Journal of Energy Storage 20 (2018) 1–15

Contents lists available at ScienceDirect

Journal of Energy Storage

journal homepage: www.elsevier.com/locate/est

A series hybrid “real inertia” energy storage system T

a,⁎ a a b
J.P. Rouse , S.D. Garvey , B. Cárdenas , T.R. Davenne
a
Department of Mechanical, Materials and Manufacturing Engineering, University of Nottingham, Nottingham, Nottinghamshire NG7 2RD, United Kingdom
b
Rutherford Appleton Laboratory, Didcot OX11 0QX, United Kingdom

A R T I C LE I N FO A B S T R A C T

Keywords: The wide scale market penetration of numerous renewable energy technologies is dependent, at least in part, on
Real inertia developing reliable energy storage methods that can alleviate concerns over potentially interrupted and un-
Hybrid energy storage certain supplies. Many challenges need to be overcome, not least among them is allowing capacity for the wide
Kinetic energy storage range of time scales required to ensure grid stability. In thermal power plant, high frequency/short duration
Frequency response
demand fluctuations, acting at the milliseconds to several seconds time scale, are addressed passively by the
inertia of the grid. Here, grid inertia can be thought of as the mechanical inertia of spinning steel in steam and
gas turbines. This allows time for active control measures to take effect at the tens of second to hours time scale
and for the system to recover without a supply frequency deviation that is noticeable to the customer. It is of
paramount importance that, as thermal plant is retired, renewable energy generation and storage systems ac-
count for the loss of this inertia. In the literature, strategies to address the loss of “real” inertia have often relied
on emulation rather than actual replacement. The present work focuses on the preliminary development of a
novel energy storage system that makes use of real inertia to address short term supply/demand imbalances
while simultaneously allowing for extended depths of discharge. The concept looks to combine flywheel and
compressed fluid energy stores in order to power a synchronous generator. By combining these energy storage
technologies through a differential drive unit, DDU, it is anticipated that the benefits of high system inertia can
be exploited in the short term while allowing energy to be continually extracted from the flywheel in the long
term during storage discharge. The use of a DDU makes the present design particularly novel and distinct from
other hybrid systems. In essence, this inclusion allows energy to be extracted entirely from the flywheel, in-
ducing “real” inertia, or entirely from the secondary store, inducing “synthetic” inertia, or some combination of
the two. Fundamental sizing calculations for a 50 MW system with 20 MWh of storage capacity are presented and
used to design a suitable control system that allows for the operation of both primary flywheel and secondary
compressed fluid energy stores. The transient behaviour of the system is simulated for several charge/discharge
time profiles to demonstrate response stability for the system. Comments on system turnaround efficiency, which
is dependent upon loading history but for the intended applications can be considered to be greater than 90% are
also made here, along with a case study application to an isolated Californian solar powered grid.

1. Introduction known as RoCoF, when a sudden change in load is encountered [3].

Systems such as thermal energy storage and pumped hydroelectric have
The intermittent and irregular nature of renewable energy sources very little associated inertia and may be thought of as providing slow
necessitates at least some form of energy storage if uninterrupted response energy storage. Slow energy storage in the present context
supply is to be achieved [1]. Mismatches in supply and demand need to may be thought of as reactions to grid imbalances that take place over
be accounted for on a wide range of time scales, from the order of weeks time periods greater than several minutes. Conversely, fast energy
or months as a result of diurnal and seasonal variations [2], to seconds storage addresses momentary load imbalances on the millisecond to
and milliseconds. In order to ensure a stable grid, it is critically im- second time scale.
portant that a balance is maintained between consumption and gen- In thermal power plants, the inertia of a turbine passively controls
eration in real time over this wide range of time scales [3]. The inertia the rate of the change in speed to the synchronous machine it is coupled
response of an energy system limits the rate of change of frequency, to. This action buys time for active control systems to take effect and

⁎
Corresponding author.
E-mail address: James.Rouse@nottingham.ac.uk (J.P. Rouse).

