Instrumentation & Measurement Magazine 23-9 - 23

high-speed hardware counter, clocked at 64 MHz; once every
hour a 32-bit low-speed SW counter is updated. The cycle-to-
cycle jitter of the internal high-speed clock system is 300 ps,
whereas its accuracy for soldered parts working in the -10 to
85 °C temperature range is -1.9% to 2.3% with respect to the
nominal value.
Measurements taken on the implemented sensor network
showed cycle-to-cycle jitter of 239.5 ps, a minimum deviation
of -0.069% and a maximum deviation of 0.026% over a time period of 2400 s. The synchronization algorithm exploited in this
work is based on a SW implementation of the classical threeway handshake adopted by the RFC 793 Transmission Control
Protocol, according to which the reception of each data packet
must be acknowledged by the receiver before the next packet is
sent by the transmitter. The maximum divergence between the
sensor nodes' clocks, which is due to inherent clock's drift, was
reduced to 4.7 ms by issuing the synchronization command
once every 5 s. The obtained value is acceptable for vibration-based structural inspection [10]. The sensor-to-GW data
transmission is performed sequentially, in packets, by exploiting a proprietary lossless encoding technique.

Fig. 2. Proposed layered architecture for structural health monitoring and
data analysis.

and small footprint accelerometer sensor nodes [9]. Each of
them features an ST Microelectronics STM32F303 32bit, 3.3 V
low-power microcontroller unit (MCU) embedding Digital Signal Processing (DSP) functionalities and a floating-point unit
(FPU). The sensing element consists of a 6 Degree-of-Freedom
(DoF) system-in-package LSM6DSL device, a MEMS-based
inertial measurement unit (IMU) able to simultaneously provide triaxial accelerations and as many angular velocities.
Admitted values for the output data rate span from 1.6 Hz to
6.664 kHz, whereas the minimum linear and angular sensitivity per Least-Significant Bit correspond to 0.598·10−3 m/s2 and
0.074·10−3 rad/s, respectively. Multiple devices can be joined
in a daisy-chain fashion by means of a multidrop Sensor Area
Network (SAN) bus, which exploits data-over-power communication leveraging the EIA RS-485 standard. This protocol can
be used effectively over long distances and in electrically noisy
environments, such as industrial sites. A wired connection was
preferred over a wireless one to grant the user the possibility
to acquire data from the structure at high data-rates while preserving data confidentiality from external attacks. Moreover,
this choice led to the design of lighter nodes, which did not require the presence of a battery.
Meaningful information sensed by each device is transmitted to an Edge Controller (EC) through a companion Gateway
(GW) network interface (Fig. 2), which can orchestrate up to
64 nodes at a time. Nonetheless, it should be mentioned that
the maximum number of sensors per GW could be arbitrarily
increased by means of repeater nodes. During acquisition,
signals are collected simultaneously by each sensor node. A
unique time-stamp is provided by means of an internal 32-bit
December 2020

Data Acquisition Layer
The EC in Fig. 2 is configured to gather the measurements produced by each sensor of the monitored structure, consequently
presenting them towards a remote cloud. To cope with sensor
heterogeneity while minimizing the need for manual configuration and intervention, a SW layer was specifically added to
each EC to virtualize the sensor operations by making them accessible and discoverable from a remote client. Following this
design, we leveraged the Web of Things (WoT) paradigm [3],
a recent standard promoted by a W3C working group that enables mutual interworking of different IoT eco-systems and
devices by means of web technologies. In detail, the WoT architecture identifies the concept of a Thing as a physical or a
virtual entity whose interfaces are described by a WoT Thing
Description (TD). The TD includes a list of machine-understandable meta-data that specify, among others, the list of
properties (e.g., state variables), actions and events exposed
by a Thing as well as its communication strategies (protocol
bindings). Hence, the TD does not define the implementation of the IoT physical devices but rather its services and the
way they can be accessed by other SW components by means
of a uniform and well-defined interface. To this aim, the TD is
usually encoded in JSON-LD language and likely annotated
with semantic labels that provide a machine-understandable
knowledge representation of each property/action/event.
An example of such annotations can be found in [11] and [12],
where two of the most popular semantic ontologies for the
SHM domain are described. In our case, each sensor is represented by a dedicated Web Thing (WT); the properties that can
be read from a remote Web client include, for instance, the raw
sensing values (e.g., 3-axial accelerometer values) and the aggregated features extracted from the raw signal (e.g., min/max
peak values). A small subset of the TD associated to each accelerometer sensor is sketched in Table 1.

IEEE Instrumentation & Measurement Magazine 23

Instrumentation & Measurement Magazine 23-9

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-9