Autonomous Energy Management system achieving

piezoelectric energy harvesting in Wireless Sensors

1
Sara Kassan, 2Jaafar Gaber, 1Pascal Lorenz
1
Univ. Haute-Alsace UHA, 34 rue Grillenbreit, 68008 Colmar Cedex France, email: {sara.kassan,
pascal.lorenz}@uha.fr
2
FEMTO-ST Institute, Univ. Bourgogne Franche-Comté UBFC, Univ. Technology Belfort- Montbéliard UTBM,
13 rue E.T.Mieg, 90010 Belfort Cedex France, email: [email protected]

Abstract— Wireless Sensor Networks (WSNs) are extensively magnetostrictive and piezoelectric [6] [7] [8]. Table 1 compares
used in monitoring applications such as humidity and temperature the features of different type of vibration energy harvesting
sensing in smart buildings, industrial automation, and predicting techniques.
crop health. Sensor nodes are deployed in remote places to sense
the data information from the environment and to transmit the TABLE Ⅰ: comparison of the different vibrational types of harvesting
sensing data to the Base Station (BS). When a sensor is drained of mechanisms
energy, it can no longer achieve its role without a substituted
source of energy. However, limited energy in a sensor’s battery Advantages Disadvantages
Technique type
prevents the long-term process in such applications. In addition,
-Strong coupling -Difficulty to be
replacing the sensors’ batteries and redeploying the sensors is very
-No smart materials integrated by small
expensive in terms of time and budget. To overcome the energy Electromagnetic Wireless Sensors
limitation without changing the size of sensors, researchers have -Low voltage
proposed the use of energy harvesting to reload the rechargeable generated
battery by power. Therefore, efficient power management is -Compatible with -Voltage source or
required to increase the benefits of having additional manufacturing load Required
environmental energy. This paper presents a new self- Electrostatic Wireless Sensors -Mechanical
management of energy based on Proportional Integral Derivative -Strong tensions constraints required
controller (PID) to tune the energy harvesting and Microprocessor generated -Capacitive
Controller Unit (MCU) to control the sensor modes. - High coupling - Difficult small
- No depolarization Wireless Sensor
Keywords—WSN; Network lifetime; Energy harvesting; - Flexibility integration
Piezoelectric; Power consumption; Proportional Integral Derivative Magnetostrictive - Fragility
controller PID; energy consumption; Microprocessor Controller - Nonlinear Effects
Unit MCU; Mica2 motes sensors. - Possible need for
magnets polarization
I. INTRODUCTION Compact -Depolarization
-Compatible with -Fragility of the piezo
The possibility to avert replacing drained batteries is integration small layers
extremely important in Wireless Sensor Networks to avoid the Wireless Sensors -High impedance
Piezoelectric
high cost and time to replace batteries and redeploy them. -Strong tensions
Therefore, energy harvesting has attracted researchers and generated
developers to recharge low power devices as Vivo-Nano- -Strong coupling for
single crystals
Robots, MEMS, Claytronics and WSN [1] [2] [3]. Energy
harvesting is the method of extracting energy from the
Among them in this paper, we focus on the control of the
environment through different sources of energy. It is
low-level energy harvesting by the most prevalent technique
considered an emerging and reasonably mature technology to
piezoelectric vibration system for WSN. The diagram of a
overcome the limited lifetime of battery-operated wearable
piezoelectric energy harvesting system is presented in Figure
devices and allows continuous recharging of the energy stored
Fig. 1; it can be shortened into three essential components:
during use. The environmental energy for scavenging is mostly
provided by ambient light (artificial lighting and solar lighting), piezoelectric devices, converters, and electrical energy storage.
In addition to these components, we introduce two controllers
radio frequency, thermal sources, kinetic and vibration sources
one to tune the energy harvesting and the other to manage the
[4] [5]. To harvest energy from vibration, there are different
energy in the sensor according to its activities needs. This new
techniques such as electrostatic, electromagnetic,
self-organization power system will be able to insure the A local lossless neighborhood indexing sequence (NIS)
prolongation of the WSN lifetime. compression algorithm for data compression in WSN is
proposed in [12]. The NIS approach dynamically assigns
shorter length code information for each character in the input
sequence, by exploiting the occurrence of neighboring bits, and
every data packet is decompressed independently from others.
Therefore, NIS algorithm helps to decrease the network load,
which results in low data packet loss.
A topology control protocol based on learning automaton is
proposed in [13]. The mechanism chooses the proper smallest
transmission range of the node using the reinforcement signal
produced by the learning automaton of neighbor sensor nodes.
Figure 1: Block diagram of energy harvesting model Consequently, the choice of transmission range affects the
energy consumed by nodes and the overall network lifetime.
The paper is organized as follows: Section Ⅱ provides a A hierarchical clustering protocol based on sensors location
review with the different mechanisms to save energy in WSN. and energy consumption using non-cooperative game theory
In section Ⅲ, we present the global proposed model to manage (GT) approach to extend the WSN lifetime is proposed in [14].
the harvested energy in WSN. In section Ⅳ, we describe the The GT permits a sensor to decide between two actions: to enter
piezoelectric harvesting energy model. In particular, we show a game and transmit a message or to stay out of the game and
how to adapt the PID controller to tune power in the vibration- harvest energy to charge its battery to reach the Nash
based model in WSN and how to tune the residual energy in the Equilibrium (NE) solution for mixed strategies.
rechargeable batteries according to the sensor’s needs using In [15], a Genetic algorithm-based Energy-efficient
MCU. In section Ⅴ, we present the simulation results. Finally, Clustering and Routing Approach (GECRA) is presented. This
we provide a conclusion of the paper in section Ⅵ. algorithm aims to calculate the total energy consumed by all
sensor nodes where the algorithm encodes the clustering
scheme and routing scheme together in the same chromosome.
II. RELATED WORK
Prolonging WSN lifetime is one of the most critical It is worth noting that these research efforts aim mainly to
challenges in WSNs. In the last few years, researchers have reduce the energy consumption in the stage of wireless sensors
proposed different methods to increase the WSN lifetime via deployment and during sensors activities, such as
saving approaches. The main goal of these methods is saving communication and sensing data, via energy saving
energy, during the communication between sensors, without mechanisms. Therefore, the WSN lifetime will be extended, but
compensation for the energy dissipated via scavenging ambient for a limited time. Instead of applying the energy saving
energy. Specifically, we can classify energy methods in two methods, energy harvesting can be used to recharge the sensor’s
types: energy saving mechanisms to decrease energy battery. The objective is to increase the power stored in the
consumption and energy harvesting to recompense the energy sensor’s battery thereby extending the overall network lifetime
consumed during sensors’ activities. depending on the availability of ambient energy harvesting
Most of the proposed approaches in literature are saving resources. This article presents a new self-management of
energy-based mechanisms such as sensors deployment energy based on a Proportional Integral Derivative (PID)
strategies, data-oriented techniques, topology control methods, controller to tune the energy harvesting and a Microprocessor
and energy routing protocols. For example, in [9], the authors Controller Unit (MCU) to control the sensor modes.
propose a statistical strategy for the nodes’ deployment. The
statistical node deployment strategy uses the Quasi-random
III. PROPOSED SELF POWER MANAGER SYSTEM MODEL
method of low-discrepancy sequences to increase the lifetime
and the coverage of the network. The target is an autonomous system to harvest energy
The attentiveness to data-oriented techniques has increased spontaneously upon reaching a threshold of energy
recently. These techniques reduce the size of information via consumption. The system proposed is composed of two main
compression. Accordingly, data-oriented techniques are blocks as designed in figure Fig. 2: The Battery Management
efficient to save energy during transmission and reception of this Subsystem block that manages a rechargeable battery or a super-
data information. In addition, the reduction of data is not only capacitor of the sensor and monitors its states via two
efficient in saving energy but also in saving more memory in (Management Controller Unit) MCU modes. In on mode, the
the sensors [10] [11]. For example, in [11], the authors propose sensor is active and can sense information from its environment,
a Distributed Distortion-Rate Optimized Compressed Sensing write/read data from memory, and communicate messages with
(DQCS) method to compress data in WSN under a complexity- other wireless sensors. During achieving activities, the sensor
constrained encoding, which minimizes a weighted sum consumes energy until a threshold and the MCU turns to off
between the mean square error (MSE) signal reconstruction mode. In the off mode, the block piezoelectric energy harvesting
distortion and the average encoding rate. structure is triggered. This block is controlled via a control loop
feedback using a Proportional Integral Derivate controller PID
that scavenges ambient energy to recharge the sensor’s battery. 𝐸𝑟 (𝑡) = 𝐸ℎ (𝑡) + 𝐸𝑟 (𝑡 − 1) − 𝐸𝑐 (𝑡) (3)

