See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/340333818

EVALUATION OF A COMMERCIAL FORCE SENSOR FOR REAL TIME

APPLICATIONS

Article in ICIC Express Letters · May 2020

DOI: 10.24507/icicelb.11.05.421

CITATIONS READS

8 565

4 authors:

Ahmed Almassri Chikamune Wada

Kyushu Institute of Technology Kyushu Institute of Technology
23 PUBLICATIONS 279 CITATIONS 180 PUBLICATIONS 607 CITATIONS

SEE PROFILE SEE PROFILE

Keiichi Horio W. Z. Wan Hasan

Kyushu Institute of Technology Universiti Putra Malaysia
113 PUBLICATIONS 538 CITATIONS 222 PUBLICATIONS 2,379 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by W. Z. Wan Hasan on 01 April 2020.

The user has requested enhancement of the downloaded file.

ICIC Express Letters
Part B: Applications ICIC International ⃝2020
c ISSN 2185-2766
Volume 11, Number 5, May 2020 pp. 421–426

EVALUATION OF A COMMERCIAL FORCE SENSOR

FOR REAL TIME APPLICATIONS

Ahmed M. M. Almassri1,∗ , Wan Zuha Wan Hasan2 , Chikamune Wada1

and Keiichi Horio1
1
Graduate School of Life Science and Systems Engineering
Kyushu Institute of Technology
2-4 Hibikino, Wakamatsu-ku, Kitakyushu 808-0196, Japan
∗
Corresponding author: [email protected]
2
Department of Electrical and Electronic Engineering
Faculty of Engineering
Universiti Putra Malaysia
Serdang 43400, Selangor Darul Ehsan, Malaysia
[email protected]
Received November 2019; accepted February 2020

Abstract. A commercial piezoresistive FlexiForce sensor can be of great interest in

biomedical robotic applications due not only to its advantageous material which will not
be affected by metallic and ferromagnetic materials but to its dimension as well. In this
paper, the evaluation of FlexiForce sensor is performed in the case of grasping robotic
hand glove for medical application. The evaluation method is done through a dynamic
calibration based on three experimental tests which are performed with different calibra-
tion features to accurately identify the behaviour of force sensor over time. Our em-
pirical evaluation shows that the method used is computationally efficient, and that it
has advantages over the traditional method. Quantitative results show and identify the
characteristics of the force sensor pattern including the systematic errors such as lack of
linearity, hysteresis and non-repeatability. The proposed method is a useful approach for
evaluating the performance of any measurement force sensor in real time environments.
Keywords: Force sensor, Real time applications, Force measurement system, Grasping
robotic hand, Evaluation, Dynamic calibration

1. Introduction. In various robotics applications especially, in grasping robotic hand,

force measurement is required to ensure safety and comfort in human and robot interaction
[1]. For example, the force distribution is essential in most human movements such as
grasping, writing and walking. An increasing number of medical robotics applications
are involved by human/robot interactions and force control. For these applications, high
accurate force measurement system is becoming very necessary.
Different technologies produce various types of force sensors which have been used and
applied in several fields and applications [2]. Piezoresistive force sensor such as FlexiForce
A201 (1-617-464-4500) with a standard force range of 0-25 lb (110 N) is one of the famous
sensors widely used in force control tasks [3]. The evaluation of such sensors to deter-
mine its performance for real time or dynamic applications is important. For a growing
number of medical applications involving the robotic hand glove for real time grasping
purpose, these sensors seem to be well adapted to force-control tasks [4,5]. In literature,
many researchers have proposed a characterization of piezoresistive force sensors using
traditional calibration. Most of them have focused on static evaluation but only a few re-
searchers have considered their response to dynamic loading [6]. However, the traditional
calibration process for the FlexiForce sensor is a time-consuming task because it is usually
DOI: 10.24507/icicelb.11.05.421

