actuators

Article
Hysteresis Curve Fitting Optimization of Magnetic
Controlled Shape Memory Alloy Actuator
Fuquan Tu, Shengmou Hu *, Yuhang Zhuang, Jie Lv, Yunxue Wang and Zhe Sun
College of Machinery and Automation, Wuhan University of Science and Technology, Wuhan 430081, China;
tufuquan@wust.edu.cn (F.T.); zhuang2074@126.com (Y.Z.); lj881013@sina.com (J.L.); yunxue357@163.com (Y.W.);
M201570668@hust.edu.cn (Z.S.)
* Correspondence: 15527071619@163.com; Tel.: +86-27-6886-2283

Academic Editor: Jose Luis Sanchez-Rojas

Received: 2 September 2016; Accepted: 2 November 2016; Published: 8 November 2016

Abstract: As a new actuating material, magnetic controlled shape memory alloys (MSMAs) have
excellent characteristics such as a large output strain, fast response, and high energy density. These
excellent characteristics are very attractive for precision positioning systems. However, the availability
of MSMAs in practical precision positioning is poor, caused by weak repeatability under a certain
stimulus. This problem results from the error of a large magnetic hysteresis in an external magnetic
field. A suitable hysteresis modelling method can reduce the error and improve the accuracy of the
MSMA actuator. After analyzing the original hysteresis modelling methods, three kinds of hysteresis
modelling methods are proposed: least squares method, back propagation (BP) artificial neural
network, and BP artificial neural network based on genetic algorithms. Comparing the accuracy
and convergence rate of three kinds of hysteresis modelling methods, the results show that the
convergence rate of least squares method is the fastest, and the convergence accuracy of BP artificial
neural networks based on genetic algorithms is the highest.

Keywords: magnetic controlled shape memory alloy; actuator; hysteresis modelling method;
hysteresis curve; optimization

1. Introduction
During the end of last century, Ullakko et al. [1] found and proposed a magnetic controlled
shape memory alloy (MSMA). Then, Murray et al. [2,3] conducted research into the mechanism and
characteristics of MSMA. When a sample of MSMA is placed in a suitable external magnetic field, it
can produce a deformation as high as 10%. Moreover, compared with traditional shape memory alloys,
MSMA has characteristics of larger strain and better energy efficiency. These excellent characteristics
are very attractive for precision positioning systems [4]. Since then, some scholars and organizations
have designed many kinds of MSMA actuators. However, there are still relatively few applications
for MSMA actuators in precision positioning. The deficiency in practical precision positioning is
caused by the weak repeatability under a certain stimulus. This is the result of the error caused by a
large magnetic hysteresis [5,6] in an external magnetic field. To reveal the hysteresis characteristic of
MSMA [7], related experiments with NiMnGa (which is the typical MSMA material) are carried out.
According to the research and analysis of Spanish scholars, suitable hysteresis modelling methods
can reduce errors and improve the accuracy of MSMA actuators [8]. Thus, it is very important to
reference and establish hysteresis modelling methods.
At present, several kinds of commonly hysteresis models exist:

(1) Preisach model: Preisach model [9–11] can fit a complex hysteresis curve. However, for some
materials, the fitting error is large.

Actuators 2016, 5, 25; doi:10.3390/act5040025 www.mdpi.com/journal/actuators

Actuators 2016, 5, 25 2 of 13

Actuators 2016, 5, 25
(1) Preisach model: Preisach model [9–11] can fit a complex hysteresis curve. However, for 2some of 13

materials, the fitting error is large.

(2) Jiles–Atherton
(2) Jiles–Atherton model:
model: The
The advantage
advantage of of the
the Jiles–Atherton
Jiles–Atherton model
model [12,13]
[12,13] is its excellent
is its excellent
convergence, while the disadvantage is that it is difficult to achieve fast modelling
convergence, while the disadvantage is that it is difficult to achieve fast modelling due due toto
numerous
numerous undetermined
undetermined parameters.
parameters.
(3) Maxwell model: Maxwell
(3) Maxwell model: Maxwell model
model [14,15]
[14,15] can
can improve
improve the the convergence
convergence rate
rate by reducing the
by reducing the
number of subsystems, but the accuracy of the model is reduced
number of subsystems, but the accuracy of the model is reduced accordingly.accordingly.
In addition, the proportional-integral-derivative (PID) model and other models recently became
researchaddition,
In hotspotsthe proportional-integral-derivative
[16–18]. To date, none of the modelling (PID) model and
methods other models
developed recently
is perfect. became
By analyzing
research hotspots [16–18]. To date, none of the modelling methods developed
the original hysteresis modelling methods, three kinds of hysteresis modelling methods are proposed is perfect. By analyzing
the
for original
fitting the hysteresis modelling
hysteresis methods,
curve: least squaresthreemethod
kinds of[19],
hysteresis modelling methods
back propagation are proposed
(BP) artificial neural
for fitting the hysteresis curve: least squares method [19], back propagation
network and BP artificial neural network based on genetic algorithm. Among these methods, there (BP) artificial neural
network
are special and BP artificial neural
improvements. First, network
BP artificialbased on genetic
neural network algorithm.
that adoptsAmong these methods, there
the Levenberg–Marquard
are special improvements. First, BP artificial neural network that
algorithm is a novel method and has some advantages: (a) because the Levenberg–Marquard adopts the Levenberg–Marquard
algorithm
algorithm isisaanovel
type method
of least and has some
squares method,advantages:
it can it (a)
canbecause the Levenberg–Marquard
be directly algorithm
used for fitting the hysteresis
is
curve; (b) mean square error of the Levenberg–Marquard algorithm [20] can be used to test(b)
a type of least squares method, it can it can be directly used for fitting the hysteresis curve; themean
rate
square
of convergence and error size; therefore, it can be used to test the reliability and accuracy of theand
error of the Levenberg–Marquard algorithm [20] can be used to test the rate of convergence BP
error size;neural
artificial therefore, it can be
network. used to test
Moreover, thethe reliability
basic principleandofaccuracy of theneural
BP artificial BP artificial neural
network network.
based on a
Moreover, the basic[21,22]
genetic algorithm principle offollows:
is as BP artificial neural
through network
the based
selection, on a genetic
crossover algorithmoperation
and mutation [21,22] is asof
follows: through the selection, crossover and mutation operation of genetic
genetic algorithms, optimize the weights and thresholds of the neural network, and then return the algorithms, optimize the
weights
optimal and thresholds
weights and theofoptimal
the neural network,toand
thresholds thethen return the
BP artificial optimal
neural weights
network. Byand the optimal
comparing the
thresholds to the BP artificial neural network. By comparing the precision
precision of fitting the hysteresis curve, the best hysteresis modelling method will be selected. of fitting the hysteresis
curve, the best hysteresis modelling method will be selected.
2. Performance Experiment of MSMA Actuator
2. Performance Experiment of MSMA Actuator
2.1. Sample
2.1. and Device
Sample and Device of
of the
the Experiment
Experiment
The sample
The sample ofof MSMA
MSMA in in the
the experiment
experiment isis an
an NiMnGa
NiMnGa alloy
alloy purchased
purchased from
from the
the Goodfellow
Goodfellow
company (Huntingdon,
company (Huntingdon, UK). The overall
UK). The overall dimensions
dimensions ofof the
the sample are 22×
sample are × 33 ×
× 15
15 mm
mm3 ,, and
and its
its
appearance is shown in Figure 1. Moreover, the atomic percentages of the Ni element, Mn
appearance is shown in Figure 1. Moreover, the atomic percentages of the Ni element, Mn element,element,
and Ga
and Ga element
element are
are 50%,
50%, 28%,
28%, and
and 22%,
22%, respectively.
respectively.