In this paper, we assume that the vibration resource exists It is required to find the total energy consumption of a node
continuously. Moreover, the piezoelectric harvesting energy in the treatment of one data packet information. The total
system is modeled by an equivalent Mass-Spring-Damper energy consumed by a wireless sensor node includes sensing,
(MSD) model (discussed in the section below Ⅳ) with a PID processing, communicating data information, switching radio
controller that adapts the quantity of energy harvested by the model states and switching MCU modes.
sensor node to still alive and continue its processes. The
harvesting energy will be tuned to reduce the error 𝑒 between  Sensing energy consumption
the setpoint or reference energy 𝑅𝐸.
Sensing energy costs depends on the type of sensors. For
example, the temperature sensors consumed less important
When the error 𝑒 is bigger than a threshold 𝑇, the sensor energy than gas sensors. The sensor node can contain diverse
node dissipates the majority of its residual energy and the sensors, and each one has its individual energy consumption
microcontroller MCU switches to off mode and lets the sensors attributes. Generally, the sensing energy consumption for a
node harvest energy. Otherwise, the MCU is still in on mode wireless sensor can be expressed as follows:
and continues normally its activities. The energy consumption
for one sensor depends on the energy consumed for its 𝐸𝑆 = 𝛼 × 𝑁 × 𝑉𝑑𝑐 × 𝐼𝑠𝑒𝑛𝑠 × 𝑇𝑠𝑒𝑛𝑠 (4)
activities. The system block diagram is shown in Figure Fig. 2.
where 𝐼𝑠𝑒𝑛𝑠 is the required amount of current, and 𝑇𝑠𝑒𝑛𝑠 is the
duration to detect and collect 𝑁 bits data of information and 𝛼 is
the percentage to compress 𝑁 bits sensing data information.
 Processing energy consumption
The sensor consumes energy to read the data message and to
write it in its memory. The processing energy consumption
could be calculated by:
𝛼 × 𝑁 × 𝑉𝑑𝑐
𝐸𝑃 = × (𝐼𝑊𝑟𝑖𝑡𝑒 × 𝑇𝑊𝑟𝑖𝑡𝑒 + 𝐼𝑅𝑒𝑎𝑑 × 𝑇𝑅𝑒𝑎𝑑 ) (5)
8

