision. In the context of classification modelling, the system integrates Random Forest (RF), Gaussian Mixture Model (GMM), and Support Vector Classifier (SVC) through a multimodal fusion strategy, with the objective of enhancing classification stability. The research incorporates the Finite Element Method (FEM) for the simulation of sensing coils, with the objective of evaluating the impact of coil parameters (such as number of turns, wire diameter, spacing) and sample thickness on induction characteristics. In addition, the research investigates the electromagnetic coupling effects of these coils on flexible copper foil substrates. The experimental results demonstrate that the system achieves overall classification accuracy exceeding 95% across the three machine learning models. This indicates that the proposed method possesses stable and precise measurement capabilities suitable for quality monitoring and process optimization in actual production lines.

[1]C. C. Tai, J. H. Rose, and J. C. Moulder, "Thickness and conductivity of metallic layers from pulsed eddy‐current measurements," Review of scientific Instruments, vol. 67, no. 11, pp. 3965-3972, 1996.
[2]T. Ying, P. Meng, and C. FeiLu, "Feature extraction based on the principal component analysis for pulsed magnetic flux leakage testing," in 2011 International Conference on Mechatronic Science, Electric Engineering and Computer (MEC), 2011: IEEE, pp. 2563-2566.
[3]V. Nagarkar et al., "CCD-based high resolution digital radiography system for non destructive evaluation," IEEE Transactions on Nuclear Science, vol. 44, no. 3, pp. 885-889, 1997.
[4]J. Coste and F. Lakestani, "Description of a method for the measurement of the Rayleigh wave velocity: application to the thickness measurement of metallic coatings," in 1994 Proceedings of IEEE Ultrasonics Symposium, 1994, vol. 2: IEEE, pp. 1233-1236.
[5]M. T. Ghasr, M. J. Horst, M. Lechuga, R. Rapoza, C. J. Renoud, and R. Zoughi, "Accurate one-sided microwave thickness evaluation of lined-fiberglass composites," IEEE Transactions on Instrumentation and Measurement, vol. 64, no. 10, pp. 2802-2812, 2015.
[6]C.-C. Tai, H.-C. Yang, and Y.-H. Liu, "Modeling the surface condition of ferromagnetic metal by the swept-frequency eddy current method," IEEE transactions on magnetics, vol. 38, no. 1, pp. 205-210, 2002.
[7]H.-C. Yang and C.-C. Tai, "Pulsed eddy-current measurement ofa conducting coating on a magnetic metal plate," Measurement science and technology, vol. 13, no. 8, p. 1259, 2002.
[8]C. C. Tai and S. F. Wang, "Time-domain and frequency-domain eddy current simulations by the finite element method," Key Engineering Materials, vol. 270, pp. 585-592, 2004.
[9]王勝豐, "有限元素法分析掃頻式和脈衝式渦電流檢測," 碩士, 電機工程學系碩博士班, 國立成功大學, 台南市, 2003. [Online]. Available: https://hdl.handle.net/11296/vh9r6u
[10]Y.-L. Pan and C.-C. Tai, "Thickness and conductivity analysis of molybdenum thin film in CIGS solar cells using resonant electromagnetic testing method," IEEE transactions on magnetics, vol. 48, no. 2, pp. 347-350, 2012.
[11] 翁新發, "基於諧振原理之渦電流檢測儀設計與應用," 碩士, 電機工程學系專班, 國立成功大學, 台南市, 2012. [Online]. Available: https://hdl.handle.net/11296/p97r38
[12]賴行健, "倒傳遞神經網路及主成份分析應用於金屬扣件硬度篩選檢測：系統設計與評估," 碩士, 電機工程學系, 國立成功大學, 台南市, 2019. [Online]. Available: https://hdl.handle.net/11296/8pjy4e
[13]李耘多, "基於K-近鄰分類法及梯度提升決策樹模型之印刷電路板銅膜厚度渦電流量測," 碩士, 電機工程學系, 國立成功大學, 台南市, 2020. [Online]. Available: https://hdl.handle.net/11296/2pwdse
[14]莊宗霖, "隨機森林法應用於渦電流金屬扣件硬度分類," 碩士, 電機工程學系, 國立成功大學, 台南市, 2020. [Online]. Available: https://hdl.handle.net/11296/7gwtha
[15]李柏毅, "非接觸式微米級金屬薄膜檢測系統研製及田口法最佳感測線圈設計," 碩士, 電機工程學系, 國立成功大學, 台南市, 2017. [Online]. Available: https://hdl.handle.net/11296/f2cg52
[16]蔡侑侖, "非接觸式微米級印刷電路板銅箔厚度量測系統及多重品質特性之最佳感測線圈設計," 碩士, 電機工程學系, 國立成功大學, 台南市, 2018. [Online]. Available: https://hdl.handle.net/11296/wmk942
[17]K. Tsukada, M. Hayashi, Y. Nakamura, K. Sakai, and T. Kiwa, "Small eddy current testing sensor probe using a tunneling magnetoresistance sensor to detect cracks in steel structures," IEEE Transactions on Magnetics, vol. 54, no. 11, pp. 1-5, 2018.