Figure 1.
Figure Sample of
1. Sample of magnetic
magnetic controlled
controlled shape
shape memory
memory alloy
alloy (MSMA)
(MSMA) for
for the
the experiment.
experiment.

This experiment
experiment adopts
adoptsaamagnetic
magneticforce
forcecoupling
couplingthermal
thermalloading
loading measuring
measuring system
system(shown
(shown in
Figure
in 2) 2)
Figure to to
study
study multifield
multifieldcoupling
couplingproperties
propertiesofofforce
forceand
andmagnetic
magneticfieldfield of
of the
the MSMA under
variable temperature magnetics. During the experiment, the load is provided by a hydraulic system
and is measured by a force sensor.
sensor. Moreover,
Moreover, the
the load
load ranges
ranges from
from 00 toto 2.5
2.5 MPa.
MPa. The external control
magnetic field
magnetic field isis provided
providedby byananelectric
electriccurrent:
current:the
themagnetic
magneticfield
fieldrange
rangeisis0–1.5
0–1.5T,T, and
and thethe current
current is
is proportional
proportional to to the
the magnetic
magnetic field.The
field. Theoperating
operatingtemperature
temperatureisisregulated
regulatedby byaacirculating
circulating oil
oil bath
bath
temperature loading
temperature loadingsystem,
system,andandthe
the working
working temperature
temperature range
range ◦ C. When
is 16–35
is 16–35 °C. When the magnetic
the magnetic field
 properties include the following: the entire structure is sealed, stable performance, good
flexibility, and applicability to the general accuracy of the sensor. The strain limit is 1.5%, and
usage temperature range is −30 °C–+80 °C.
(2) The strain gauge sampling frequency is 4 Hz.
ActuatorsExperimental
2016, 5, 25 steps are as follows: 3 of 13

(1) Turn off the power, place the sample in the cup, and fix the cup on the hydraulic loading device
intermediate.
is perpendicular or parallel to the preload stress, axial and transverse deformation of the MSMA sample
(2)beThe
will hydraulic
produced. drive
These willloading device iswith
be measured directly placed in
a resistance the working
strain gauge. Therange. Atcharacteristics
basic the same time,of
the magnetic field is set from
the resistance strain gauge are as follows: 0 to 1.5 T.
(3) Turn on the power, and set up the pre load of the sample.
(1)(4) The
Turnstrain gauge
on the power is supply
a seriesfor ofthe
BXtemperature
(BX strain gauge
loadingrefers
system,to set
phenolic foil type strain
the temperature gauge)
parameters,
whose properties
and check whetherincludethe the following:
outer thesystem
circulation entire structure
makes good is sealed, stable
contact. performance,
When everythinggood
is
flexibility,
acceptable,and
turnapplicability
on the heating to the general
power accuracy
supply and theofoilthe sensor.
pump Thesupply,
power strain and
limitregulate
is 1.5%,the
and
usage
flow temperature
rate of siliconerange is −30 ◦ C–+80
oil circulation ◦ C. the silicone oil spilling.
to prevent
(2)(5) The
Start the test.
strain gaugeThesampling
corresponding deformation
frequency is 4 Hz. of MSMA is measured.
(6) After unloading, view and save the experimental data.

Figure
Figure 2. 2.Experimental
Experimentalsetup:
setup:(1)
(1)power;
power; (2)
(2) loading
loading device;
device; (3)
(3)power
powersupply
supplyfor
fortemperature loading
temperature loading
system; (4) computer with data acquisition card, which collects input magnetic field intensity H and
system; (4) computer with data acquisition card, which collects input magnetic field intensity H and
output magnetic induction intensity B; and (5) control device for magnetic field.
output magnetic induction intensity B; and (5) control device for magnetic field.