https://doi.org/10.1016/j.est.2018.08.006
Received 21 December 2017; Received in revised form 3 August 2018; Accepted 6 August 2018
2352-152X/ © 2018 Elsevier Ltd. All rights reserved.
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Nomenclature KI integral gain (kg m2/s2)

ωn, ζ, k DDU transfer function parameters
ω rotational speed (rad/s) T instantaneous DDU torque (Nm)
ωFW flywheel rotational speed (rad/s) TR DDU rated torque (Nm)
ω̂FW flywheel burst rotational speed (rad/s) TC controller torque (Nm)
ωD flywheel design rotational speed (rad/s) TW windage torque (Nm)
ωSM synchronous machine rotational speed (rad/s) θEM electromechanical load angle (°)
ωSM synchronous machine nominal rotational speed (rad/s) θ̂EM electromechanical load angle at rated synchronous ma-
ω̂SM synchronous machine target rotational speed (rad/s) chine power (10°)
ωSMLL synchronous machine rotational speed lower limit (rad/s) KSM synchronous machine spring constant (Nm/°)
ωSMUL synchronous machine rotational speed upper limit (rad/s) err rotational speed error, difference between instantaneous
EK kinetic energy (J) synchronous machine speed and target speed (rad/s)
ΔEFW change in flywheel store energy (J) t simulation time (s)
ΔEHY change in secondary store energy (J) ts total simulation time (s)
ETrans transaction energy (J) PPV photovoltaic cell power (MW)
J moment of inertia (kg m2) PDemand total demand power (MW)
JFW flywheel moment of inertia (kg m2) PGen total generated power (MW)
JSM synchronous machine moment of inertia (kg m2) PBase total baseline power (MW)
m flywheel mass (kg) P̂SM rated synchronous machine power (MW)
RI internal flywheel radius, bore radius (m) CW windage loss coefficient
RO external flywheel radius (m) ηG generator efficiency
REP elastic/plastic transition radius (m) ηM motor efficiency
r radial coordinate (m) ηP pump efficiency
ρ flywheel material density (kg/m3) ηPV photovoltaic cell efficiency
σr radial stress component in flywheel (MPa) ηINV inverter efficiency
σθ hoop stress component in flywheel (MPa) ηT total system efficiency
A, B, C integration constants in flywheel stress expressions hpv, ppv, qpv, mpv, rpv, spv, upv photovoltaic cell model coefficients
σY flywheel material yield stress (MPa) θPV photovoltaic cell temperature (°C)
σUTS flywheel material ultimate tensile strength (MPa) θAir instantaneous air temperature (°C)
Δσ “Endurance” alternating stress, fully reversed condition θ̂Air peak air temperature (°C)
(MPa) trise time of sunrise
σe flywheel material endurance limit (MPa) tset time of sunset
SA alternating stress for a particular loading condition (MPa) t θ̂Air time of peak air temperature
SM mean stress for a particular loading condition (MPa) MAir normalised air mass
Nf number of cycles to failure Gβ global solar irradiation (W/m2)
α, β Basquin model parameters ψZ zenith angle (°)
pi hydraulic fluid working pressure (bar) dn day number
ΔP hydraulic machine pressure differential (bar) Cs, Cv , Cf hydraulic machine model coefficients
PAtmos atmospheric pressure (bar) QP/M pump/motor flow rate (m3/s)
p number of machine poles TP/M pump/motor torque (Nm)
f frequency (Hz) D hydraulic machine displacement (cc/rev)
G synchronous machine basic rating (MVA) x hydraulic machine displacement fraction
H synchronous machine inertia constant (MJ/MVA) AL loading cycle amplitude (MW)
GC PID controller transfer function tL loading cycle period (s)
KP proportional gain (kg m2/s) RoCoF rate of change of (grid) frequency (Hz/s)
KD differential gain (kg m2)