where 𝐼𝑊𝑟𝑖𝑡𝑒 and 𝐼𝑅𝑒𝑎𝑑 are the necessary amount current to write
and read one byte data. 𝑇𝑊𝑟𝑖𝑡𝑒 and 𝑇𝑅𝑒𝑎𝑑 are the necessary
duration to treat the 𝐿(𝑆𝑖 ) data information.
 Communicating energy consumption
The energy consumed to transmit and receive data
messages is computed following the first-order wireless
Figure 2: Architecture of the global system of wireless sensor communication model for the radio hardware illustrated in
composed of two main blocks one to manage energy consumed and Figure Fig. 3 [16].
residual energy in the battery and another one to harvest energy from
the ambient source 𝐸𝑇 (𝑑)

In what follows, the model is formulated mathematically in

terms of energy consumption by sensor nodes. d
The energy error 𝑒 can be written as follows: 𝛼 × 𝑁 × 𝐸𝑒𝑙𝑒𝑐
𝐸𝑅 (𝑑)
𝑒(𝑡) = 𝑅𝐸 − 𝐸𝑟 (𝑡) (1)

where the reference energy 𝑅𝐸 is a fixed energy harvesting to

charge the sensor’s battery and 𝐸𝑟 is the current residual energy 𝛼 × 𝑁 × 𝐸𝑒𝑙𝑒𝑐
in the sensor’s battery.
1 Figure 3: First order radio transceiver model
𝑒(𝑡) = 𝐶(𝑉𝑟𝑒𝑓 2 − 𝑉𝑟 (𝑡)2 ) (2)
2
Transmitter expends energy to turn on the radio electronics and
where the reference voltage 𝑉𝑟𝑒𝑓 is the desired fixed voltage and the power amplifier. The energy required to transmit 𝐿 bits data
message is:
𝑉𝑟 (𝑡) is the current voltage for the sensor’s battery in time 𝑡.
 𝛼 × 𝑁 × 𝐸𝑒𝑙𝑒𝑐 + 𝛼 × 𝑁 × 𝐸𝑓𝑠 × 𝑑 2 𝑤ℎ𝑒𝑛 𝑑 < 𝑑0 The mechanical characteristics of a piezoelectric heap can be
𝐸𝑇 = { represented by an equivalent Mass-Spring-Damper (MSD)
𝛼 × 𝑁 × 𝐸𝑒𝑙𝑒𝑐 + 𝛼 × 𝑁 × 𝐸𝑚𝑝 × 𝑑 4 𝑤ℎ𝑒𝑛 𝑑 > 𝑑0
model with one degree of freedom based on the proposed model
(6) in [17]. It consists of an equivalent mass 𝑀𝑝𝑧𝑡 linked with a
where 𝐸𝑒𝑙𝑒𝑐 represents the energy consumed to transmit or spring that has a constant coefficient 𝐾𝑝𝑧𝑡 and a damper with a
receive 1 bit message, the constants 𝐸𝑓𝑠 and 𝐸𝑚𝑝 depend on the damping coefficient 𝐷𝑝𝑧𝑡 and input displacement of the
transmitter amplifier model. 𝐸𝑓𝑠 is for the free space model, 𝐸𝑚𝑝 frame 𝑦(𝑡).
is for multipath model, 𝑑 is the distance between the transmitter The MSD model of the piezoelectric heap is presented in Figure
and the receiver and 𝑑0 is a threshold distance calculated as Fig. 5.
follows:

𝐸𝑓𝑠
𝑑0 = √ ⁄𝐸 (7)
𝑚𝑝

and the energy consumed by the radio to receive 𝐿(𝑆𝑖 ) bits data
information is defined by:
𝐸𝑅 = 𝛼 × 𝑁 × 𝐸𝑒𝑙𝑒𝑐 (8)
 Switching the microcontroller (MCU) mode energy
consumption
The sensor wastes energy by switching between the MCU
modes. In this paper, we just take into consideration the on mode
and the off mode. The energy cost for computational MCU mode
can be expressed as:
𝐸𝑆𝑤𝑖𝑡𝑐ℎ−𝑀𝐶𝑈 = 𝑉𝑑𝑐 × (𝐼𝐴𝑐𝑡𝑖𝑣𝑒 × 𝑇𝐴𝑐𝑡𝑖𝑣𝑒 + 𝐼𝑆𝑙𝑒𝑒𝑝 × 𝑇𝑆𝑙𝑒𝑒𝑝 ) (9)

IV. PIEZOELECTRIC ENERGY HARVESTING MODEL

In this section, the piezoelectric energy harvesting model is Figure 5: Equivalent MSD model
presented and then a PID controller is adapted to the model to
tune the energy harvesting and the stability of the system. The displacement 𝑧(𝑡) of the mass relative to the
frame is characterized by the differential equation (3), with the
For the piezoelectric energy harvesting model, the mass 𝑀𝑝𝑧𝑡 , the spring constant; the equivalent short circuit
piezoelectric layers connected on a cantilever beam can be stiffness; 𝐾𝑝𝑧𝑡 , the damping coefficient 𝐷𝑝𝑧𝑡 and the input
considered as a simple energy-harvesting device. The displacement of the frame y(t). 𝑉 is the voltage across the load
dimensions of piezoelectric, schematic and coordinate directions resistance, 𝐶 is the capacitance, 𝑅 is the resistance and 𝑖 is the
are shown in Figure Fig. 4. When the cantilever vibrates, at the electric current. The governing equation can be found by
first mode, the force which proceeds on the piezoelectric layers applying Newton’s second law.
can be simplified to a 1-D model and regarded as a force 𝑓 acting
on the lateral surface. 𝑀𝑝𝑧𝑡 𝑧̈ (𝑡) + 𝐷𝑝𝑧𝑡 𝑧̇ (𝑡) + 𝐾𝑝𝑧𝑡 𝑧(𝑡) = 𝑓(𝑡) (10)

where
𝑓(𝑡) = −𝑀𝑝𝑧𝑡 𝑦̈ (11)

We can present the equivalence mechanical equation of (13)

by the electrical equation by applying Kirchhoff’s law for an
RLC circuit:

1
𝐿𝑞̈ (𝑡) + 𝑅𝑞̇ (𝑡) + 𝑞(𝑡) = 𝑉(𝑡) (12)
𝐶

Where 𝐿 is the inductance, 𝑅 is the resistance, 𝐶 is the battery’s

capacitance and 𝑞(𝑡) is the quantity of electricity generated by
the electromotive force voltage 𝑉(𝑡).

The electro-mechanical analogy of our system is represented as

Figure 4: Schematic of piezoelectric harvesting structure follows in Table 2:
 For the closed-loop, there are three transfer functions between
TABLE Ⅱ: Electro-mechanical analogy of piezoelectric energy the setpoint and the output under the PID controller and they are
harvesting system given by:
 The sensitivity function:
Mechanical system Electrical system
1
Position 𝑧(𝑡) Electrical charge 𝑞(𝑡) 𝐸(𝑠) = (17)
1 + 𝑃(𝑠)𝑈(𝑠)
Mass 𝑀𝑝𝑧𝑡 Inductance 𝐿

Resistance 𝑅  The complementary sensitivity function:

Damping coefficient 𝐷𝑝𝑧𝑡
Inverse of capacitance 𝐶
Spring constant 𝐾𝑝𝑧𝑡 𝑃(𝑠)𝑈(𝑠)
Voltage 𝑉(𝑡)
𝑇(𝑠) = (18)
External force 𝑓(𝑡) 1 + 𝑃(𝑠)𝑈(𝑠)

Applying the Laplace transform on equation (13) with zero (𝑘𝑑 𝑠 2 + 𝑘𝑝 𝑠 + 𝑘𝑖 )

𝑇(𝑠) = (19)
initial conditions: (𝑀𝑝𝑧𝑡 𝑠 3 + (𝐷𝑝𝑧𝑡 + 𝑘𝑑 )𝑠 2 + (𝐾𝑝𝑧𝑡 + 𝑘𝑝 )𝑠 + 𝑘𝑖 )