2.2. Experimental Results

Experimental steps are as follows:
In order to achieve good experimental results, the experimental conditions must meet the
(1)following
Turn off the power,First,
requirements. placemagnetic
the sample
field in the cup,
intensity H isand fix the
parallel cup
to the on the
external hydraulic
load. loading
The operating
device intermediate.
temperature is 16 °C, and the size of external load is 3.9 kg. In the experimental conditions, the key
(2)original experimental
The hydraulic drivedata of magnetic
loading device isfield intensity
directly placedHinand magneticrange.
the working induction intensity
At the B are
same time, the
shown in Table
magnetic 1. is set from 0 to 1.5 T.
field
(3) Turn on the power, and set up the pre load of the sample.
(4) Turn on the power supply for the temperature loading system, set the temperature parameters,
and check whether the outer circulation system makes good contact. When everything is
acceptable, turn on the heating power supply and the oil pump power supply, and regulate the
flow rate of silicone oil circulation to prevent the silicone oil spilling.
(5) Start the test. The corresponding deformation of MSMA is measured.
(6) After unloading, view and save the experimental data.

2.2. Experimental Results

In order to achieve good experimental results, the experimental conditions must meet the
following requirements. First, magnetic field intensity H is parallel to the external load. The operating
temperature is 16 ◦ C, and the size of external load is 3.9 kg. In the experimental conditions, the key
Actuators 2016, 5, 25 4 of 13

original experimental data of magnetic field intensity H and magnetic induction intensity B are shown
in Table 1.

Table 1. Key original experimental data.

Serial Serial Serial

H (T) B (T) H (T) B (T) H (T) B (T)
Number Number Number
1 −0.5258 −3.4152 11 0.0301 0.2790 21 0.0707 1.9553
2 −0.4640 −3.3831 12 0.0503 0.8144 22 0.0446 1.2415
3 −0.31118 −3.4097 13 0.0966 1.4583 23 0.0070 0.6694
4 −0.2506 −3.4061 14 0.1284 2.0656 24 −0.0133 0.0627
5 −0.1235 −2.7931 15 0.2066 3.2812 25 −0.0857 −1.2237
6 −0.0887 −2.0787 16 0.3252 4.9266 26 −0.1088 −1.6881
7 −0.0569 −1.6138 17 0.4439 6.7145 27 −0.1754 −2.7962
8 −0.0395 −1.1853 18 0.3741 6.7028 28 −0.2188 −3.3261
9 −0.0134 −0.7208 19 0.2126 4.4432
10 0.0096 −0.2920 20 0.1692 3.6708

3. Fitting and Results

3.1. Least Squares Method

As we know, MATLAB software (MATLAB 2013, MathWorks Company, Natick, MA, USA) is the
most common software numerical calculation and graphics processing. In this chapter, we apply the
least squares method with MATLAB software to fit the original experimental data.
Compared with some other least squares methods, we adopt the quadratic polynomial method
for the following reasons. First, the quadratic polynomial method is one of the most common least
squares methods. Second, the calculation time for the quadratic polynomial method is the shortest of
the least squares methods. In this paper, we assume a fitting equation as follows:

y = a4 x 4 + a3 x 3 + a2 x 2 + a1 x + a0 . (1)

The programming procedure to fit the original experimental data is as follows:

clear
a = load(”data.txt”);
x = a(:, 1);
y = a(:, 2);
P = polyfit(x, y, 4)
Application results of the programming procedure are shown as follows:
P = [−8.2964, −33.4283, 8.7761, 17.8643, 0.0325]
Therefore,
a4 = −8.2964, a3 = −33.4283, a2 = 8.7761, a1 = 17.8643, a0 = 0.0325.
Using Equation (1), the mathematical equation of the fitting curve is as follows:

y = −8.2964x4 − 33.4283x3 + 8.7761x2 + 17.8643x + 0.0325 (2)

Through a simulation analysis using MATLAB software, the fitting curve and the original
experimental data curve are shown in Figure 3.
Through the original experimental data curve in Figure 3, magnetic induction intensity B is
maintained at approximately −3.4 T when magnetic field intensity H ranges from −0.25 T to −0.54 T.
Shown as curve 1 in Figure 3, magnetic induction intensity B gradually increases from −3.4 T to 6.8 T
Actuators 2016, 5, 25 5 of 13

when magnetic field intensity H ranges from −0.25 T to 0.4 T. However, shown as curve 2 in Figure 3,
magnetic induction intensity B gradually decreases from 6.8 T to −3.4 T when magnetic field intensity
H ranges from 0.4 T to −2.5 T. Obviously, curve 1 and curve 2 do not coincide. Therefore, as shown in
Actuators
Figure 2016,
3, the5,relation
25 curve between magnetic field intensity H and magnetic induction intensity 5Bofis13
Actuators 2016, 5, 25 5 of 13
the hysteresis curve of the MSMA.

Figure 3. Fitting curve and original experimental data curve. ”Processed 1” represents the fitting
Figure 3.3.Fitting
Figure Fittingcurve
curveand
and original
original experimental
experimental datadata curve.
curve. ”Processed
”Processed 1” represents
1” represents the fitting
the fitting curve
curve through
curve through least squares
least squaresmethod.
method. ”Original
”Original data” represents
data” represents the original
the original experimental
experimental data.
through least squares method. ”Original data” represents the original experimental data. data.