stabilise the system frequency by adjusting prime mover inputs. Note rotating at rated speed, to the rated electrical power of the system.
that prime mover adjustment may not be required for a particular load Inertia constants have time units and indicate how long it would take
imbalance scenario if, for example, sufficient kinetic energy can be for a rotating mass to de-accelerate to stationary if continuously dis-
extracted from synchronous machine rotors. Renewable energy sources charged at rated power [6].
on the other hand are commonly connected to the grid via power The present work proposes an electricity in/electricity out (EIEO)
converters rather than synchronous generators. As such they operate to storage system that bridges the gap between the extremes of energy
generate maximum available power and do not respond to system load storage time scales, with sudden load imbalances addressed through the
directly. Power converters require control technology in order to keep introduction of “real system inertia” (in a flywheel) and secondary
line frequencies, voltages and power oscillations within acceptable energy stores (compressed fluid) exploited for sustained delivery over
tolerances while also guarding against power circulation [4]. Renew- longer time periods. Real inertia is distinct to emulated or synthetic
able energy sources typically have little in the way of intrinsic inertia in inertia, and may be thought of as energy storage that acts in an entirely
the first instance. Wind turbines (for example), have relatively low in- passive manner. That is to say, the transfer of energy is determined
ertia constants, 2–6 s [5], and it is debatable as to whether or not this completely by the reluctance of the system to change speed. By way of
inertia is truly seen by the grid due to the interconnecting power example, a 50 MW system with a capacity of 20 MWh is sized here with
electronics. For comparison, turbo generators used in conventional the view that such a design could form part of the energy storage re-
steam power plants have inertia constants of 7–9 s [6]. Inertia constants quirement for an offshore wind farm.
may be expressed as the ratio of stored kinetic energy in a system, Numerous techniques have been proposed for emulating inertia in