𝑍(𝑠) = ℒ{𝑧(𝑡)} 𝑎𝑛𝑑 𝐹(𝑠) = ℒ{𝑓(𝑡)} (13)  The input sensitivity function:
𝑈(𝑠)
𝑀𝑝𝑧𝑡 𝑠 2 𝑍(𝑠) + 𝐷𝑝𝑧𝑡 𝑠𝑍(𝑠) + 𝐾𝑝𝑧𝑡 𝑍(𝑠) = 𝐹(𝑠) (14) 𝐼(𝑠) = (20)
1 + 𝑃(𝑠)𝑈(𝑠)
The open-loop plant transfer function between the input Proportional Integral Derivative (PID) controllers are
vibration force; 𝐹(𝑠) ant the output displacement 𝑍(𝑠) is given extensively used to provide robustness and optimal performance
by: for stable, unstable, and nonlinear processes [18]. It can be
𝑍(𝑠) simply implementable in analog or digital form. Further, it
𝑃(𝑠) = (15)
𝐹(𝑠) supports tuning and online retuning based on the performance
requirement of the process to be controlled. The output of a PID
1 controller is calculated in the time domain from the response
𝑃(𝑠) = (16)
(𝑀𝑝𝑧𝑡 𝑠2 + 𝐷𝑝𝑧𝑡 𝑠 + 𝐾𝑝𝑧𝑡 ) error and it can be expressed mathematically by the equation
(21) as follows:
𝑇
Figure Fig. 6 shows the displacement of the mass 𝑀𝑝𝑧𝑡 produced 𝑑𝑒(𝑡)
𝑢(𝑡) = 𝑘𝑑 + 𝑘𝑝 𝑒(𝑡) + 𝑘𝑖 ∫ 𝑒(𝑡) (21)
by an external vibration effort 𝐹. The DC gain (i.e, the amplitude 𝑑𝑡 0
ratio between the steady state response and the step input)
is 1/𝐾𝑝𝑧𝑡 , so 0.8116 is the final value of the output for a unit where 𝑢(𝑡) and 𝑒(𝑡) are the control and the error signals
step input. The rise time is about 0.4 𝑠, the settling time is too respectively, 𝑘𝑑 , 𝑘𝑝 and 𝑘𝑖 are the parameters to be tuned.
long, about 6.562 𝑠 and the overshoot is high, about 50.6432. Taking the Laplace transform of equation (21), the transfer
function for a PID controller can be expressed as follows:

𝑘𝑑 𝑠 2 + 𝑘𝑝 𝑠 + 𝑘𝑖
𝑈(𝑠) = (22)
𝑠
In this section, tuning the energy harvested by the
piezoelectric system is indispensable to control the quantity of
energy harvested, the stability of the system. Therefore, a
Proportional Integral Derivative controller (PID) is adapted to
the model represented above. The PID controller uses a control
loop feedback mechanism to control the piezoelectric energy
harvesting system. The parameters of the PID controller are the
proportional gain 𝑘𝑝 , the integral parameter 𝑘𝑖 and the
derivative parameter 𝑘𝑑 that affect the control of the system.
The overall effects of controller parameters 𝑘𝑝 , 𝑘𝑖 and 𝑘𝑑 on a
closed-loop system are summarized in Table 3.

Improving proportional gain 𝑘𝑝 has the effect of equivalently

increasing the control signal for the same level of error. The fact
that the controller will drive harder for a given level of error lets
Figure 6: Transfer function between the input vibration force and the the closed-loop system react more speedily, but also to
output displacement for an open loop without the PID controller
overshoot more. In addition, increasing 𝑘𝑝 helps to decrease
the steady-state error. TABLE Ⅳ: The PID controller parameters according to Ziegler-
Nichols tuning method on a closed-loop system

The addition of the integral parameter 𝑘𝑖 will continually 𝒌𝒑 𝒌𝒊 𝒌𝒅

increase over time to drive the steady-state error to attain zero. 𝐾𝑢 0.075𝐾𝑢 𝑇𝑢
0.6𝐾𝑢 1.2
If there is a steady error, the integral response will slowly 𝑇𝑢
increase the control signal to the error down. However, the
integral term can make the system more sluggish since when the The Flow Chart as shown in Figure Fig. 7 explains how to find
error signal changes sign. It can take a while for the integral the PID controller parameters based on Ziegler-Nichols
windup phenomenon when integral action saturates a controller heuristic tuning method.
without the controller driving the error signal toward zero.

The derivative controller parameter 𝑘𝑑 appends the ability of

the control signal to become large, if the error begins sloping
upward, even while the amplitude of the error is still quite small.
This anticipation helps to add damping to the system, thereby
decreasing overshoot. However, the addition of this parameter
will not affect the steady-state error.