With
With regard
regard to to Equation (2),
Equation (2), because
(2), because
because the the number
the number
number of of undetermined parameters
of undetermined
undetermined parameters during
during the
the
With regard to Equation parameters during the
hysteresis curve
hysteresis curve fitting
curve fitting is lower, the fitting speed of the least squares method is faster. A wealth of
hysteresis fitting isis lower,
lower, the
the fitting
fitting speed
speedof ofthe
theleast
leastsquares
squaresmethod
methodisisfaster.
faster.A A
wealth
wealthof
information
information can
can be
bebefound
found in the statistics figures. For example, with
with regard toto Figures 33 and
and4,4,the
the
of information can foundininthe
thestatistics
statisticsfigures.
figures.For
Forexample,
example, withregard
regard to Figures
Figures 3 and 4, the
fitting curve
fitting curve using
curveusing
usingthethe least
theleast squares
leastsquares method
squaresmethod
method can
can show
show the trend of the hysteresis curve as 0.33% of
fitting can show thethe trend
trend of the
of the hysteresis
hysteresis curve
curve as 0.33%
as 0.33% of
of the
the
thelargest
largest fitting
fitting error rate.
largest fitting errorerror
rate.rate.

Figure4.
Figure
Figure 4.The
4. fitting error
The fitting
fitting error rate
rate curve
curve of
of the least squares
the least squares method.
squaresmethod.
method.

Moreimportantly,
More
More importantly,the
importantly, the largest
the largest fitting
fitting error
error rate
rate of
of 0.33%
0.33% isis quite
quite accurate
accurate and
accurateand can
andcan improve
canimprove the
improvethethe
accuracyof
accuracy
accuracy ofthe
of theMSMA
the MSMAactuator.
MSMA actuator. In
actuator. In order
In order to
order to further
further improve
improve thethe fitting
fitting accuracy
fitting accuracy of
accuracyof the
ofthe fitting
thefitting curve,
fittingcurve,
curve,
thispaper
this
this paperadopts
paper adoptsaaaBP
adopts BPartificial
BP artificialneural
artificial neuralnetwork
networkto
network tocarry
to carryout
carry outthe
out the fitting.
thefitting.
fitting.

3.2.BP
3.2.
3.2. ArtificialNeural
BPArtificial
Artificial NeuralNetwork
Neural Network
Network
Thispaper
This
This paperadopts
paper adopts aaa three-layer
adopts three-layer BP
three-layer BP artificial
BP artificial neural
artificial neural network.
neural network. The
network. network
The network
network is is divided
isdivided into
dividedinto an
intoan
an
input
input layer,
layer, hidden
hidden layer,
layer, and
and output
output layer.
layer. The
The input
input and
and output
output layers
layers are,
are, respectively,
respectively,
input layer, hidden layer, and output layer. The input and output layers are, respectively, magnetic magnetic
magnetic
fieldintensity
field intensityHHand
andmagnetic
magnetic induction
induction intensity
intensity B.
B. The
The number
number of of input
input layer
layer nodes
nodes and
andoutput
output
layer nodes are both 1. Like the least squares method, we apply a BP artificial neural networkwith
layer nodes are both 1. Like the least squares method, we apply a BP artificial neural network with
MATLABsoftware
MATLAB softwareto
tofit
fitthe
the original
original experimental
experimental data.
data. We
We apply
apply equation
equation calculations
calculations combined
combined
withprogramming
with programming algorithms
algorithms to to establish
establish aa magnetic
magnetic hysteresis
hysteresis model
model that
that is
is faster
faster and
and more
more
accurate.
accurate.
Actuators 2016, 5, 25 6 of 13

field intensity H and magnetic induction intensity B. The number of input layer nodes and output
layer nodes are both 1. Like the least squares method, we apply a BP artificial neural network with
MATLAB software to fit the original experimental data. We apply equation calculations combined with
programming algorithms to establish a magnetic hysteresis model that is faster and more accurate.
Actuators 2016, 5, 25 6 of 13
Through various analyses, the number of hidden layer nodes is determined as
√
mm== √ a +
+ b ++c. . (3)
(3)

In Equation
In Equation(3),
(3),a aand
andb are,
b are, respectively,
respectively, the the number
number of input
of input layerlayer
nodesnodes and output
and output layer
layer nodes.
nodes.
In In addition,
addition, c rangesc ranges
from 1from
to 10.1 When
to 10. When
c is 3, cthe
is 3,number
the number of hidden
of hidden layerlayer
nodesnodes
is 5. is 5. Therefore,
Therefore, the
the structure of the artificial neural network is 1-5-1, as shown
structure of the artificial neural network is 1-5-1, as shown in Figure 5. in Figure 5.

Figure 5. Structure of the back propagation (BP) artificial neural network.

The activation
The activation function
function of
of neurons
neurons g(x)
g(x) in
in the
the hidden
hidden layer
layer is
is aa bipolar
bipolar compressed
compressed function:
function:
1 1
( )
g ( x ) = 2 + 1 + −λx . .
= 1 + 1 (4)
(4)
2 1+e
Correspondingly, the activation function of neurons h(x) in the output layer is a symmetric
Correspondingly, the activation function of neurons h(x) in the output layer is a symmetric
saturated linear satlins function:
saturated linear satlins function:
 −1, , < −1
ℎ( ) =  −, 1,−1 , x≤< −1≤ 1 . (5)
h (x) = x, − 1 ≤
1, ≥ 1 x ≤ 1 . (5)
 1, x ≥ 1

