Hand-Arm Vibration Estimation Using A Commercial Smartwatch

Abstract

Measuring Hand-Arm Vibration (HAV) exposure is important to prevent permanent injuries, such as the White Finger / Raynaud Syndrome. Current measuring solutions require an individual attachment of those work tools that emit considerable vibrations. These sensing instruments are expensive and usually require a setup by experts. Additionally , these attached sensors are bulky and wired, which may further increase the risk of accidents in occupational safety. For an easy use, we propose using a Smartwatch to estimate the HAV doses gathered throughout the day. By utilizing the Smartwatch's Inertial Measuring Unit (IMU) that is sampling up to 800Hz, we are capable of reconstructing vibrations up to 400Hz. This range sufficiently covers the majority of harmful HAV loads that occurs with work tools. Our approach is an inexpensive solution that provides a rough estimation to indicate a vibration overload. Our solution does not require the specific tool type or datasheet.

Full text

Hand-Arm Vibration Estimation Using A Commercial Smartwatch
Denys J.C. Matthies1, Marian Haescher2, Gerald Bieber2, Suranga Nanayakkara1
1 Augmented Human Lab, Auckland Bioengineering Institute, The University of Auckland, NZ {firstname}@ahlab.org
2 Fraunhofer Institute for Computer Graphics Research IGD, Rostock, GER {firstname.surname}@igd-r.fraunhofer.de
Abstract
Measuring Hand-Arm Vibration (HAV) exposure is im-
portant to prevent permanent injuries, such as the White
Finger / Raynaud Syndrome. Current measuring solutions
require an individual attachment of those work tools that
emit considerable vibrations. These sensing instruments
are expensive and usually require a setup by experts. Ad-
ditionally, these attached sensors are bulky and wired,
which may further increase the risk of accidents in occu-
pational safety. For an easy use, we propose using a
Smartwatch to estimate the HAV doses gathered through-
out the day. By utilizing the Smartwatch’s Inertial Measur-
ing Unit (IMU) that is sampling up to 800Hz, we are capa-
ble of reconstructing vibrations up to 400Hz. This range
sufficiently covers the majority of harmful HAV loads that
occurs with work tools. Our approach is an inexpensive
solution that provides a rough estimation to indicate a vi-
bration overload. Our solution does not require the spe-
cific tool type or datasheet.
Keywords:
Hand-Arm Vibration Estimation; Smartwatch; Sensing;
Accelerometer; HAV Exposure Dose.
Introduction
There have been many research investigations looking
into understanding the risks of injury from hand-transmit-
ted vibration and whole-body vibration by means of epi-
demiological studies [1]. The most crucial impact is the
Raynaud Syndrome [3], which is a vascular spasm that
negatively affects vessel blood flow. This can be caused
when exposed to cold or stress, such as operating work
tools that emit considerable vibrations [4] to the hand and
arm. Vibrotactile perception in the fingertips can become
numb on a short-term temporal or long-term basis [6].