TABLE Ⅲ: The effects of PID controller parameters on a closed-loop

system

PID Rise Oversho Settling St Steady-State

parameters Time ot time error
Decrease Increase Small
𝑘𝑝 Decrease
changes
𝑘𝑖 Decrease Increase Increase Decrease
Small Decrease Decrease
𝑘𝑑 No change
changes

There are different methods to find the PID controller

parameters that satisfy these goals. The fast rise time, the
minimal overshoot, the minimal steady-state error and the
minimal settling time. The gains of a PID controller can be found
by trial and error method. Different methods have been proposed
for setting the PID controller parameters. Some of these
methods are based on characterizing the dynamic response
of the dynamic system to be controlled with a first-order
model or second-order model with a time delay. General
methods for control design can be applied to PID control. A
number of special methods that are tailor made for PID control
have also been developed, these methods are often called
tuning methods. The most well-known tuning methods are
those that are specified by Ziegler and Nichols. The
Ziegler-Nichols Oscillation Method, Ziegler-Nichol Process
Reaction Method and Frequency Response method, is basic
Self-Tuning methods. Oscillation method is based on system
gain, in other words, system gain is redounded until the
system makes oscillation, then PID parameters can be found
from system response graphic. In our work, we apply Ziegler-
Figure 7: Flowchart to find the transfer function for the PID
Nichols heuristic tuning method to find the PID controller
controller based on Ziegler-Nichols heuristic tuning method
parameters [19] [20]. It consists to find the ultimate gain 𝐾𝑢 and
the period 𝑇𝑢 by setting the Integral and Derivative gains to zero.
The Proportional gain 𝐺 increases until it reaches the ultimate Applying the Ziegler-Nichols heuristic tuning method, we find
gain, at which the output displacement of the control loop has 𝐾𝑢 = 33.727 and 𝑇𝑢 = 3.90176. The transfer function for
consistent oscillations. For a classic PID controller, we can the PID controller can be found from eq. 2 and can be
find 𝑘𝑝 , 𝑘𝑖 and 𝑘𝑑 by the relations as in Table 4: rewritten as follows:
 addition, the terminal voltage reaches a steady value of
9.8699𝑠 2 + 20.2366𝑠 + 10.3729 0.8116 𝑝. 𝑢. at about 9,2 𝑠, which is not acceptable. We can
𝑢(𝑠) = (23) enhance the MSD system by using a controller. The using of a
𝑠
Applying Routh-Hurwitz criterion, we can verify if the proportional controller can reduce the rise time of a response,
system in a closed loop converges, stable or not [21]. but the steady-state error cannot be removed. An integral
The following structure of Routh-Hurwitz criterion matrix controller can improve the steady-state performance, but it can
for the complementary sensitivity transfer function is also badly affect the transient response. A derivative controller
shown as follows: increases the performance of the transient response by reducing
overshoot, thereby improving the stability margin for the
𝑀𝑝𝑧𝑡 (Kpzt + kp ) system. The behavior of the MSD system can be improved by
(𝐷𝑝𝑧𝑡 + 𝑘𝑑 ) 𝑘𝑖 including a PID controller as shown in Figure Fig. 8.

(((𝐷𝑝𝑧𝑡 + 𝑘𝑑 ) × (Kpzt + kp )) − (𝑘𝑖 × 𝑀𝑝𝑧𝑡 )) (24)

0 V. SIMULATION RESULTS AND PERFORMANCE
(𝐷𝑝𝑧𝑡 + 𝑘𝑑 )
( 𝑘𝑖 0 ) The values of the hardware parameters used in our simulations
are those of Mica2 Motes. All parameters used in our energy
All the values of column 1 of the table are positive. There is model are listed in Table 5. We also indicate the references
no sign changed. Therefore, the system controlled by pour where these values originated. Our results are based on existing
PID controller is stable for the closed loop. device parameters. Moreover, that is why they can reflect a real
energy model for a type of wireless sensors.
Figure Fig. 8 shows the displacement of the mass 𝑀𝑝𝑧𝑡
TABLE Ⅴ: Simulation parameters
produced by an external vibration effort 𝐹 controlled by our
PID controller. This controller is designed based on Ziegler- Parameter value
Nichols tuning method. The new system has a fast rise The supply voltage to the sensor 𝑽𝒅𝒄
2.7
time 0.0311 𝑠, it reduces also the settling time 0.0509 𝑠, and (V)
minimize the overshoot 1.1887. Reference Energy RE (J) 0.2