In this paper, the original BP artificial neural network that adopts the Levenberg–Marquard
algorithm
In thishas somethe
paper, advantages:
original BP(a)artificial
becauseneural
the Levenberg–Marquard
network that adoptsalgorithm is a type of least
the Levenberg–Marquard
squares method, it can be directly used for fitting the hysteresis curve;
algorithm has some advantages: (a) because the Levenberg–Marquard algorithm is a type and the (b) mean square
of error
least
of the Levenberg–Marquard
squares method, it can be directly algorithm
used forcan be used
fitting to test the
the hysteresis rate and
curve; of convergence
the (b) meanand error
square size;
error of
therefore, it can be used to test the reliability and accuracy of the BP artificial neural
the Levenberg–Marquard algorithm can be used to test the rate of convergence and error size; therefore,network.
it canThe equation
be used to testforthe
mean squareand
reliability error
accuracyinofthetheLevenberg–Marquard algorithm is
BP artificial neural network.
The equation for mean square error Emse in the Levenberg–Marquard algorithm is
1
= ( − ) . (6)
1 n
Emse = ∑ (origi − expi ) . 2
(6)
n i =1
In Equation (6), is original experimental data, and is expected data. If mean square
errorIn Equation (6), origi equally
is important, is original experimental
so is data, and∆exp
weight adjustment : i is expected data. If mean square error
Emse is important, equally so is weight adjustment ∆w:
∆ = ( ) ( )+ ( ) ( ), (7)
i −1
(W ) e((w)),is the error vector, and ( (7))
h
where is the unit matrix, ∆wis=theJ Tuser-defined
(W ) J (W ) +learning
µ1 I T
Jrate,
is the jacobian reciprocal matrix of the error to weights, as shown in Equation (8):
where I is the unit matrix, µ1 is the user-defined ( ) rate, e (w) is
( ) learning ( )the error vector, and J (W ) is the
jacobian reciprocal matrix of the error to weights, as shown⋯in Equation (8):
( ) ( ) ( )
⋯
( )= . (8)
⋮ ⋮ ⋱ ⋮
( ) ( ) ( )
⋯

Moreover, the equation of threshold adjustment ∆ is shown in Equation (9):

∆ = ( ) , (9)
Actuators 2016, 5, 25 7 of 13

∂e1 ( x ) ∂e1 ( x )
 
∂e1 ( x )

 ∂w1 ∂w2 ··· ∂wn


 ∂e2 ( x ) ∂e2 ( x ) ∂e2 ( x ) 

∂w1 ∂w2 ··· ∂wn

J (W ) =  . (8)
 
.. .. .. ..

 . . . . 

 ∂en ( x ) ∂en ( x ) ∂en ( x ) 

∂w1 ∂w2 ··· ∂wn


Moreover, the equation of threshold adjustment ∆θ is shown in Equation (9):

Actuators 2016, 5, 25   7 of 13
T
∆θ = µ2 f θ x x, (9)
where is the user-defined learning rate, is the input vector, and ( ) is the output of the
µ2 is the user-defined learning rate, x is the input vector, and f θ T x is the output of the neurons.
neurons.


where
Through an
Through an analysis
analysis of
of the
the activation
activation function,
function, weight
weight adjustment,
adjustment, and and threshold
threshold adjustment,
adjustment,
this paper uses
this usesMATLAB
MATLABsoftware
softwaretoto realize a hysteresis
realize curve
a hysteresis fitting
curve basedbased
fitting on a BP
on artificial neural
a BP artificial
network.
neural network.
The fitting
The fitting curve
curveand
andthe
theoriginal
originalexperimental
experimentaldata
datacurve
curveare
areshown
shownin inFigure
Figure6.6.

Figure 6.6.Fitting
Figure Fittingcurve
curveand
and original
original experimental
experimental datadata curve.
curve. “Processed
“Processed 2” represents
2” represents the fitting
the fitting curve
curve through
through the BP
the original original BP neural
artificial artificial neural“Original
network. network.data”
“Original data”the
represents represents the original
original experimental
experimental
data curve. data curve.

As shown
As shownin in Figure
Figure 7, fitting
7, the the fitting error
error rate rate between
between the fittingthe
curvefitting curve
and the and experimental
original the original
experimental data curve has large fluctuations. It was found that the largest fitting error
data curve has large fluctuations. It was found that the largest fitting error rate for the original rate for BP
the
original BP
artificial artificial
neural neural
network network
(0.07%) (0.07%)
is much is much
smaller smaller
than than the
the largest largest
fitting fitting
error rate error ratethe
through through
least
the least squares method
squares method (0.33%). (0.33%).

Figure 7.
Figure 7. Fitting-error-rate
Fitting-error-rate curve
curve for
for the
the original
original BP
BP artificial
artificialneural
neuralnetwork.
network.

To further analyse the fitting effect using an original BP artificial neural network, the mean
square error is introduced. As shown in Figure 8, the mean square error sharply decreases during 0–
10 iterations. After 10 iterations, the mean square error slows until it stops changing. Finally, the
mean square error reaches its minimum of 9.9413 × 10 at the 100th iteration. Therefore,
Actuators 2016, 5, 25 8 of 13

To further analyse the fitting effect using an original BP artificial neural network, the mean square
error is introduced. As shown in Figure 8, the mean square error sharply decreases during 0–10
iterations. After 10 iterations, the mean square error slows until it stops changing. Finally, the mean
square error reaches its minimum of 9.9413 × 10−5 at the 100th iteration. Therefore, compared with
the least squares method, the BP artificial neural network based on a genetic algorithm can decrease
the fitting error and improve the accuracy of the MSMA actuator. In addition, its convergence rate is
Actuators 2016,
relatively 5, 25
slow. 8 of 13

Figure 8. Mean square error curve.

