A Self-healing Framework for Online Sensor Data Tuan Anh Nguyen, - PowerPoint PPT Presentation

A Self-healing Framework for Online Sensor Data Tuan Anh Nguyen, Marco Aiello Takuro Yonezawa Kenji Tei Keio University, National Institute of Informatics, University of Groningen, Japan Japan The Netherlands takuro@ht.sfc.keio.ac.jp tei@nii.ac.jp {t.a.nguyen, m.aiello}@rug.nl

Self-healing process of natural systems

MAPE-K architecture

The framework

The framework Fault correction Neighbourhood median Fault detection and • value classification Model learning: f(x) = v f • Models: Environment • model Fault model • Pre-processing: Execute: Historical data • Correct faults • Neighbours • Notify users •

Fault model

Fault detection and classification

Actual readings of a Temperature sensor Ground Truth Median of Neighbour readings Forecast value with ARMA

Actual readings of a Temperature sensor Ground Truth Median of Neighbour readings Intersection Result Forecast value with ARMA

Experiment: Intel Lab Dataset

Environment Model

Neighbouring: k-means++ with Dynamic Time Warping (DTW) as a distance K = 2 K = 3

First results with Intel Lab Dataset

City Data Process

The City Data Processing architecture

Santander sensors

K = 2 K = 3 Neighbouring: k-means++ with Dynamic Time Warping (DTW) as a distance

K = 2 K = 3

NodeID = 171

NodeID = 183

Self-healing for Santander sensors http://sox.ht.sfc.keio.ac.jp:54380/show/9652237040c8e344a2d553773f5feea0

Next steps: Fault correction Model learning with statistical pattern recognition 1. expectations of correct behaviour established at the calibration phase 2. historical sensor data.

Real-world implementation

Thank you very much for your attention! A Self-healing Framework for Online Sensor Data Tuan Anh Nguyen t.a.nguyen@rug.nl