Threshold energy 0.1

Sensing time for one bit of data
0.5
𝑇𝑆𝑒𝑛𝑠 (ms)
Sensing current 𝐼𝑆𝑒𝑛𝑠 (mA) 25
Writing current 𝐼𝑊𝑟𝑖𝑡𝑒 (mA) 18.4
Writing time for one bit 𝑇𝑊𝑟𝑖𝑡𝑒 (ms) 12.9
Reading current 𝐼𝑅𝑒𝑎𝑑 (mA) 6.2
Reading time for one bit 𝑇𝑅𝑒𝑎𝑑 (µs) 565
On mode for the MicroController
8
Unit (MCU) current 𝐼𝐴𝑐𝑡𝑖𝑣𝑒 (mA)
On mode for the MicroController
1
Unit (MCU) time 𝑇𝐴𝑐𝑡𝑖𝑣𝑒 (ms)
Off mode for the MicroController
1
Unit (MCU) current 𝐼𝑆𝑙𝑒𝑒𝑝 (µA)
Off mode for the MicroController
299
Unit (MCU) time 𝑇𝑆𝑙𝑒𝑒𝑝 (ms)
Initial energy (J) 𝐸0 0.5 𝛼
Energy dissipation: electronics 𝐸𝑒𝑙𝑒𝑐
50
(nJ/bit)
Figure 8: Transfer function between the input vibration force and the parameters of amplifier energy
0.0013
output displacement for a closed-loop with an adaptive PID controller consumption 𝐸𝑚𝑝 (pJ/bit/m4)
and 10
and 𝐸𝑓𝑠 (pJ/bit/m2)
based on Ziegler-Nichols tuning method.
Data aggregation energy (J) 5×10-12
The response of the MSD to a unit step input without a PID Size of the data packet (bits) 𝑁 4000
controller is shown in Figure Fig. 6, where it is observed that
the MSD system without a PID controller is oscillating in Compression percentage (%) 𝛼 20
nature. It is clear that the settling time is about 6.562 𝑠 and the Mass 𝑀𝑝𝑧𝑡 (g) 182
peak overshoot is about 50.6432 𝑝. 𝑢. The rise time is The damping ratio of the structure
about 0.4 𝑠 and the settling time is too long, about 6.562 𝑠. In 0.2
𝐷𝑝𝑧𝑡 (N. m−1 . s −1)
 Parameter value
The supply voltage to the sensor 𝑽𝒅𝒄
2.7
(V)
Spring stiffness constant 𝐾𝑝𝑧𝑡
0.12320
(N. m−1 )
Proportional controller parameter 𝐾𝑝 20.2366

Integral controller parameter 𝐾𝑖 10.3729

Derivative controller parameter 𝐾𝑑 9.8699

Figure Fig. 9 shows the residual energy in the sensor’s

battery. When the battery is out of energy, the MCU switches
to off mode. During the off mode, the battery is charged with a
vibration harvesting energy without the adaptive PID
controller. The battery is recharged to 0.1245 𝐽𝑜𝑢𝑙𝑒𝑠 and
cannot reach more than this value. Therefore, there is a need to
include an adaptive PID controller to the system that can
enhance the stability and the performance of the system.
Figure 10: The self-power manager controller organizes its wake-up
mode to do its activities and its off mode to harvest energy controlled
by the adaptive PID controller depending on its residual energy

The self-power manager system including the PID

controller can recharge 65% of the battery as shown in Figure
Fig. 9 while the system without the adaptive PID controller
cannot charge more than 25% at the same period of time as
shown in Figure fig. 9. Therefore, the self-power manager
system must be tuned by an adaptive PID controller, to recharge
the battery spontaneously (according to a critical threshold
value), and stabilize the system whatever are the sensor’s
activities.