2
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

renewable generators [7,8]. For example, offshore wind farms (parti- better for demand load variations. The benefits of matching demand
cularly in remote locations) are often connected to the grid through variations are still evident in the synchronous system proposed here,
high voltage DC (HVDC) links, thereby decoupling offshore frequency however potentially significant losses associated with the presence of
fluctuations from grid fluctuations. Several articles have been published two hydraulic pump/motor assemblies are avoided and a stiffer (solid)
that detail methods for inertia emulation in such systems using, say, link is provided between the main source of inertia (the flywheel) and
capacitance within the HVDC [9,10]. The system presented here uses the synchronous machine.
only a minimum amount of voltage regulation electronics, similar to Flywheel energy stores have historically been widely used for un-
that which is deployed on thermal power plant synchronous machines. interruptible power supply (UPS) applications [15] and have had a
Several hybrid energy storage systems that employ a flywheel have great deal of success in this area with many manufacturers in the
been proposed in the literature with a renewable energy (particularly market [16]. 97% of AC outages are resolved in under 3 s [16], sug-
wind) emphasis. The work of Prodromidis and Coutelieris [11,12], for gesting that high cycle lives are highly valuable and limiting the ef-
example, has detailed experimental investigations into flywheel energy fectiveness of electrochemical energy storage solutions. Diesel gen-
store design/efficiency and the application of hybrid battery/flywheel erator/flywheel UPS systems may be considered “series” energy stores
standalone systems to small grids. In the former study steel and alu- (see Section 2) in a similar sense to the hybrid system developed in the
minium flywheel designs, which utilised different cross section geo- present work. It should be noted however that, in the system developed
metries, were compared at the laboratory scale (0.7 kW and 0.2 Wh) for here, it is possible to fully decouple to flywheel from the synchronous
building energy storage applications. Aluminium designs were favoured machine and supply power solely from some secondary source.
due to the possibility for greater levels of energy storage in the given
application, however interactions with a grid were not considered and 2. Series hybrid kinetic energy storage (SHyKESS)
their system only has a limited relevance to the one proposed in the
present work. Combined battery and flywheel hybrid energy stores The presented hybrid energy storage system, refereed to as
were also analysed by Prodromidis and Coutelieris for a standalone SHyKESS, falls into a category of systems that the authors would term
system on the Greek island of Naxos. Specifically, flywheels were used “series” type systems. These are distinct to “parallel” type systems,
to supplement the energy storage of Hoppecke, Surrette, and Vision which make up the vast majority of designs proposed in the literature.
batteries, all with approximately 3000 Ah of storage. Such capacity In parallel systems, energy storage technologies are individual elements
levels were chosen so that the considered systems could feasibly supply feeding a common bus linked to the grid. This type of hybrid system is
a typical house. Net present costs of the flywheel and battery systems considered the norm. The overview of energy store hybrids in papers by
were comparable to simple battery systems (approximately $6000). Amrouche et al. [17] and Bocklisch [18] highlight this fact and little
Again however, the systems considered in Prodromidis and Coutelieris’ exploration into alternative areas can be found in the literature. The
work did not interact directly with any kind of grid and were sized at a complexity of parallel hybrid designs is only increased when it is re-
vastly different scales to the system proposed in the present work. membered that each generator must be controlled by some overarching,
Sebastián and Peña-Alzola [13] focused on the development of a ideally autonomous, system that is sensitive to the various supply and
control system and simulation method for a flywheel energy store demand scenarios that a grid may experience. See, for example, the
which was to be used alongside a hybrid wind/diesel power systems. work of Shankar and Mukherjee [19] and Hamzaoui et al. [20]. The
The flywheel component was sized using elastic stress analysis and a proposed system may be termed a “series” hybrid energy storage
Tresca failure criterion (a similar study will be carried out later in this system. Electrical power to the grid is supplied by a single generator
paper). Machine and grid side control systems were established in order (per base unit) which categorically generates in a synchronous manner,
to maintain link voltages by varying the low speed flywheel power, thereby removing the need for frequency control power electronics. The
thereby resulting in smoother operation during power transients. The grid “sees” precisely the same type of generator that one would expect
system considered in this work was however limited to isolated micro- from a conventional fossil fuel power plant. Furthermore, the proposed
grid applications, with power/capacity ratings of approximately hybrid allows energy to be stored and extracted from multiple stores
150 kW and 5 × 10−3 MWh, respectively. simultaneously at rates that are appropriate for the specific grid con-
Of particular interest here are publications by Carrillo, Feijóo and dition. In the case of a system demand for example, short term gen-
Cidrás [14], where synchronous and asynchronous machines were at- eration can be provided by running down a flywheel. If prolonged en-
tached to diesel generators and flywheel systems in order to compare ergy recovery is required the generator still runs at a synchronous speed
their performance in supplementing wind power in isolated locations by extracting energy from both the flywheel and a compressed fluid
(i.e. for low power applications of approximately 50 kW). Building on store. The ratio of power from the flywheel to the power from the
the author's previous work, one synchronous machine/flywheel con- compressed fluid would clearly decreases as the flywheel spins down.
figuration featured a hydraulic transmission linking the two compo- SHyKESS comprises of several simple components and, during charge/
nents. This was done in order to “allow energy transfer between two discharge cycles, can be thought to operate in one of several modes. For
systems rotating at different speeds”. During discharge, the flywheel clarity, an overview of the SHyKESS system is presented here with a
would be spun down and used to drive a fixed displacement pump that step by step discussion of a typical operation pattern.
circulates a pressurised fluid in line connected to a variable displace- An overview of SHyKESS is given in Fig. 1. A synchronous machine
ment motor. Work can thus be extracted by a motor to power a syn- (item 4) is connected to a flywheel (item 1) through a differential drive
chronous machine. Broadly speaking, variable speed configurations unit (DDU, item 3). In the present work, the DDU may most readily be
(asynchronous machines) were concluded to be superior for accom- interpreted as a hydraulic pump/motor, the rotor of which is connected
modating wind speed fluctuations and synchronous machines were directly to the synchronous machine and the “stator” of which is

Fig. 1. The proposed flywheel based energy storage concept

Series Hybrid Kinetic Energy Storage System (SHyKESS). Note
that, in the present work, the DDU may be through of as a
hydraulic motor/pump assembly.