When the human body to the exposure to vibrations with-
out limits, symptoms such as coldness of the hands, the
legs, hypesthesia of the fingers, tremor/shivering of the
fingers, dexterity disturbance, weakness of the hands,
mobility disturbance of the elbow, shoulder/neck stiffness,
low back pain, fatigue, headache, dizziness, tinnitus, and
hearing loss can occur. These symptoms have been evi-
dent among quarry workers in developing countries such
as Vietnam [7], where occupational safety is not highly
practiced.
Different methods and technologies based on measuring
Hand-Arm Vibration [5], such as using high-sensitive ac-
celerometers [8], are used to prevent such symptoms.
These dosimeters are precise and provide sampling rates
up to 5kHz. Since these technologies are usually expen-
sive and instrumenting work tools with additional sensors
may create an increased risk in occupational safety, using
wearable technology such as a Smartwatch is a logical
step. IMUs, in particular Accelerometers and Gyroscopes,
that are implemented in Smartwatches so far only ena-
bled sampling rates of up to 100Hz without kernel modifi-
cations. Determining an accurate HAV is insufficient with
this sampling rate, since emitted vibrations can exist be-
yond this frequency. Research explored a work-around
when attempting to measure HAV exposure doses with
Smartwatches [2], such as using the accelerometer in
conjunction with the microphone to identify the tool the
worker used. Once the tool is known, it’s specific HAV ra-
tio is being looked up from a database. However, this re-
quires the system to have access to a complete database
with all HAV ratios from a great variety of work tools.
Method
In this paper, we propose an alternative approach, in
which we use the IMU of a commercial Smartwatch to cal-
culate a rough estimation of the HAV received at the
user’s wrist. This way, measuring the exact HAV expo-
sure doses is not possible because of the signal absorp-
tion, signal coupling, and transmission loss between the
vibration emitter (tool) and wrist (Smartwatch IMU).
In fact, the current Android Wear OS (2.9 – based on An-
droid 8.0.0) provides a new direct channel to assess the
acceleration sensor. Apparently, new devices will be able
to sample the IMU with a frequency up to 800Hz. Follow-
ing the Nyquist–Shannon sampling theorem, frequencies
of 400Hz can be reconstructed accordingly. Although, this
may still appear too low to sense the full spectrum of the
vibration exposure, it enables us to read most critical root-
mean-square (r.m.s.) acceleration magnitude - repre-
sented as a frequency weighting curve Wh (see Figure 1).
Figure 1. Hand-arm vibration frequency weighting curve
Wh following ISO 5349-1:2001 [9]. The highlighted areas
shows the coverable frequencies.
Therefore, our hypothesis is that a commercial Smart-
watch is sensitive enough to provide an acceptable ap-
proximation of the actual HAV exposure doses.
14th Internatio nal Conference on Ha nd-Arm-Vibration,
21-24 May 2 019, Bonn, Germany.
© 2019 Copyright is held by the owner/author(s).

