觸控面板現今已廣泛用於許多設備，例如智能手機，平板電腦或筆記本電腦。而超聲波觸控系統有別於主流的電容式觸控，其應用彈性極大，意即凡是能激振蘭姆波的薄板結構，皆有可能實現觸控應用。
本文介紹了超聲波觸控系統的來由與設計一個可行的實驗架構，並且進行分析和實現；利用手指觸控面板產生蘭姆波能量的衰減進行分析進而達到觸控辨識之目的。進一步的說，本研究是發展即時觸控辨識且解析度大約為手指觸控面積的超聲波觸控面板。
觸控面板材料使用304不鏽鋼，大小為A4(297 mm×210 mm)，厚度為0.8 mm，定義不鏽鋼表面正中央110 mm×90 mm為觸控面積。激振訊號與接收訊號皆使用壓電材料(PZT)，透過產波器與訊號擷取卡實現。
後端用於辨識觸控位置程式主要的演算法使用卷積類神經網路(Convolution Neural Network, CNN)，輔以自動化收集訓練資料的方式，控制實驗變因且有效率收集大量資料。實驗結果為觸控精準度約95%、觸控解析度5 mm和反應時間為20 ms~25 ms之間。

Touch panels are now widely used in many devices, such as smartphones, tablets or laptops. The ultrasonic touch system is flexible and different from the current mainstream methods such as capacitive touch. The strength and advantage of using ultrasound for touch screen sensor is that any thin-plate structure which can carry the Lamb waves can be realized for touch applications.
This paper investigates the origin idea and implement a new design of an ultrasonic touch system. the basic theory is relying on the finger touch panel to perturb the propagation behaviors of Lamb wave. By correlating the touching position with the signal changes, one can achieve the purpose of touch recognition. Further, this study is to develop an ultrasonic touch panel capable of instant touch recognition and having a resolution of approximately the same contact area of a finger.
The touch panel material is a 304 stainless steel plate about A4(297 mm×210 mm) size and a thickness of 0.8 mm. In the central area of the stainless steel plate, an area of 110 mm × 90 mm in size is defined as the touch sensing area. Both the excitation signal and the received signal are made of piezoelectric material (PZT), which is realized by a wave generator and a data acquisition card (DAQ).
The main algorithm used to identify the touch location is Convolution Neural Network (CNN), which is supplemented by training data collected by an automated system. Using this system and an artificial finger, sufficient training data are obtained for efficiently establish a reliable and correct CNN model. The experimental results show that the touch accuracy is almost 95%, the touch spatial resolution is 5 mm, and the reaction time is between 20 ms to 25 ms.

[1]G. Walker, “A review of technologies for sensing contact location on the surface of a display,” Journal of the society for information display, vol. 20, no. 8, pp. 413-440, 2012.
[2]R. Adler and P. Desmares, "An Economical Touch Panel Using SAW Absorption", IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, vol. 34, no. 2, pp. 195-201, 1987.
[3]W. S. McCuulloch and W. Pitts, “A logical calculus of the ideas immanent in nervous activity,” Bulletin of mathematical biophysics, vol. 5, pp. 115-133, 1943.
[4]F. Rosenblatt, “The perceptron: A probabilistic model for information storage and organization int the brain,” Psychological Review, vol. 65, no. 6, 1958.
[5]John William Strutt Baron Rayleigh(1894), The Theory of Sound : Volume 1, London : Macmillan.
[6]K. Firouzi, A. Nikoozadeh, T. E. Carver, and Butrus (Pierre) T. Khuri-Yakub, ” Lamb Wave Multitouch Ultrasonic Touchscreen,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 63, no. 12, pp. 2174-2186, 2016.
[7]K. Fukushima, “Cognitron: A self-organizing multilayered neural network,” Biological Cybernetics, vol. 20, no. 3-4, pp.121-136, 1975.
[8]K. Fukushima, “Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position,” Biological Cybernetics, vol. 36, no. 4, pp.193-202, 1980.
[9]林彥廷, “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, 2017.
[10]K. Firouzi, A. Nikoozadeh, T. E. Carver, and Butrus (Pierre) T. Khuri-Yakub, ” Numerical modeling of ultrasonic touchscreen,” 2014 IEEE International Ultrasonics Symposium, Chicago, Illinois USA.
[11]A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Communications of the ACM, vol. 60, no. 6, pp.84-90, 2017.
[12]J. Melville, K. S. Alguri, C. Deemer, and J. B. Harley, “Structural damage detection using deep learning of ultrasonic guided waves,” 44th annual review of progress in quantitative nondestructive evaluation, vol. 37, no. 230004, pp.1-7, 2018.
[13]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,” Sensors and Actuators A: Physical, vol. 162, pp.394-399, 2010.
[14]V. Barat, P. Kostenko, V. Bardakov and D. Terentyev, “Acoustic Signals Recognition by Convolutional Neural Network,” International Journal of Applied Engineering Research, vol. 12, no. 12, pp.3461-3469, 2017.
[15]V. Ewalda, R. M. Grovesa, R. Benedictusb, “DeepSHM: A Deep Learning Approach for Structural Health Monitoring Based on Guided Lamb Wave Techniques,” SPIE Smart Structures + Nondestructive Evaluation 2019, Denver, USA.
[16]D. P. Kingma and J. L. Ba, “Adam: a method for stochastic optimization,” in , San Diego, USA, International Conference on Learning Representations(ICLR), May 2015, pp. 1-13.
[17]P.-T. de Boer, D. P. Kroese, S. Mannor, R. Y. Rubinstein, “A tutorial on the cross-entropy method,” Annel Operation Research, vol.134, no. 1, pp. 19-67, 2005.