3
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

connected to the flywheel shaft. Clearly, if no fluid is allowed to flow The charging condition is represented in Fig. 3(b). At this point the
through such a machine, the pump/motor assembly will act as a rigid DDU will lock and the system will move back into mode 1. The shaft
coupling. If fluid is however allowed to flow through the machine the speed (identical to the flywheel and synchronous rotor speed in this
synchronous machine shaft will spin relative to the flywheel shaft, case) will now increase until it reaches the upper tolerance limit for the
thereby offering a path for the introduction of “slip” energy into the machine (ωSMUL). At this point the DDU will again unlock, with the
system. Both the synchronous machine and flywheel are mounted on secondary store drive acting as a motor (spinning in the opposite direct
hollow shafts. These are connected only through the DDU and allow for to the charging instance) to increase the flywheel speed while keeping
the transmission of high pressure compressed fluid (the secondary en- the synchronous machine speed within tolerance (mode 2B). By doing
ergy store). It is important to note that this feature of SHyKESS facil- this we may increase the amount of energy stored in the flywheel,
itates one of the system's key advantages over other proposals. From a however a structural limit will exist which prohibits any more being
mechanical perspective (i.e. in terms of torque transmission), none of stored in the system (mode 3B). At this point a secondary (separate)
the components are grounded to earth. Torque transmission is accom- store must be employed. The above example is demonstrated using a
plished through the DDU, the casing of which is free to rotate with, say, transient response model in Section 4 (Fig. 7).
the synchronous machine rotor. The DDU allows slip between the fly-
wheel and synchronous machine rotor speeds and houses a variable
3. SHyKESS basic design calculations and development of a
displacement compressed fluid machine. This machine may operate as a
control system model
motor (during system discharge, thereby extracting energy from the
compressed fluid store) or as a pump (allowing for charging of the
To make a true impact on the energy market high power designs are
secondary energy store). A sprag clutch is incorporated in the coupling
required. The present work will therefore focus on the development of a
between the flywheel and DDU in order to prevent the flywheel's di-
50 MW 20 MW hr capacity system, meaning that contributions com-
rection of rotation being reversed during full discharge (i.e. when the
parable to hydro-storage can be made realised [22] and the system
speed of the flywheel relative to ground is 0). There is a subtlety in the
would be well suited for offshore wind applications. What follows are
design of the synchronous machine that results from the use of a hollow
fundamental sizing calculations for the 50 MW system and the devel-
rotor shaft. A 4 pole design is favoured over a 2 pole variant as it is
opment of a transient response model.
expected that this will minimise disturbances to the magnetic field
The kinetic energy (EK) stored in a flywheel may be given by Eq. (1),
which, in the 2 pole machine must pass through changing media,
where J is the moment of inertia (see Eq. (2)). Note m is flywheel mass,
namely the steel rotor shaft and the internal compressed fluid.
RI/RO are the internal and external radii, respectively, and ω is the
The multiple operating modes of SHyKESS are explained in the
flywheel's rotational velocity. A 4 pole synchronous machine is to be
flowchart given in Fig. 2. Fig. 3 shows a schematic representation of
