本論文將發展一套以PZT壓電材料為基礎的聲波發送與接收系統，激發出在板中的導波 (蘭姆波)，並以蘭姆波在板上受到干擾所產生的變化為數據，結合人工智慧的深度學習 (Deep Learning)，設計一套可行的觸控辨識方法。
架構上，以多通道的資料擷取 (Data Acquisition, DAQ) 卡擔任電壓的輸入輸出，藉由LabVIEW程式控制DAQ將電壓輸入PZT壓電材料，經由壓電片的激震後產生蘭姆波，蘭姆波在經由板中傳遞後由另一端PZT接收，並輸出電壓信號給DAQ，並將波形信號輸出至以人工智慧所建立之模型，並進行及時的觸控位置判讀。為了對此一系統進行監督式的人工智慧訓練與模型建置，本研究以LabVIEW控制一個XYZ的三軸龍門式位移平台，並搭配一個仿真的假手指，可以依序觸碰指定位置之鋼板表面的觸控區域，並同時儲存因觸碰產生的波形變化資料，最後將資料匯入程式語言Python進行卷積神經網路 (Convolutional Neural Networks, CNN) 來訓練其辨識模型，最後將模型檔案送入DAQ的C++架構底下進行即時辨識，觀察準確程度。
針對本研究所完成之具備人工智慧判讀能力的超音波觸控位置感測系統，實際上以真手指進行多次與反覆的實驗測試，發覺在12x12 cm2 的感測面積與1 cm2 的感測解析度之下，可以達到 95 % 的正確度，而且反應的時間約為89 ms，亦即具有11.25 Hz 的更新速率，驗證此一設計與架構的可行性與實用價值。本研究將特別討論一種有益於觸控辨識的 PZT 壓電片配置方法，可以利用系統的對稱性，大幅度提高CNN觸控模型的辨識度。

This thesis developed an ultrasonic signal transmission and receiving scheme based on PZT ceramics piezoelectric transducers. Four of the transducers in a thin plate can excite Lamb waves and the others can detect the Lamb wave signals. When the plate surface is locally touched, for example by a finger, the changes in the received Lamb wave signals can be used to actually identify the position of the finger touch. To fulfill this ultrasonic touch screen, the deep learning of artificial intelligence (AI) is introduced for signal recognition and position detection.
A data acquisition (DAQ) system is constructed for handling the excitation and waveform receiving. To excite the Lamb waves, a LabVIEW program regulates the voltage of DAQ’s analog input to the PZT piezoelectric transducers. As the Lamb waves passing through the plate and reaching another PZT transducer, a output voltage is recorded by the DAQ. The changes in these waveforms caused by the finger touch are used by the AI system for recognizing the event of touch as well as the x-y position of the touch. For the purpose of AI model training, a XYZ three-axis stage is used to carry an artificial finger so that it can sequentially touch the steel plate surface for touch region. All the data collected by the simulated finger touches are used as the training data for the AI modeling based on Convolution Neural Network (CNN). For the CNN training and modeling, the programming language Python was used. However, for immediate identification and precision observation, the trained model is transferred to TensorFlow built by C++.
Finally, we are able to test the ultrasonic touch screen by actual human finger touching. Multiple tests are carried out and data are collected. It shows that, for a 12x12 cm2 sensing area and a spatial resolution of 1 cm2, the accuracy of position determination can reach 95 %. It is notices that, for achieving such a high fidelity of position sensing, the deployment of PZT piezoelectric transducers and the way for forming data for the CNN can be very critical and hence should be done wisely.

[1] G. Walker, “A review of technologies for sensing contact location on the surface of a display,” J. Soc. Inf. Display, vol. 20, no. 8, pp. 413-440, 2012.

[2] R. Adler and P. J. Desmares, “An economical touch panel using SAW absorption,” IEEE Trans. Ultrasonics, Ferroelectrics, Freq. Control, vol. 34, no. 2, pp. 195-201, 1987.

[3] K. North and H. D’Souza, “Acoustic pulse recognition enters touch-screen market,” Inf. Display, vol. 12, pp. 22-25, 2006.