421
422 A. M. M. ALMASSRI, W. Z. W. HASAN, C. WADA AND K. HORIO

done through manual and repetitive identification. In addition, a traditional computa-

tional method of FlexiForce sensor is inadequate for solving the problem of systematic
errors since it is extremely difficult to resolve the mathematical formula among multiple
confounding pressure variables [7]. Particularly, during the grasping object using robotic
hand glove, the sensor loading will vary due to the patient hand movements and thereby
inaccurate force measurements appear. Accordingly, adequate dynamic evaluation of the
low-cost piezoresistive force sensors offers an interesting solution for force measurement in
this context, because of successful identification force patterns including the systematic
errors such as lack of linearity, hysteresis and non-repeatability.
In this paper, the focus is put on the dynamic evaluation of the multi commercial low-
cost piezoresistive sensors (FlexiForce). The existence of nonlinearities, non-repeatability
and hysteresis in their dynamic response is outlined, and the opportunity to compensate
these systematic errors is suggested.
2. Problem Statement. The advantages of FlexiForce sensors include low cost, inter-
esting dimensions with a very small thickness, easy to integrate and insensitive to magnetic
fields which can be of great interest in biomedical applications [1]. Nevertheless, in real
time applications, as time elapses, the FlexiForce sensor parameters change its charac-
teristics due to its material creep and cause measurement errors leading to nonlinearity
output [7-10]. These negatively affect the calibrated output data and as a result, inac-
curate force measurements appear. Furthermore, the elastic modulus and hardness that
leads to the creep effects of material are dependent on holding time, maximum load, and
loading/unloading rates for nanoindentation in polymer. Therefore, an appropriate set
of calibration method to dynamically evaluate the force sensor is necessary to avoid such
measurement error. In addition, it is considered as a first step towards successful sensor
compensation.
3. Sensor Evaluation. In this work, a piezoresistive FlexiForce sensor was utilized to
be evaluated. This sensor relies on the resistive method where the resistance is inversely
proportional to applied force. It can emit signals within millivolts range; therefore an
amplification is needed. However, prior to the evaluation tests, all FlexiForce sensors
were calibrated in real time environments.
3.1. Calibration. To evaluate how the FlexiForce sensor behaves over time, a dynamic
calibration was performed using the CT3 Texture Analyser machine as demonstrated in
Figure 1. It can be observed that different hardware components incorporated together
in the construction of a force measurement system in order to successfully calibrate the
multi force sensors. The components used are load cell with strain amplifier to be used
as a reference signal, data acquisition device (DAQ) to calibrate signal from load cell,
FlexiForce sensor’s adapter to amplify sensor’s signal, readout circuit to extract sensor’s
signal, trigger circuit using Arduino to synchronize between two signals (sensor and load
cell signals). In experimental setup, there are two signals to be extracted and the com-
ponents are connected accordingly. The hardware connections were as follows. In sensor
signal, the FlexiForce sensor was connected to the adapter. The output of adapter was
connected to the input of readout circuit which was connected to the PC through USB
connection. On the other hand, the load cell signal (reference) is extracted by connect-
ing the load cell with the strain amplifier which amplified the signal output and then
transmitted to the analogue input of DAQ connected to the PC. The Arduino is used to
synchronize between two signals by applying a trigger circuit which enhances the calibra-
tion accuracy and acquires only the signals in the area we need. In addition, it can save
hardware bandwidth and memory. LabVIEW program was used to acquire the calibrated
signals to be evaluated.
 ICIC EXPRESS LETTERS, PART B: APPLICATIONS, VOL.11, NO.5, 2020 423

Figure 1. Dynamic calibration of FlexiForce sensor using CT3 Texture Analyser

Subsequently, the calibration process was done by using CT3 Texture Analyser machine
to apply compressive loads to the sensing area of FlexiForce sensor. However, in order to
ensure the compatibility of the proposed calibration method with the desired application
which is grasping robotic hand glove, three experimental calibration tests were performed.
During each test, different loads of maximum pressure and holding times were applied.
By this calibration method, the features of force pattern are effectively identified and the
relationship between applied load and sensor output is sufficiently established. Table 1
explains the differences between three calibrations executed. Based on these experiments,
the dynamic evaluations of multi force sensors were investigated.

