Object oriented time series exploration: Applied to power consumption analysis of embedded systems

Correlating Time Series Signals and Event Logs in Embedded Systems

“…Unfortunately, using these approaches in external device monitoring is not satisfactory due to limited capabilities of accurate data acquisition systems. We faced this problem while developing some embedded and IoT devices [2]. The monitored embedded devices quite often do not provide hardware/software synchronisation capabilities or do not accept the impact of additional synchronisation processes on their operation.…”

“…As opposed to classical approaches dealing with TS at sample level, we consider TS objects aggregating samples, e.g., pulses, series of pulses, snippets, states, and state sequences. We showed the usefulness of aggregating TS samples into objects in [2,28] while analysing operations of embedded and complex computer systems, respectively. The presented studies in subsequent sections of this paper correlate not individual events but event sequences.…”

“…Within TS, we can distinguish objects defined as subsets of samples with specified properties, e.g., higher average values within a predefined window time and pulses of specified shapes. Decomposing TS into objects can be performed manually or using special algorithms targeted at pulses or more complex object features, as illustrated in our previous papers [2,28], respectively. Subsequent object instances can appear in diverse moments of the TS time scale.…”

“…In the case of external monitoring of signals (typically, with special data acquisition equipment), we face the problem of time correlation of the derived time series (collected data samples) and other logs generated by the monitored system, usually not synchronised. This problem is neglected in the literature, and we showed its practical importance [2] while analysing time series features at a higher observation level involving data sample aggregation into time series objects (e.g., pulses, characteristic sample sequences).…”

“…In practice, this process needs efficient real-time monitoring of the device's operation. It involves observation of selected physical signals and device and environment states ( [1,2] and references therein). From an analytical point of view, this leads to correlation analysis of time series depicting considered signal states and relevant device state/event logs.…”