Preliminary Field Study
We developed a smartwatch app running on an autarkic
Android Smartwatch. We used the model Simvalley
AW420-RX running Android 4.2. The watch has a Cortex
A7, 1 GB RAM and incorporates a Bosch BMC050 IMU,
which quantizes +/- 2 g (19,62m/s2) with 12bit. The
weight of the watch is approximately 90 grams.
Figure 2. We ran a preliminary field study in metalworking
/ manufacturing. The participants were equipped with a
smartwatch running our app, as well as with a microphone
and a GoPRO to measure the ground truth data.
We selected a window size of 128 samples of accelera-
tion tuples while using 50Hz. We assume that any harmful
HAV occur between 0–25Hz, which can be measured by
the Smartwatch. We calculated the significant accelera-
tion of the Smartwatch within the 3D-area by this formula:
!"# $
%
&' ( '
)
*+, &- ( -
.
*+, &/ ( /
)
*+
Our assumption concludes that tools with a slow motor
and slow motions would also lead to a low acceleration.
Figure 3. Distribution of measured 3D-acceleration (jerk)
measured at the wrist.
It becomes clear by Figure 3 that not only emitted tool vi-
bration is sensed, but also any other motion the worker
performs in the workshop. Therefore, we suggest classi-
fying typical motions that occur with <no Tool> and to sub-
tract this from the daily exposure doses.
In our preliminary field study, we used a low sampling rate
that is only capable of reading limited vibrations. For fu-
ture research, we suggest utilizing Smartwatches with
high sampling rates to capture a wider spectrum.
Moreover, we found that the measured 3D-acceleration
force does not directly correlate to the HAV. We believe
this is mainly due to the non-linear absorption of the vi-
bration frequencies at the fingers, the hand, and the wrist.
Also, we noticed that the wristband’s tightness to have a
significant influence on sensing the 3D-acceleration force.
Taking these factors into account, we believe reproducing
the correct HAV exposure doses could be possible.
Discussion
As demonstrated in Figure 1, the most harmful HAV oc-
curs in the lower frequency spectrum between 2.5–50Hz.
Capturing these should be prioritised. While professional
sensing tools are usually expensive and impractical, we
propose using a Smartwatch to estimate these. However,
an exact measurement is not possible due to various pa-
rameters such as the contact pressure between hand and
tool, signal absorption by the joints, tightness of the wrist-
band, etc. Aside from these factors, smartwatches are be-
coming increasingly powerful. They can provide a greater
sampling rate and are capable of sensing an increased
range of the frequency load. Nevertheless, a professional
measurement equipment is still superior. The proposed
work-around in AGIS [2] may still be the state-of-the-art
when measuring a more accurate HAV with unmodified
Smartwatches. In fact, the advance of the increased sam-
pling rate with Smartwatches can also benefit a greater
tool detection. Sampling the IMU with 800Hz may provide
enough signal characteristics to identify tools based on
the accelerometer only. Once the tool is identified, looking
up the HAV intensity ratio from the datasheet for each tool
would still be next step to calculate the daily dose. We see
Smartwatches as the gatekeeper for calculating the expo-
sure duration of harmful vibrations. In the future, we envi-
sion smart wearables to enter different industry branches,
provided that the legislator paves the way. Furthermore,
we see this technology as being capable of registering the
exposures in a cadaster, namely to distinguish between
regional and branch specific workloads.
Conclusion
The advancements in IMU sensing enables an estimation
of HAV exposure with commercial Smartwatches. Still, an
accurate measurement is yet problematic. In particular,
we are required to account for the non-linear absorption
of vibration frequencies into the hand and the tightness of
the wristband. Calculating the exact HAV dose using a
Smartwatch is feasible when running a tool detection, but
which requires a large database and thus is impractical.
References
[1] Griffin, M., Lewis, C., Bovenzi, M., Lemerle, P., & Lundström,
R. (2004). Risks of Occupational Exposures to Hand-transmit-
ted Vibration: VIBRISKS. In Proceedings of the 10th Interna-
tional Conference on Hand-Arm Vibration, Las Vegas, Nevada.
[2] Matthies, D. J. C., Bieber, G., & Kaulbars, U. (2016). AGIS:
automated tool detection & hand-arm vibration estimation
using an unmodified smartwatch. In Proceedings of iWOAR
2016 (p. 8). ACM.
[3] Blunt, R. J., & Porter, J. M. (1981). Raynaud syndrome.
In Seminars in arthritis and rheumatism (Vol. 10, No. 4, pp.
282-308). WB Saunders.
[4] Bovenzi, M. (1998). Exposure-response relationship in the
hand-arm vibration syndrome: an overview of current epidemi-
ology research. In International archives of occupational and
environmental health, 71(8), 509-519.
[5] Rossi, G. L., & Tomasini, E. P. (1995). Hand-arm vibration
measurement by a laser scanning vibrometer. In Measure-
ment, 16(2), 113-124.
[6] Yu, Gongqiang, Brammer, Anthony J., Cherniack, Martin G.
(2015). Vibrotactile perception thresholds at the fingertip meas-
ured with and without a surround. In Proceedings of the 13th
International Conference on Hand-Arm Vibration.
[7] Futatsuka, M., Shono, M., Sakakibara, H., & Quan, P. Q.
(2005). Hand arm vibration syndrome among quarry workers in
Vietnam. In Journal of occupational health, 47(2), 165-170.
[8] Maeda, S., & Dong, R. G. (2004). Measurement of hand-trans-
mitted vibration exposures. In Proceedings of the 10th Interna-
tional Conference on Hand-Arm Vibration (p. 89).
[9] Bovenzi, M. (2012). Epidemiological evidence for new
frequency weightings of hand-transmitted vibration. In
Industrial health, 50(5), 377-387.
Corresponding Address
d.matthies@auckland.ac.nz