[18]N. Yusa, H. Hashizume, R. Urayama, T. Uchimoto, T. Takagi, and K. Sato, "An arrayed uniform eddy current probe design for crack monitoring and sizing of surface breaking cracks with the aid of a computational inversion technique," NDT & E International, vol. 61, pp. 29-34, 2014.
[19]Z. Qu, Q. Zhao, and Y. Meng, "Improvement of sensitivity of eddy current sensors for nano-scale thickness measurement of Cu films," NDT & E International, vol. 61, pp. 53-57, 2014.
[20]C.-C. Tai and Y.-L. Pan, "A Novel Multiphysics Sensoring Method Based on Thermal and EC Techniques and Its Application for Crack Inspection," in 2008 IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Computing (sutc 2008), 2008: IEEE, pp. 475-479.
[21]H. Wang and Z. Feng, "Ultrastable and highly sensitive eddy current displacement sensor using self-temperature compensation," Sensors and Actuators A: Physical, vol. 203, pp. 362-368, 2013.
[22]A. L. Ribeiro, H. G. Ramos, and J. C. Arez, "Liftoff insensitive thickness measurement of aluminum plates using harmonic eddy current excitation and a GMR sensor," Measurement, vol. 45, no. 9, pp. 2246-2253, 2012.
[23]M. Fan, B. Cao, P. Yang, W. Li, and G. Tian, "Elimination of liftoff effect using a model-based method for eddy current characterization of a plate," NDT & E International, vol. 74, pp. 66-71, 2015.
[24]E. Pinotti and E. Puppin, "Simple lock-in technique for thickness measurement of metallic plates," IEEE Transactions on Instrumentation and Measurement, vol. 63, no. 2, pp. 479-484, 2013.
[25]P. K. Mayakuntla, A. Ganguli, and D. Smyl, "Gaussian mixture model-based classification of corrosion severity in concrete structures using ultrasonic imaging," Journal of Nondestructive Evaluation, vol. 42, no. 2, p. 28, 2023.
[26]O. Bíró, "Edge element formulations of eddy current problems," Computer methods in applied mechanics and engineering, vol. 169, no. 3-4, pp. 391-405, 1999.
[27]J. García-Martín, J. Gómez-Gil, and E. Vázquez-Sánchez, "Non-destructive techniques based on eddy current testing," Sensors, vol. 11, no. 3, pp. 2525-2565, 2011.
[28]戴政祺, "「非破壞性檢測」上課講義," 國立成功大學電機系.
[29]A. Dempter, "Maximum likelihood from incomplete data via the EM algorithm," Journal of Royal Statistical Society, vol. 39, pp. 1-22, 1977.
[30]J. Fan, S. Chen, X. Luo, Y. Zhang, and G. Y. Li, "A machine learning approach for hierarchical localization based on multipath MIMO fingerprints," IEEE Communications Letters, vol. 23, no. 10, pp. 1765-1768, 2019.
[31]M. Min et al., "Estimating summertime precipitation from Himawari-8 and global forecast system based on machine learning," IEEE transactions on geoscience and remote sensing, vol. 57, no. 5, pp. 2557-2570, 2018.
[32]N. Reamaroon, M. W. Sjoding, K. Lin, T. J. Iwashyna, and K. Najarian, "Accounting for label uncertainty in machine learning for detection of acute respiratory distress syndrome," IEEE journal of biomedical and health informatics, vol. 23, no. 1, pp. 407-415, 2018.
[33]L. Breiman, J. Friedman, R. A. Olshen, and C. J. Stone, Classification and regression trees. Chapman and Hall/CRC, 2017.