3.3. BP
3.3. BP Artificial
Artificial Neural
Neural Network
Network Based
Based on
on Genetic
Genetic Algorithm
Algorithm
Based on
Based on an
an analysis
analysis of
of Figures
Figures 77 and
and 8,8, the
the convergence
convergence at at the
the 100th
100th iteration
iteration is
is slow,
slow, and
and 0.07%
0.07%
of the
of the largest
largest fitting-error
fitting-error rate
rate is
is not
not small.
small.
In order
In order to
to accelerate
accelerate the
the convergence
convergence rate rate of
of the
the original
original BP
BP artificial
artificial neural
neural network,
network, this
this paper
paper
uses aa genetic
uses geneticalgorithm
algorithmtotoimprove
improveBPBP artificial
artificial neural
neural network.
network. As As shown
shown in Figure
in Figure 9, the9, weight
the weight
and
and threshold
threshold valuesvalues ofBP
of the theartificial
BP artificial neural
neural network
network are optimized
are optimized through
through selection,
selection, crossover,
crossover, and
and mutation
mutation of theofgenetic
the genetic algorithm.
algorithm.
In this paper, the initial parameters of genetic algorithm are as follows: the population size is 20,
crossover probability is 0.1, and mutation probability is 0.0001.
Through selection, crossover and mutation of genetic algorithm, the optimized weight and
threshold value are as follows:

W1 = [−2.0774, −2.7653, −1.6402, −3.3241, 3.1488] T , (10)

W2 = [−0.0032, 1.8282, −1.8009, −0.6494, 0.9973], (11)

θ1 = [−2.7870, 0.1735, 1.3575, −0.3503, 0.6810] T , (12)

θ2 = [0.6449] (13)

The following require special attention: W1 in Equation (10) is the weight between the input layer
and the hidden layer, W2 in Equation (11) is the weight between the hidden layer and the output layer,
θ1 in Equation (11) is the threshold value between the input layer and the hidden layer, and θ2 in
Equation (12) is the threshold value between the hidden layer and the output layer.

Figure 9. Flowchart of the improved original BP artificial neural network.

 Based on an analysis of Figures 7 and 8, the convergence at the 100th iteration is slow, and 0.07%
of the largest fitting-error rate is not small.
In order to accelerate the convergence rate of the original BP artificial neural network, this paper
uses a genetic algorithm to improve BP artificial neural network. As shown in Figure 9, the weight
and threshold
Actuators values of the BP artificial neural network are optimized through selection, crossover,
2016, 5, 25 9 of 13
and mutation of the genetic algorithm.

Actuators 2016, 5, 25 9 of 13

Through selection, crossover and mutation of genetic algorithm, the optimized weight and
threshold value are as follows:

= −2.0774, −2.7653, −1.6402, −3.3241, 3.1488 , (10)

= −0.0032, 1.8282, −1.8009, −0.6494, 0.9973 , (11)

= −2.7870, 0.1735, 1.3575, −0.3503, 0.6810 , (12)

= 0.6449 . (13)
Figure
Figure 9.
9. Flowchart
Flowchart of
of the
the improved
improved original
original BP
BP artificial neural network.
artificial neural network.
The following require special attention: in Equation (10) is the weight between the input
In this
layerAsand paper,
the in the
hidden initial parameters of genetic algorithm are asbetween
follows: the
the hidden
population size is 20,
shown Figurelayer, in Equation
10, it is found (11)
that 0.03% of is
thethe weight
largest fitting error rate through layer and the
an improved
crossover
output probability
layer, in is 0.1,
Equation and mutation
(11) is the probability
threshold valueis 0.0001.
between the input layer and the hidden layer,
BP artificial neural network is much smaller than 0.33% of the largest fitting error rate through an
and
original in Equation
BP (12) is network.
artificial neural the threshold value between the hidden layer and the output layer.

Figure 10.
Figure 10. Fitting-error-rate
Fitting-error-ratecurve.
curve.“Original” andand
“Original” “improved” represent
“improved” the fitting-error-rate
represent curve
the fitting-error-rate
of the BP artificial neural network and BP artificial neural network based on genetic
curve of the BP artificial neural network and BP artificial neural network based on genetic algorithms,
respectively.
algorithms, respectively.

As shown in Figure 10, it is found that 0.03% of the largest fitting error rate through an improved
To further show the advantages of the BP artificial neural network based on a genetic algorithm,
BP artificial neural network is much smaller than 0.33% of the largest fitting error rate through an
as shown in Figure 11, the mean square error of the BP artificial neural network based on a
original BP artificial neural network.
genetic algorithm reaches its minimum of 4.9804 × 10−5 at the 16th iteration. Compared with
 Figure 10. Fitting-error-rate curve. “Original” and “improved” represent the fitting-error-rate curve
of 2016,
Actuators the BP
5, 25artificial neural network and BP artificial neural network based on genetic algorithms,
10 of 13
respectively.

the mean-square-error
As shown in Figure curve
10, itof
is the original
found BP artificial
that 0.03% neuralfitting
of the largest network
errorinrate
Figure 8, a an
through BPimproved
artificial
neural network based on a genetic algorithm not only accelerates the convergence
BP artificial neural network is much smaller than 0.33% of the largest fitting error rate through rate but alsoan
decreases theartificial
original BP fitting error.
neural network.