implemented (for the reasons given in the previous section), giving rise
how contributions to total power are made from the flywheel and
to a synchronous rotational velocity of 50 πrad/s (ωSM = (4πf)/p,
secondary energy stores as the flywheel speed, ωFW, varies from the
where p is the number of poles in the machine and f is the operating
nominal synchronous machine speed, ωSM , to 0 (relative to earth). Both
frequency, here taken to be 50 Hz). Using the requirement for a
charging and discharging operating conditions are represented. Con-
20 MW hr storage capacity and Eq. (1), it can be found that the SHy-
sider SHyKESS in mode 1, where both the flywheel and synchronous
KESS flywheel requires a moment of inertia of
machine rotor are rotating at the same speed. Note that, ωFW = ωSM,
JFW = 5.836 × 106 kg m2.
where ωFW and ωSM are the flywheel and synchronous machine rota-
tional speeds, respectively. For the purposes of this example, we shall 1
EK = JFW ω2
assume that this is the synchronous speed (50 πrad/s for a 4 pole ma- 2 (1)
chine) and initially there is no load on the system. During discharge,
when the synchronous machine acts as a generator, the DDU is initially 1
JFW = m (RO2 + RI2)
locked and the system has the sum of flywheel and rotor inertias. As the 2 (2)
system continues to discharge ωSM will approach a lower limit, ωSMLL, The inertia of a synchronous machine rotor may be approximated in
which is the minimum speed the synchronous machine can run such the following way (taken from Kothari and Nagrath [6]). Eq. (3) relates
that generated frequency is within the grid tolerance. Note that in the G, the machine base rating (MVA) and a machine based inertia constant
UK, the 2002 electricity safety, quality and continuity regulations H (MJ/MVA, note therefore that GH is equivalent to the rotor's kinetic
(ESQCR 2002) stipulate that frequency must be with ± 1% of the energy) to Me (the moment of inertia of the rotor based on the electrical
nominal operating frequency (in this case 50 Hz) [21]. When this limit field rotational velocity ωs, MJ s/rad(electrical)). The relationship be-
is reached, the DDU will unlock, allowing an amount of slip between tween ωs (rad(electrical)/s) and ωSM (rad(mechanical)/s) is given in Eq.
ωFW and ωSM (during discharge, this is mode 2A). In this example, en- (4) for a machine with p poles. Using Eqs. (4) and (4), the definition of
ergy is extracted from the secondary (compressed fluid) store in order kinetic energy in Eq. (1) and assuming H = 6 MJ/MVA (which is re-
to maintain a constant torque over the DDU, acting to slow down the presentative of a condensing turbine steam generator),
flywheel (thereby extracting stored kinetic energy) and return the rotor JSM = 24.317 × 103kg m2.
speed to the target (synchronous) speed (ω̂SM) ). This condition is re-
presented in Fig. 3(a). Note that, as the flywheel speed approaches 0,
the secondary energy store accounts for a greater contribution to the
total power. Control of a variable displacement motor in the DDU can
therefore be achieved by comparing ωSM to ω̂SM) (thereby modifying an
output torque). Power can continue to be supplied by the flywheel until
its rotational speed (relative to ground) becomes 0 and the sprag clutch
disengages (mode 3A).
In the case of charging (starting in mode 3A), the DDU acts as a
pump, thereby providing a negative torque to the synchronous machine
rotor in an attempt to slow it down. As such the sprag clutch will re-
engage and the speed of the flywheel will increase (mode 2A). With
continued charging of the system the speed of the flywheel will con-
tinue to increase and approach the speed of the synchronous machine. Fig. 2. SHyKESS operating mode flowchart.

