Grillo sensor benchmarking

Outline

This article evaluates the performance of Grillo sensors for seismic applications such as earthquake early-warnings. Historic accelerations from the Grillo sensor network in Mexico will be compared to the UNAM Engineering Institute Accelerometric Network (RAII-UNAM) which uses observatory grade accelerometers such as
Kinemetrics Episensor with 24bit digitizers. UNAM sites are a combination
of low-noise free field sites outside of cities and basin sites within large
population centers.

UNAM Engineering Institute Accelerometric Network

Grillo sensors

The Grillo Sensor is a three component accelerometer with a 20 bit digitizer and dynamic range of +/- 2g. It is paired with a microcontroller that continuously sends samples to the cloud, as well as status updates.
The device is configured to send 125 samples for each axis per second to Grillo’s cloud, where the data is stored, processed, and visualized.

Grillo sensor and dashboard

Installation

Grillo Sensors require specific installation conditions to ensure acceptable data quality:

  1. Located on ground or basement floor.
  2. Fixed mechanically to a rigid structural element such as column, beam or near a wall corner edge.
  3. Connects to local wifi using a free mobile app (or can be pre-programmed if required). The network needs to be of reasonable quality so that packets of telemetry aren’t lost. Grillo has deployed sensors with a 4g cellular connection in scenarios where local Wifi is not available.
  4. The device has a 5W wall adapter that needs to be connected to a mains socket.
  5. (optionally) a GPS antenna needs to be fixed with adhesive to a nearby window to get clear view of satellites.
  6. (optionally) a GPS antenna needs to be fixed with adhesive to a nearby window
  7. to get clear view of satellites.
Typical sensor installation against structural element.

Grillo Cloud

The sensor data typically arrives to the cloud in 100ms or so, depending on the local network. Grillo uses AWS data centers situated near the physical locations of sensors, although the processing could also be done on a local server or alternative cloud provider.


Once in the cloud, the data is ingested via a scalable and reliable message broker, and then routed to different services:

  1. Cold document storage for future analysis
  2. Real-time stream for processing (eg PGA, PVA or other transformations such as frequency domain analysis or spectral accelerations).
  3. Database and dashboard for visualizing traces, sensor activity, latency and more.

Once in the cloud, Grillo can service parallel real-time ingestions with ease.

Grillo PGAs compared to UNAM PGAs

To arrive at a comparison for PGAs during the February 2018 M7.2 Pinotepa event, peak ground accelerations were taken from the UNAM database and also from the Grillo traces stored in AWS.

The maximum value for a single component was chosen for the PGA on either dataset, and plotted here. It can be observed that the Grillo sensors show similar PGAs and distance decay consistent with the UNAM sensors. Clusters of UNAM sites above the mean acceleration are within city basins (Acapulco, Oaxaca, and Puebla).

PGA comparison for 7.2M event

Grillo PGVs compared to UNAM PGAs

To arrive at a comparison for PGAs during the February 2018 M7.2 Pinotepa event, peak ground accelerations were taken from the UNAM database and also from the Grillo traces stored in AWS.

The maximum value for a single component was chosen for the PGA on either dataset, and plotted here. It can be observed that the Grillo sensors show similar PGAs and distance decay consistent with the UNAM sensors. Clusters of UNAM sites above the mean acceleration are within city basins (Acapulco, Oaxaca, and Puebla).

PGV comparison for 7.2M event

Comparison with ground motion models

We routinely observe that shaking intensities measured by Grillo compare favorably to regional and global ground motion models (GMMs).
Shown here are 3 separate events. The scatter plots show the comparison to published GMMs (Zhao, Atkinson, McVerry, LinLee).
It can be observed that the Grillo PGAs align well with predictions from GMMs.

We note slightly high residual ground motions for the M7.2 earthquake (observed in UNAM data too). These are likely due to the unusually small source dimensions of the earthquake which denotes a high stress drop. See Li et al (2020).

1

Sample waveforms for one Grillo sensor for 3 events

Sensor 6, M7.2, 2/6/2018
Sensor 6, M5.9, 2/6/2018
Sensor 6, M6.0, 2/6/2018

M6.2 2018.04.10 (Chile)

Grillo also operates a network within Chile for early warning focused on Santiago. The largest event recorded thus far was an M6.2 in April 2018 at 200 km distance .

Conclusion

It is shown that Grillo sensors perform as well as one of Mexico’s national strong motion networks for the purpose of PGA and PVA generation.
Grillo accelerometers have enough fidelity and dynamic range to easily record all felt and damaging events at local and regional distances.
The lower cost and ease of  deployment allows Grillo sensors to form dense networks to correctly capture the variability of ground motion across an urban
center.
As well as generating high quality data, Grillo also offers very low-latency transmission and processing through the cloud, resulting in near-instant and accessible data for end users.

Share on facebook
Facebook
Share on twitter
Twitter
Share on linkedin
LinkedIn

This website uses cookies to ensure you get the best experience on our website. Learn more