Actuators 2016, 5, 25 10 of 13

To further show the advantages of the BP artificial neural network based on a genetic algorithm,
as shown in Figure 11, the mean square error of the BP artificial neural network based on a genetic
algorithm reaches its minimum of 4.9804 × 10 at the 16th iteration. Compared with the mean-
square-error curve of the original BP artificial neural network in Figure 8, a BP artificial neural
network based
Figure
Figure 11. on a square
11.Mean
Mean genetic algorithm
squareerror
error ofnot
curveof
curve only
theBP
the accelerates
BPartificial
artificial neuralthe
neural convergence
network
network basedon
based rate but
ongenetic
genetic also decreases
algorithms.
algorithms.
the fitting error.
Therefore, it can improve the accuracy of the MSMA actuator and then enhance the utilization
Therefore, it can improve the accuracy of the MSMA actuator and then enhance the utilization
rate of the MSMA actuator in precision positioning.
rate of the MSMA actuator in precision positioning.
4. Validation
4. Validation
To validate
To validate the
the reliability
reliability and
and accuracy
accuracy ofof the
the BP
BP neural
neural network
network based
based on
on aa genetic
genetic algorithm,
algorithm,
this paper
this paper makes
makes useuse ofof additional
additional original
original experimental
experimental data
data ininexperimental
experimentalconditions.
conditions. The
The
experimental conditions are as follows: magnetic field intensity H is parallel to the
experimental conditions are as follows: magnetic field intensity H is parallel to the external load; external load;
operating temperature
operating temperature is is 16
16 ◦°C; the size
C; the size of
of the
the external
external load
load is
is 4.9
4.9 kg;
kg; and
and the
the range
range ofof the
the magnetic
magnetic
field intensity H is 0–1
field intensity H is 0–1 T. T.
To test
To test the
the fitting
fitting effect,
effect, through
through aa simulation
simulation analysis
analysis using
using MATLAB
MATLAB software,
software, the
the fitting
fitting curve
curve
and the
and the original
original experimental
experimental datadata curve
curve are
are shown
shown in in Figure
Figure 12.
12. As
As preliminarily
preliminarily seen
seen in
in Figure
Figure 12,
12,
the fitting accuracy of the BP neural network based on a genetic algorithm
the fitting accuracy of the BP neural network based on a genetic algorithm is excellent. is excellent.

Figure 12. Fitting curve and the original experimental data curve. “Processed 4” represents
Figure 12. represents the
the fitting
fitting
curve
curve through
through thethe BP
BP neural
neural network
network based
based on
on aa genetic
genetic algorithm.
algorithm. “Original
“Original data”
data” represents
represents thethe
original
original experimental
experimental data.
data.

As shown in Figure 13, the largest fitting error rate of BP artificial neural network is 0.06%, and
the largest fitting error rate of the BP artificial neural network based on a genetic algorithm is 0.025%.
Thus, a BP artificial neural network based on a genetic algorithm can significantly reduce the fitting
error.
Actuators 2016, 5, 25 11 of 13

As shown in Figure 13, the largest fitting error rate of BP artificial neural network is 0.06%, and the
largest fitting error rate of the BP artificial neural network based on a genetic algorithm is 0.025%. Thus,
Actuators
a BP 2016, neural
artificial 5, 25 11 of 13
network based on a genetic algorithm can significantly reduce the fitting error.

Actuators 2016, 5, 25 11 of 13

Figure
Figure 13.13. Fitting
Fitting error
error raterate curve
curve of theofBPthe BP artificial
artificial neural network
neural network andartificial
and the BP the BPneural
artificial neural
network
based on genetic
network based algorithms.
on genetic algorithms.

To further show the advantage of the BP artificial neural network based on a genetic algorithm,
as shown in Figure 14, the mean square error of the BP artificial neural network based on a genetic
algorithm reaches its minimum of 5.0384 × 10−5 at the 19th iteration. There is no doubt that a BP
Figure 13. Fitting error rate curve of the BP artificial neural network and the BP artificial neural
artificial neural network based on a genetic algorithm can accelerate the convergence rate and decrease
network based on genetic algorithms.
the fitting error.

Figure 14. The mean square error curve of the BP artificial neural network based on a genetic
algorithm.

To further show the advantage of the BP artificial neural network based on a genetic algorithm,
as shown in Figure 14, the mean square error of the BP artificial neural network based on a genetic
−5
algorithm reaches
Figure 14. its minimum
The mean square errorof 5.0384
curve × 10
of the BP at the
artificial 19th
neural iteration.
network basedThere is no doubt
on a genetic that a BP
algorithm.
Figure 14. The mean square error curve of the BP artificial neural network based on a genetic
artificial neural network based on a genetic algorithm can accelerate the convergence rate and
algorithm.
5.decrease the fitting error.
Conclusions
To further
Through theshow
above theexperiments
advantage of thesimulation
and BP artificial neural the
analysis, network based on
conclusions areaasgenetic algorithm,
follows:
5. Conclusions
as shown in Figure 14, the mean square error of the BP artificial neural network based on a genetic
(1) The
algorithm fitting
Through
reachesaccuracy
the above
its rate is quite
experiments
minimum high
andand
of 5.0384 can− 5 improve
simulation
× 10 the accuracy
analysis,
at the 19th ofThere
MSMA
the conclusions
iteration. isareactuators.
noasdoubt
follows:Because
that a BP
there
artificial are
neuralfewer undetermined
network based onparameters
a genetic during
algorithm the hysteresis
can curve
accelerate fitting,
the the fitting
convergence speed
rate of
and
(1) The fitting accuracy rate is quite high and can improve the accuracy of MSMA actuators.
leastthe
decrease squares
fittingmethod is fast.
error. are
Because there fewer undetermined parameters during the hysteresis curve fitting, the
fitting speed of least squares method is fast.
5. Conclusions
(2) Compared with BP artificial neural networks, BP artificial neural networks based on genetic
algorithms
Through the above can accelerate and
experiments the convergence rate and
simulation analysis, thedecrease the fitting
conclusions error. Therefore,
are as follows:
they can improve the accuracy of MSMA actuators and enhance the utilization rate of
(1) The fitting accuracy rate is quite high and can improve the accuracy of MSMA actuators.
Actuators 2016, 5, 25 12 of 13

(2) Compared with BP artificial neural networks, BP artificial neural networks based on genetic
algorithms can accelerate the convergence rate and decrease the fitting error. Therefore, they can
improve the accuracy of MSMA actuators and enhance the utilization rate of MSMA actuators in
precision positioning.