Table 1. The calibration of FlexiForce sensor based on the three experiments

Experiment No. Calibration

Applied force Holding time
Data points No. of cycles
(N) (sec)
1 0∼44.13 10 132,243 28
2 0∼19.61 5 130,990 39
3 > 30 – 122,149 83
Note Sampling rate is 100 Hz

In the first experiment, a dynamic load ranging from zero to 44.13 N was applied to
each of the FlexiForce sensor via standard cylinder probe (diameter 52.4 mm, sized to fit
within the active area of the sensors (9.53 mm) and the puck (8.75 mm)). The puck put
between the probe and the sensor’s active area as well as the cylinder probe made from
plastic clear acrylic were inserted to ensure that the load was evenly distributed across
the active sensing area, avoiding local high pressures at edges, for example. The larger
load values were chosen because the typical maximum finger force produced by humans
is no higher than 30 N [11].
Consequently, and similarly, with previous calibration, the experiment was repeated
but with different applied forces and holding times. In the second experiment, an applied
424 A. M. M. ALMASSRI, W. Z. W. HASAN, C. WADA AND K. HORIO

force from zero to 19.61 N was loaded while more than 30 N was performed in the third
experiment as a random applied force. A set of holding times was also selected to imitate
the desired robotic hand grasping as well as to find an optimum holding time limit for
the material of sensor. The selected holding times were 10, 5 and zero sec at maximum
load for experiments 1, 2 and 3 respectively. The sensor output was recorded by readout
circuit at 100 Hz with multiple load-unload cycles tests run on each FlexiForce sensor.
Simultaneously, the load cell output (reference) was recorded by DAQ at the same sample
rate to match the sensor output. Then the sensor output was aligned with the applied
load. During each experimental test, load-unload cycles were applied and repeated with
more than 20 min and the data points output were recorded as outlined in Table 1.

4. Results. In this section, we present the real time calibration data of the force sensor
and its evaluation. Figure 2 shows the calibrated data of force sensor obtained from
three experiments within one hour of calibration as discussed previously. The blue, green
and red colours were related to the data of experiment 1, experiment 2 and experiment
3 respectively. It can be observed that the force patterns were clearly different from
experiment 1, experiment 2 and experiment 3 due to the dissimilar features of applied
forces and holding time. Indeed, different input force patterns and holding times for the
same sensor enhance the accuracy and efficiency towards identification of the behaviour
of force sensor over time so that it is easy to compensate the systematic errors in later
process.

Figure 2. The calibration data output of FlexiForce sensor based on three

experiments in real time

Evaluation. To additionally clarify the evaluation of FlexiForce sensor over time, Figure
3 represents the behaviour of sensor output and its corresponding load cell reference
based on experiment 2 as example. Throughout the repeated load-unload cycles tests,
a decrease with fluctuation in the output voltage over time was observed even the same
pressure forces were applied in each cycle. This causes the systematic errors that would
be evaluated in this paper. On the whole, for evaluation and performance purpose of
calibration method, the Mean Absolute Deviation (MAD), Mean Square Error (MSE),
Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were
0.288, 0.098, 0.313 and 79.07% respectively. The results were based on the calibration
data of experiment 2 while 39 cycles tests were repeated. However, the compensation of
systematic errors for FlexiForce sensor can be found in the paper published in [7] under
the same author.
 ICIC EXPRESS LETTERS, PART B: APPLICATIONS, VOL.11, NO.5, 2020 425

Figure 3. The relative measured voltage changes in the sensor over time
based on experiment 2

Figure 4. Measured sensor versus applied force for hysteresis, nonlinearity

and non-repeatability calculation

Figure 4 investigates the evaluation of FlexiForce sensor including the repeatability,

hysteresis and linearity. The first and last load-unload cycles of sensor calibration over
time are presented in order to evaluate the sensors behavior under a full scale.
1) Repeatability. The repeatability refers to a sensor’s ability to give identical output
for the same input. It is evident from the results that, the FlexiForce has a lack of
repeatability under the same applied force and experimental conditions. Indeed, the
characteristics of first and last cycle were totally different especially after 5 min of sensor
calibration. The mean and SD of the readings for 39 cycles test based on experiment 2
at maximum applied force (19.61 N) were 0.3299 and ± 0.02, respectively. Furthermore,
the coefficient of variation was 6.44%.
 426 A. M. M. ALMASSRI, W. Z. W. HASAN, C. WADA AND K. HORIO