4
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Fig. 3. Schematic representations of power contributions for the SHyKESS system under (a) discharging and (b) charging operating conditions.

GH =
1
Me ωs differential and the required machine power, 50 MW, the rated flow
2 (3) rate for the machine can be calculated, namely 1.0 × 105 cc/rev at
synchronous speed. Assuming a uniform velocity profile and imposing a
p
ωs = ⎛ ⎞ ωSM laminar flow condition consequently indicates that RI = 0.4 m.
⎝2⎠ (4) In the present work, flywheel sizing is based on elastic-perfectly-
Assuming a constant thickness flywheel design, the optimum solu- plastic (EPP) material assumptions with a Tresca yield criterion. Note
tion in terms of packaging and resource usage of the flywheel mass is that this solution is well established (see Refs. [24,25]) and only a brief
dependent on setting a burst speed (ω̂FW ). This, in turn, is dependent on overview is given here. Attentional is focused on solutions for non-zero
accurately estimating the stress distribution in the flywheel. Elastic σr values at the internal bore. Eq. (7) describes equilibrium in a disc
solutions for the radial and hoop stresses (σr and σθ, respectively) in a rotating at speed ω, where ρ is the material density and r is the radial
rotating disc are widely known (see Eqs. (5) and (6), where B and C are coordinate, assuming plane stress conditions. It is expected that casting
constants of integration) but would only allow for very conservative a flywheel of the dimensions discussed here as a single entity will afford
designs (i.e. those that deploy mass poorly) to be realised. This is par- many non-trivial challenges for both manufacture and transport. As
ticularly true for the present case, where the presence of an internal such, it is expected that the present work's flywheel will be made up of
bore (for the compressed fluid supply) results in a stress riser at the several laminates or plates which are constrained axially. As such, the
internal radius. Given the hardening characteristics of many en- plane stress conditions assumed here are taken to be valid. The limiting
gineering steels, the complete avoidance of yielding is unreasonable for case (burst) is considered to be when the material has yielded through
a large number of engineering applications and it is the author's as- the entire radius of the flywheel. The Tresca criterion and EPP material
sertion that SHyKESS is one of them. assumption therefore indicate that this occurs when σθ = σY for all r.
Substituting this in Eq. (7) and integrating gives Eq. (8), where the
C ρω2 (3 + ν ) 2 constant A is determined using the boundary condition at r = RI,
σr = B − − r
r 2 8 (5) namely σr =− pi, the internal shaft pressure (see Eq. (9)). With this in
hand, the external flywheel radius RO may be found at a design speed,
C ρω2 (1 + 3ν ) 2 ωD, which gives rise to a completely plastic flywheel by enforcing the
σθ = B + − r
r2 8 (6) external boundary condition (σr = 0 at r = RO). For SHyKESS using
Before sizing the hollow flywheel an estimation of the internal bore EN24 (BS 970 817M40) steel (σY = 940 MPa, ρ = 8170 kg/m3 [26]), it
radius (RI) is required. This is constrained by the flow of fluid in the may be found that RO = 3.18 m. Using Eq. (2), the length of the fly-
secondary store. ISO VG150 oil is taken as the secondary store working wheel (L) may be found to be 4.45 m. As the Tresca criterion is already
fluid (νoil ≈ 7.0 × 10−4 m2/s and ρoil ≈ 870 kg/m3 [23]) and a pressure conservative it is assumed here that ωD need only marginally greater
differential over the DDU of 20 MPa is assumed. Using this pressure than the maximum flywheel operational speed (ω̂FW ). For the purposes

