Application of parallel computing for sensor data annotation

Sensor data annotation involves automated marking of a time series of readings taken from the sensor, which highlights various activities specified by the specified series. Activity marking has a wide range of practical applications: predictive maintenance, intelligent management of life support systems, climate modeling, etc. Previously, we developed a parallel PSF algorithm for annotating sensor data using a GPU based on the concept of snippets. Snippet is a subsequence that many other subsequences of a given series resemble in the sense of a specialized similarity measure based on Euclidean distance. This article describes two case studies performed using the PSF algorithm: annotation of the readings of a wearable vibration accelerometer mounted on a person and a stationary vibration accelerometer mounted on a small crusher. As part of the research, computational experiments were conducted to evaluate the speed and accuracy of the developed algorithm. Also there was the research on the dependence of the efficiency of the algorithm on the values of the input parameters: the number of the desired snippets and the length of the subsequence.