[4] S. Reis et al., “Touchscreen based on acoustic pulse recognition with piezoelectric polymer sensors,” in Proc. IEEE Int. Symp. Ind. Electron., pp. 516-520, 2010.

[5] V. G. Reju, A. W. H. Khong, and A. B. Sulaiman, “Localization of taps on solid surfaces for human-computer touch interfaces,” IEEE Trans. Multimedia, vol. 15, no. 6, pp. 1365-1376, 2013.

[6] Y. Liu, J. P. Nikolovski, N. Mechbal, M. Hafez, and M. Vergé, “A multi-touch plate based on lamb wave absorption,” Procedia Chem., vol. 1, no. 1, pp. 156-159, 2009.

[7] Y. Liu, J. P. Nikolovski, N. Mechbal, M. Hafez, and M. Vergé, “An acoustic multi-touch sensing method using amplitude disturbed ultrasonic wave diffraction patterns,” Sens. Actuators A, Phys., vol. 162, no. 2, pp. 394-399, 2010.

[8] 林彥廷, 李永春, “壓電材料與蘭姆波應用於觸控辨識,” 國立成功大學機械工程學系碩士論文, 2018.

[9] Kunihiko Fukushima, “A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position,” Biol. Cyber., vol. 36, pp. 193-202, 1980.

[10] R. Adler and P. J. Desmares, “An economical touch panel using SAW absorption,” IEEE Ultrasonics Symposium, vol. 1, pp. 499-502, 1985.

[11] R. Adler, P.J. Desmares, “Saw touch systems on spherically curved panels,” IEEE Ultrasonics Symposium, vol. 1, pp. 289-292, 1986.

[12] Y. Tanaka, M. Takeuchi, J. L. Aroyan, J. Kent, “P2I-2 Characteristics of SAW Grating Reflectors with Arbitrary Angles of Incidence in Ultrasonic Touch Sensors,” IEEE Ultrasonics Symposium, vol. 1, pp. 1778-1781, 2006.

[13] Ajay Raghavan, Carlos E. S. Cesnik, “3-D Elasticity-Based Modeling of Anisotropic Piezocomposite Transducers for Guided Wave Structural Health Monitoring,” ASME, J. Vib. Acoust., vol. 129, no. 6, pp. 739-751, 2007.

[14] Ajay Raghavan, Carlos E. S. Cesnik, “Modeling of piezoelectric-based Lamb wave generation and sensing for structural health monitoring,” SPIE Proceedings, vol. 5391, pp. 419-430, 2008.

[15] LeCun, Y., Bottou, L., Bengio, Y. & Haffner, P., “Gradient-Based Learning Applied to Document Recognition,” Proceedings of the IEEE, vol. 86, pp. 309-318, 1998.

[16] Rumelhart, D. E., Hinton, G. E. & Williams, R. J., “Learning internal representations by error propagation,” Parallel Distributed Processing, MIT Press, vol. 1, pp. 318-362, 1986.

[17] P.Y. Simard, D. Steinkraus, J.C. Platt, “Best practices for convolutional neural networks applied to visual document analysis,” Proc. Seventh Int'l Conf. Document Analysis and Recognition, 2003.

[18] Warren S. McCulloch, Walter Pitts, “A logical calculus of the ideas immanent in nervous activity,” The bulletin of mathematical biophysics, vol. 5, pp. 115-133, 1943.

[19] Yann LeCun, Corinna Cortes, Christopher J.C. Burges, “THE MNIST DATABASE of handwritten digits,”
Retrieved from http://yann.lecun.com/exdb/mnist/

[20] Vincentius Ewald, Roger M. Groves, Rinze Benedictus, “A Deep Learning Approach for Structural Health Monitoring Based on Guided Lamb Wave Techniques,” Conference: SPIE Smart Structures + Nondestructive Evaluation 2019, 2019.

[21] Karen Simonyan, Andrew Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” Published as a conference paper at ICLR, 2014.

[22] A. Krizhevsky, I. Sutskever, and G.E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” NIPS'12 Proceedings of the 25th Inter. Confer. on Neural Info. Processing Systems, vol. 1, pp. 1097-1105, 2012.