Recognizing Product Application based on Integrated Consumer Grade Sensors: A Case Study with Handheld Power Tools

Abstract

As more and more consumer grade sensors are integrated into products such as handheld power tools, data acquisition is becoming more affordable. The product development process benefits greatly from the knowledge of the incidence and the relevance of applications performed by a user. Using the collected sensor data remains a challenge due to the noise caused by the consumer grade sensors and the complexity of the applications. In this study, the use of machine learning on time series to identify which predefined applications a user performed with a power tool, using an angle grinder and a cordless screwdriver as example, is explored. An acceleration sensor, a rotation rate sensor, a geomagnetic sensor and a current sensor were used for data acquisition. We tested two sampling rates and two approaches of feature extraction: An effectively off-the-shelf method of feature extraction and an expert-tuned extraction of the features. The study shows that both methods achieve very good results within existing data sets (>95% accuracy). When applied to new experiments, overfitting occurs due to the complexity of the application and the noise of the consumer-grade sensors. Thus, this study shows first promising results and further potentials for the future application of machine learning for the recognition of applications in products such as handheld power tools.