1. Structural Health Monitoringin WSNs by theEmbedded Goertzel Algorithm Maurizio Bocca, Janne Toivola, Lasse M. Eriksson, JaakkoHollmen, Heikki Koivo Department of Automation and Systems Technology Aalto University School of Electrical Engineering, Helsinki, Finland www.wsn.tkk.fi
2. Looking back at SHM Brooklyn Bridge, NYC African Elephant AIM: accurate diagnosis of the health of civil infrastructures from data collected by sensors 1 ICCPS 2011, Chicago, IL, USA, 14.4.2011
3. Few Facts about Finland It is extremely difficult to retrieve elephants… Extremely cold climate Very rigid temperatures Lots of snow and ice 187 888 lakes… 14 000 (approx.) public roads bridges 150-200 new bridges built every year Most of the bridges built in the 60’s and 70’s: Originally not engineered for the current volume and type of traffic Approaching the critical “50-years maintenance check” 2 ICCPS 2011, Chicago, IL, USA, 14.4.2011
4. The 4 levels of SHM Damage detection Damage localization Damage quantification COMPLEXITY Assessment of the remaining lifetime of the structure PATIENT: the monitored infrastructure DOCTOR: the structural engineer ICCPS 2011, Chicago, IL, USA, 14.4.2011 3
5. Outline of the Talk 4 ICCPS 2011, Chicago, IL, USA, 14.4.2011 Goertzel algorithm (GA) and Transmissibility Functions (TFs) WSN architecture Experimental evaluation
6. Published in 1958. Classic application: DTMF Compared to the FFT, the GA: allows to efficiently calculate the amplitude of the frequency spectrum at specific bins (frequencies of interests, fi) worksiteratively: no need to store the accelerationsignals in RAM orFlashmemory for off-lineprocessing the number of collected samples (N) does not need to be a number power of 2 Why the GA for SHM? 1/2 5 ICCPS 2011, Chicago, IL, USA, 14.4.2011 t sampling START sampling END sample acquisition GA computations
8. GA: Concept & Parameters The GA implements a 2nd-order IIR filtercentered at each frequency of interest: 3 keyparameters (set by the end-user): Samplingfrequency(fs) Distance (db) betweentwoconsecutivebins on the frequencyaxis (resolutionr = 1/db) Vector of frequencies of interest (fi) 7 ICCPS 2011, Chicago, IL, USA, 14.4.2011
9. From the Goertzel Algorithm... Before the sampling: Number of samples (N) to be collected to obtain the fixed resolution (r = 1/db) Bins (k) corresponding to the selected frequencies of interest (fi) Coefficients (c) used in the GA iterations During the sampling: At the end of the sampling: 8 ICCPS 2011, Chicago, IL, USA, 14.4.2011 si: last collected sample q1 and q2 store the results of the two previous iterations Xi: squared magnitude of the spectrum
10. ...to Transmissibility Functions Transmissibility: the result of the interference of vibrations propating and reflecting along the structure TFsachieveenvironmentalinvariability Damage indicator: 9 ICCPS 2011, Chicago, IL, USA, 14.4.2011 si and sj: sensor nodes (fi ,f2): range of frequencies of interest REF: reference (undamaged) TEST: current condition (damaged?)
11. Flow of the Application 10 ICCPS 2011, Chicago, IL, USA, 14.4.2011 Sensor Node #N ... Sink Node Sensor Node #1 Sensor Node #2 Goertzel Algorithm Parameters Broadcast Accelerometer Sampling & Goertzel Algorithm Computations Damage Indicators Results Broadcast Transmissibility Functions Computations TDMA Results Sharing t t t
27. ExperimentalValidation (D5 and D6) 19 ICCPS 2011, Chicago, IL, USA, 14.4.2011 D5: 27.6% stiffness reduction D6: 55.2% stiffness reduction
28. GA vs Off-Line Modal Analysis (1/3) Modal analysis: study of the dynamic properties of structures under vibrational excitation Centralized off-line data analysis for identifying natural frequencies, mode shapes and damping ratios 20 ICCPS 2011, Chicago, IL, USA, 14.4.2011 M. Bocca, A. Mahmood, L.M. Eriksson, J. Kullaa, and R. Jäntti, A Synchronized Wireless Sensor Network for Experimental Modal Analysis in Structural Health Monitoring, Computer-Aided Civil and Infrastructure Engineering, 2011.
29. GA vs Off-Line Modal Analysis (2/3) 21 ICCPS 2011, Chicago, IL, USA, 14.4.2011 measurementperiod: 30 s, samplingfrequency: 125 Hz
30. GA vs Off-Line Modal Analysis (3/3) 22 ICCPS 2011, Chicago, IL, USA, 14.4.2011 measurementperiod: 30 s, samplingfrequency: 125 Hz
31. Conclusions 23 ICCPS 2011, Chicago, IL, USA, 14.4.2011 By the embeddedGoertzelalgorithm, it is possible to correctlydetect and localizestructuraldamages The proposedsystem is low-latency and low-power