2) Linearity. Linearity is a measure of the proportionality of the sensor’s response to

the applied load over the range of loading [12]. The mean and SD of the readings for
39 cycles test at 5, 10, 15 and 19 N were 0.039 ± 0.002, 0.109 ± 0.011, 0.211 ± 0.020,
0.307 ± 0.029, respectively. In addition, the coefficient of variation for 5, 10, 15 and 19 N
was 4.74%, 10.35%, 9.32% and 9.45% respectively. The quantitative results proved that
the FlexiForce has a lack of linearity over the range of loading.
3) Hysteresis. It is the difference in the sensor output response during increased loading
and decreased loading at the same force [12]. Similar to previous evaluation, the sensor
was examined based on procedures of experiment 2 under 39 repeated cycles test. The
hysteresis of first and last cycles of the sensor output was calculated at the midpoint in
voltage measurements and the result was 3.273% and 2.282% respectively. Furthermore,
the mean ± SD of the maximum difference in output for increasing and decreasing loads
in first and last cycles was 0.221 ± 0.021 and 0.317 ± 0.026 respectively. A typical example
of the data is shown in Figure 4.
5. Conclusions. From this study, we can conclude that the evaluation of a commercial
FlexiForce sensor for real time applications is crucial in order to accurately identify the
behaviour of sensor over time. This method definitely helps to overcome the issue of
the systematic errors which negatively affect the calibrated output data. Quantitative
results of three experimental tests were performed. The evaluation results show the lack
of linearity, non-repeatability and hysteresis in the FlexiForce sensor. Furthermore, a
fluctuation in output force over time was detected even the same experimental conditions
were applied. With this evaluation method, other sensor measurement systems can be
evaluated. Nevertheless, the sensor provided useful force data, adding to the knowledge
of what happens at the grasp during a robotic hand glove.
REFERENCES
[1] C. Lebosse, P. Renaud, B. Bayle and M. de Mathelin, Modeling and evaluation of low-cost force
sensors, IEEE Transactions on Robotics, vol.27, pp.815-822, 2011.
[2] A. M. Almassri, W. Wan Hasan, S. A. Ahmad, A. J. Ishak, A. Ghazali, D. Talib and C. Wada,
Pressure sensor: State of the art, design, and application for robotic hand, Journal of Sensors,
vol.2015, 2015.
[3] A. Almassri, C. Wada, W. Wan Hasan and S. A. Ahmad, Auto-grasping algorithm of robot gripper
based on pressure sensor measurement, Pertanika J. Sci. Technol., vol.25, pp.113-121, 2017.
[4] L. Birglen and C. M. Gosselin, Fuzzy enhanced control of an underactuated finger using tactile and
position sensors, 2005 IEEE International Conference on Robotics and Automation, pp.2320-2325,
2005.
[5] H. Zhang and N. N. Chen, Control of contact via tactile sensing, IEEE Transactions on Robotics
and Automation, vol.16, pp.482-495, 2000.
[6] C. Lebosse, B. Bayle, M. de Mathelin and P. Renaud, Nonlinear modeling of low cost force sensors,
2008 IEEE International Conference on Robotics and Automation, pp.3437-3442, 2008.
[7] A. Almassri, W. Wan Hasan, S. Ahmad, S. Shafie, C. Wada and K. Horio, Self-calibration algorithm
for a pressure sensor with a real-time approach based on an artificial neural network, Sensors, vol.18,
p.2561, 2018.
[8] J. Rivera, M. Carrillo, M. Chacón, G. Herrera and G. Bojorquez, Self-calibration and optimal re-
sponse in intelligent sensors design based on artificial neural networks, Sensors, vol.7, pp.1509-1529,
2007.
[9] J. R. Rivas, F. Lou, H. Harrison and N. Key, Measurement and calibration of centrifugal compres-
sor pressure scanning instrumentation, The Summer Undergraduate Research Fellowship (SURF)
Symposium, 2015.
[10] D. Giovanelli and E. Farella, Force sensing resistor and evaluation of technology for wearable body
pressure sensing, Journal of Sensors, vol.2016, 2016.
[11] G. H. Büscher, R. Kõiva, C. Schürmann, R. Haschke and H. J. Ritter, Flexible and stretchable
fabric-based tactile sensor, Robotics and Autonomous Systems, vol.63, pp.244-252, 2015.
[12] M. Ferguson-Pell, S. Hagisawa and D. Bain, Evaluation of a sensor for low interface pressure appli-
cations, Medical Engineering & Physics, vol.22, pp.657-663, 2000.

View publication stats