Phone Drop Impact Surface Detection: A Machine Learning Approach

Accidental phone drop is a common occurrence in the field and can lead to device failure. By knowing the severity of a drop event, the effective reliability of the device can be determined over time. One key parameter for severity is the impact surface type, which can be classified as soft, medium or hard. In this work, we propose a novel drop diagnostic solution that monitors phone accelerometer sensor data and determines the impact surface type. We present the results for two different classifier models using SVM and CNN respectively. Our method is able to achieve an accuracy of 95% across various surfaces and generalizes well on a range of smartphone devices. Additionally, we have evaluated the accuracy of the classifier in the presence of a protective case on the device. We show that the model performs fairly well with only marginal drop in accuracy on some surfaces. Our solution can be used in analyzing overall device health, buyback price estimation, and environment sensing capabilities in mobile cyber physical systems.