Ensemble Learning-Based Technique for Force Classifications in Piezoelectric Touch Panels

Human-machine interfaces (HMI) based on piezoelectric sensors have been receiving increasing attention due to their high force detection accuracy and low energy consumption, along with the ability to deploy user-oriented force sensing capability enabled by artificial intelligence. However, reports to date use a large amount of data to construct a customized model, which reduces user experience. To address this issue, an ensemble learning-based technique is presented in this paper, for building a customized classification model with a much lower data set. Experimental results demonstrate high force detection accuracy (98.32%) at low computational cost and at data collection times of less than a minute. This implies much fewer user operations yet enhancing the user experiences of HMI.

I. Introduction

Touch panels (TPs) have become the primary human–machine interface (HMI) in many electronic devices, especially in mobile phones, which are tremendously utilized in our daily life for different purposes such as online banking, surfing the internet and playing games [1], [2]. Traditional touch panels can only locate 2D (- axis) positions using capacitive and resistive architectures [3]. In contrast, 3D (force) touch detection, which was firstly commercialized in 2015, has gained increasing popularity for obtaining additional dimensional information [4], [5], owing to its merit in improving the efficiency of human-machine interactions. In the first generation of 3D touch functions, such as iPhone 6S, capacitance sensors were employed into the backlight of the display to measure the force-induced distance shift between the cover glass and the backlight. Another commercialized product, iPhone X, utilizes additional electrode to measure the force-induced resistance change. These products use additional components to differentiate two force levels [6]. The acquisition of touch force information by these products enables users and end-terminal to exchange data in a higher efficiency means, based on which, promising applications have been shown in various fields, such as gaming and drawing applications. Therefore, the detection of touch force has attracted global attention, and a lot of research on force touch sensing has emerged, which subsequently led to several interesting results [7]–[10]. Among them, piezoelectric sensors in TPs is attracting significant attention because of the high force detection sensitivity, simple panel structure, and low power consumption that is enabled by piezoelectric effect’s intrinsic ability to convert stress to voltage [8], [11]. Here, a short comparison between piezoelectric and capacitive means is given for explaining the above statement. From the working principle, capacitive sensing is an active detection technique, while piezoelectric detection is a passive sensing means, hence the piezoelectric means consumes less power. For the readout circuit, the capacitive sensing techniques usually employ more than one operational amplifier to modulate and demodulate the signal. In contrast, piezoelectric readout merely needs one operational amplifier functioning as a charge amplifier to convert force signals to voltage outputs. Hence, the piezoelectric technique consumes less energy compared to the capacitive means. A detailed comparison of these two techniques in terms of force sensing is given in [12].