VI. CONCLUSION
An adaptive energy manager is essential to harvest energy
in WSNs. Its core goal is to control the energy harvesting to
maximize the lifetime of the sensor nodes and give the sensors
the power to achieve their activities. In this paper, autonomous
Figure 9: Self-power manager behavior without the adaptive PID
power management with a PID controller provides practical
controller
adaptation to the harvested and consumed energy for a sensor
Figure Fig. 10 shows the residual energy in the sensor’s node. Moreover, an adaptive controller for the MCU modes, to
extend a sensor lifetime depending on the existing vibration in
battery. When the residual energy is less than a given threshold
the environment, is presented. Future work will cover a
value (e.g.20%,) of the initial energy 0.5 𝐽𝑜𝑢𝑙𝑒𝑠 with 𝛼 = 1, combination between simulation results and real experimental
the MCU switches to off mode. During the off mode, the battery validation of the proposed autonomous energy management
is charged with a vibration harvesting energy controlled by the system. In addition, we will focus our work on the integration
adaptive PID controller. The battery is recharged to a fixed of some existing management energy efficient methods in our
threshold 0.3249 𝐽𝑜𝑢𝑙𝑒𝑠. self-organized power manager system to save more energy
The results illustrate that the system is still cyclical stable even during communications and to reduce the rapid consumption of
if the energy consumption depends on different sensor’s residual energy in the sensor’s battery and from the
activities and it is not the same for each time period. environmental energy resources. Applying energy efficient
protocols for sensors’ communications with the autonomous
energy management system will increase the overall WSN
lifetime.
ACKNOWLEDGMENT
The authors would like to thank the editors and reviewers
for their valuable comments on earlier versions of this paper.
 REFERENCES [11] M. Leinonen, M. Codreanu and M. Juntti, “Distributed Distortion-Rate
Optimized Compressed Sensing in Wireless Sensor Networks,” IEEE
[1] S. Dolev, S. Frenkel, M. Rosenblit, R. P. Narayanan and K. M. Transaction on Communications , vol. 66, no. 4, pp. 1609-1623, 2018.
Venkateswarlu, “In-vivo energy harvesting nano robots, ” 2016 IEEE
International Conference on the Science of Electrical Engineering [12] J. Uthayakumar, T. Vengattaraman, and P. Dhavachelvan, “A new
(ICSEE), Eilat, pp. 1-5, 2016. lossless neighborhood indexing sequence (NIS) algorithm for data
compression in wireless sensor networks,” Ad Hoc Networks, vol. 83, pp.
[2] H. Nesser, H. Débeda, J. Yuan, A. Colin, P. Poulin, I. Dufour, and C. 149-154, 2019.
Ayela, “All-organic microelectromechanical systems integrating
electrostrictive nanocomposite for mechanical energy harvesting,” Nano [13] M. Javadi, H. Mostafaei, M. U. Chowdhurry, and J. H. Abawajy,
Energy, vol. 44, pp.1-6, 2018. “Learning automaton based topology control protocol for extending
wireless sensor networks lifetime,” Journal of Network and Computer
[3] H. H. R.Sherazi, L. A. Grieco and G. Boggia, “A comprehensive review Applications, vol. 122, pp. 128-136, 2018.
on energy harvesting MAC protocols in WSNs: Challenges and
tradeoffs,” Ad Hoc Networks, vol. 71, pp. 117-134, 2018. [14] S. Kassan, J. Gaber, and P. Lorenz, “Game theory based distributed
clustering approach to maximize wireless sensors network lifetime,”
[4] F. K. Shaikh and S. Zeadally, “Energy harvesting in wireless sensor Journal of Network and Computer Applications, vol. 123, pp. 80-88,
networks: A comprehensive review, ” Renewable and Sustainable Energy 2018.
Reviews, vol. 55, pp. 1041-1054, 2016.
[15] T. Wang, G. Zhang, X. Yang, and A. Vajdi, “Genetic algorithm for
[5] K. S. Adu-Manu, A. N. Adam, A. C. Tapparello, A. H. Ayatollahi and A. energy-efficient clustering and routing in wireless sensor networks,”
W. Heinzelman, “Energy-Harvesting Wireless Sensor Networks (EH- Journal of Systems and Software, vol. 146, pp. 196-214, 2018.
WSNs): A Review, ” ACM Trans. Sen. Netw., vol. 14, no. 02 pp. 1-50,
2018. [16] W. R. Heinzelman, A. Chandrakasan and H. Balakrishnan, “Energy-
Efficient Communication Protocol for Wireless Microsensor Networks,”
[6] F. Narita and M. Fox, “A Review on Piezoelectric, Magnetostrictive, and Hawaii International Conference on System Sciences, Hawaii, 2000.
Magnetoelectric Materials and Device Technologies for Energy
Harvesting Applications, ” Adv. Eng. Mater, vol. 20, no. 05, 2017. [17] C. Williams and R. Yates, “Analysis of a micro-electric generator for
microsystems,” in Proc. Transducers, Eurosensors IX, Stockholm,
[7] K. S. Adu-Manu, N. Adam, C. Tapparello, H. Ayatollahi and W. Sweden, vol. 1, pp. 369-372, 1995.
Heinzelman, “Energy-Harvesting Wireless Sensor Networks (EH-
WSNs): A Review, ” ACM Transaction on Sensor Networks, vol. 14, no. [18] K. H. Ang, G. Chong and Y. Li, “PID Control System Analysis,
10, pp. 01-50, 2018. Design,and Technology,” IEEE Transaction On Control Systems
Technology, vol. 13, no. 4, pp. 559-576 , 2005.
[8] J. Siang, M. H. Lim, and M. S. Leong, “Review of vibration‐based energy
harvesting technology: Mechanism and architectural approach,” [19] J. G. Ziegler and N. B. Nichols, “Optimum settings for automatic
International Journal of Energy Research, vol. 42, no. 05, pp. 1866-1893, controllers,” Transaction fo the ASME, vol. 64, pp. 759-768, 1942.
2018. [20] K. J. Åström, C. C. Hang, P. Persson, and W.K. Ho, “Towards intelligent
[9] N. Kulkarni, N. R. Prasad and R. Prasad, “A Novel Sensor Node PID control,” Automatica, vol. 28, no. 01, pp. 1-9, 1992.
Deployment using Low Discrepancy Sequences for WSN,” Wireless [21] J.M. Peña, “Characterizations and stable tests for the Routh–Hurwitz
Personal Communications, vol. 100, no. 02, pp. 241-254, 2018. conditions and for total positivity,” Linear Algebra and its Applications,
vol. 393, pp. 319-332, 2004.
[10] W. Wang, D. Wang, and Y. Jiang, “Energy efficient distributed
compressed data gathering for sensor networks,” Ad Hoc Networks, vol.
58, pp.112 -117, 2017.