Acknowledgments: This study was funded by the China National Natural Science Funding Project (Grant No:
51274155) and Hubei Province Natural Science Funding Project (Grant No: 2015CFA114).
Author Contributions: F.T. and S.H. conceived and designed the experiments; Y.Z. and Y.W. performed the
experiments; J.L. analyzed the data; Z.S. contributed analysis tools; S.H. wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Ullakko, K.; Huang, J.K.; Kantner, C.; O’Handley, R.C.; Kokorin, V.V. Large magnetic-field-induced strains in
Ni2 MnGa single crystals. Appl. Phys. Lett. 1996, 69, 1966–1968. [CrossRef]
2. Murray, S.J.; Marioni, M.; Allen, S.M.; O’Handley, R.C.; Lograsso, T.A. 6% magnetic-field-induced strain by
twin-boundary motion in ferromagnetic Ni–Mn–Ga. Appl. Phys. Lett. 2000, 77, 886–888. [CrossRef]
3. Murray, S.J.; Farinelli, M.; Kantner, C.; Huang, J.K.; Allen, S.M.; O’Handley, R.C. Field-induced strain under
load in Ni-Mn-Ga magnetic shape memory materials. J. Appl. Phys. 1998, 83, 7297–7299. [CrossRef]
4. Feuchtwanger, J.; Asua, E.; García-Arribas, A.; Etxebarria, V.; Barandiaran, J.M. Ferromagnetic shape memory
alloys for positioning with nanometric resolution. Appl. Phys. Lett. 2009, 95, 054102. [CrossRef]
5. Murray, S.J.; Marioni, M.; Tello, P.G.; O’Handley, R.C. Giant magnetic-field-induced strain in Ni–Mn–Ga
crystals: Experimental results and modelling. J. Magn. Magn. Mater. 2001, 226, 945–947. [CrossRef]
6. Conti, S.; Lenz, M.; Rumpf, M. Hysteresis in magnetic shape memory composites: Modelling and simulation.
J. Mech. Phys. Solids 2016, 89, 272–286. [CrossRef]
7. Krenke, T.; Aksoy, S.; Duman, E.; Acet, M.; Moya, X.; Mañosa, L.; Planes, A. Hysteresis effects in the
magnetic-field-induced reverse martensitic transition in magnetic shape-memory alloys. J. Appl. Phys. 2010,
108, 043914. [CrossRef]
8. Sadeghzadeh, A.; Asua, E.; Feuchtwanger, J.; Etxebarria, V.; García-Arribas, A. Ferromagnetic shape memory
alloy actuator enabled for nanometric position control using hysteresis compensation. Sens. Actuators A Phys.
2012, 182, 122–129. [CrossRef]
9. Tao, G. Adaptive control of systems with nonsmooth input and output nonlinearities. IEEE Trans.
Autom. Control 1996, 41, 1348–1352.
10. Ossart, F.; Davidson, R.; Charap, S.H. A 3D moving vector Preisach hysteresis model. IEEE Trans. Magn.
1995, 31, 1785–1788. [CrossRef]
11. Sutor, A.; Rupitsch, S.J.; Lerch, R. A Preisach-based hysteresis model for magnetic and ferroelectric hysteresis.
Appl. Phys. A 2010, 100, 425–430. [CrossRef]
12. Takahashi, N.; Miyabara, S.H.; Fujiwara, K. Problems in practical finite element analysis using Preisach
hysteresis model. IEEE Trans. Magn. 1999, 35, 1243–1246. [CrossRef]
13. Annakkage, U.D.; McLaren, P.G.; Dirks, E.; Jayasinghe, R.P.; Parker, A.D. A current transformer model based
on the Jiles–Atherton theory of ferromagnetic hysteresis. IEEE Trans. Power Deliv. 2000, 15, 57–61. [CrossRef]
14. Toman, M.; Stumberger, G.; Dolinar, D. Parameter identification of the Jiles–Atherton hysteresis model using
differential evolution. IEEE Trans. Magn. 2008, 44, 1098–1101. [CrossRef]
15. Rao, I.J.; Rajagopal, K.R. On a new interpretation of the classical Maxwell model. Mech. Res. Commun. 2007,
34, 509–514. [CrossRef]
16. Zhou, M.; Zhang, Q. Hysteresis Model of Magnetically Controlled Shape Memory Alloy Based on a PID
Neural Network. IEEE Trans. Magn. 2015, 51, 1–4.
17. Zhou, M.; He, S.; Zhang, Q.; Ji, K.; Hu, B. Hybrid control of magnetically controlled shape memory alloy
actuator based on Krasnosel’skii–Pokrovskii model. J. Intell. Fuzzy Syst. 2015, 29, 63–73. [CrossRef]
18. Zhou, M.; He, S.; Hu, B.; Zhang, Q. Modified KP model for hysteresis of magnetic shape memory alloy
actuator. IETE Tech. Rev. 2015, 32, 29–36. [CrossRef]
19. Liu, Y.; Liu, H.; Wu, H.; Zou, D. Modelling and compensation of hysteresis in piezoelectric actuators based
on Maxwell approach. Electron. Lett. 2015, 52, 188–190. [CrossRef]
Actuators 2016, 5, 25 13 of 13

20. Shi, Y.; Liang, C.H. The finite-volume time-domain algorithm using least squares method in solving
Maxwell’s equations. J. Comput. Phys. 2007, 226, 1444–1457. [CrossRef]
21. Wilamowski, B.M.; Iplikci, S.; Kaynak, O.; Efe, M.Ö. An algorithm for fast convergence in training neural
networks. In Proceedings of the International Joint Conference on Neural Networks, Washington, DC, USA,
15–19 July 2001; Volume 3, pp. 1778–1782.
22. Fu, Z.; Mo, J.; Chen, L.; Chen, W. Using genetic algorithm-back propagation neural network prediction and
finite-element model simulation to optimize the process of multiple-step incremental air-bending forming of
sheet metal. Mater. Des. 2010, 31, 267–277. [CrossRef]

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).