Fig. 4. Normalised flywheel stress distributions for (a) the design speed case (ωD) and (b) the maximum operating speed case (ω̂FW ).

5
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

of the present study, ωD = 1.1ωSM and ωˆ FW = 1.04ωSM. Fully plastic KSM P2 KSM (Ms3 + Cs2 + Ks)
G2 = =
flywheel stress distributions (i.e. when ω = ωD) can be seen in Fig. 4(a). 1 + GC P1 P2 JSM s (Ms3 + Cs2 + Ks) + KD s 2 + KP s + KI (12)
When ω ≤ ωD stresses will transition from plastic (towards the centre of
the flywheel) to elastic at some radial position REP. Solutions in these M KI M
KD > KP − JSM, KD + JSM > CJS M + KP
two regions are given by Eqs. (8) and (6), respectively. REP may be C KP C (13)
found by enforcing continuity in σr over this boundary. Elastic/plastic
stress distributions for ω = ω̂FW are presented in Fig. 4(b) (note REP is
indicated by a dashed vertical line). 4. SHyKESS transient behaviour

dσr Simulations have been completed using the model developed in

σθ − σr − r = ρr 2ω2
dr (7) Section 3 for several loading scenarios in order to (a) suggest appro-
ρr 2ω2 priate controller gain values and (b) to illustrate the transient response
A
σr = σY − + of SHyKESS as it switches between different modes of operation. So-
3 r (8)
lutions were found in the time domain using Matlab's ODE23tb solver
ρRI 2ωD2 ⎞ and were verified against Simulink results for simulations performed in
A = −RI ⎜⎛pi + σY − ⎟
isolation (i.e. for limited periods of time in a single mode of operation).
⎝ 3 ⎠ (9)
In all plots and for the sake of clarity, normalised values of ωSM, ωFW, T
With the basic dimensions for the system in place, attention can be (the torque transmitted across the DDU) and θEM (defined by the par-
focused on a simple controller that limits the shaft torque through the ticular loading condition) are presented over a normalised time period
DDU in modes 2A and 2B. For convenience, equations presented here (t/ts, where ts is the total simulation time). Note that normalisation
will be given in the Laplace domain. A PID controller has been applied factors are ωSM , TR (the hydraulic machine rated torque) and θ̂EM (the
(GC = KP + KDs + KI/s) that acts on the error (err, see Fig. 5) between assumed load angle at rated power, here 10°).
the instantaneous synchronous machine speed (ωSM) and the desired The partial discharge of the system under constant system load is
value (ωSM , in the present work equal to 50 πrad/s) in order to control shown in Fig. 6. Point A in this figure indicates the transition between
the shaft torque over the DDU (TC in Fig. 5) and hence regulate power system modes 1 and 2A (prior to this point ωSM = ωFW). Discharging is
from the secondary store. This system may be visualised through the continued at constant load in Fig. 7 until point B, where the flywheel
block diagram in Fig. 5. The high frequency response of the DDU is speed is 0 (relative to ground) and the system enters mode 3A. Charging
simulated by a second order damped linear system (P1), of the form of the system commences at point C (mode 2A, represented by a ne-
given in Eq. (10). Values for ωn, ζ and k (100 Hz, 0.3142 and 1, re- gative load angle), causing an increase in ωFW until this speed locks
spectively) were chosen to give a unity response at low frequency (i.e. back into ωSM at point D. The flywheel continues to be charged until
at low frequency TC = T) while allowing for the possibility of resonance point E (where ωSM → ωSMUL), meaning that the DDU is once again
in the DDU at a realistic natural frequency. unlocked. At point F ωFW = ωˆ F W and the system is saturated (mode
Load angle (sometimes called torque angle) is defined as the angle 3B).
between the rotor's magnetomotive force (mmf) and the resultant mmf Step changes in load have been applied during modes 1 and 2 in
[27]. This quantity is used here to simulate a first order approximation Figs. 8 and 9 in order to demonstrate stability in the system. In all cases,
of the mechanical load exerted on SHyKESS (TLoad in Fig. 5). Grid load ωSMLL < ωSM < ωSMUL (i.e. ± 1 % ωSM). Note that the load perturba-
will, in the next section, be approximated by varying the load angle tion in Fig. 8 takes place during mode 1 operation, therefore control is
(θEM), with the associated torque determined by approximating the achieved entirely passively (due to the high inertia of the system) and
synchronous machine as a rotational spring with constant KSM. As- the DDU does not unlock until ωSM → ωSMLL (Fig. 10).
suming a 10° load angle at rated torque, KSM can be estimated as
31,831 Nm/° for the presented design. In practice, reactive power
5. Comments on turnaround efficiency and a case study
would be balanced by adjusting the machine's field strength, in effect
application of SHyKESS
altering the value of KSM, however this is not considered in the present
work. The difference in DDU torque and grid load “torque” is integrated
Determining the turnaround efficiency for the proposed SHyKESS
in P2 in order to determine the synchronous machine speed (P2 = 1/
method is difficult due to range of distinct modes that the system may
JSMs). Note that flywheel speed is omitted from Fig. 5 as the DDU is only
operate in. For example, while operating in mode 1 (as per Fig. 2), it is
concerned with controlling ωSM. The negative of the DDU torque can of
foreseeable that the overall system efficiency is governed by the char-
course be integrated in order to determine ωFW.
acteristics of the synchronous machine (losses in the flywheel are likely
Analysis of the block diagram allows for two transfer functions to be
to be minimal [3] and there are, presumably, no transmission losses if
defined (G1 and G2) such that ωSM (s ) G1 − θEM (s ) G2 = ωSM (s ) , where G1
the DDU is locked). It is worth noting here that, due to the size of the
and G2 are given in Eq. (11, 12), respectively. The Routh-Hurwitz sta-
system, it is expected that SHyKESS will operate in a vertical orienta-
bility criterion may be applied in order to constrain the PID gains in GC
tion supported by magnetic bearings, with rolling element bearings
(see Eq. (13)), however a far more useful condition for KD is found
carrying only a small fraction of the rotor load but providing stiffness to
through inspection of the system. It is clear that err ˙ = −ωṠM and it is
the bearing system. Given that most of the rotor load is supported by
intended that the DDU in modes 2A and 2B should replicate the fly-
magnetic bearings, it is reasoned here that bearing losses will be neg-
wheel inertia. If this is to be achieved, we may say that KD = JFW
ligible in comparison to other sources. When the DDU unlocks (in all
(kg m2). Using this condition, the Routh-Hurwitz result and by testing
other modes of operation) efficiency depends on the electric machine,
candidate KP and KI values for benchmark charge/discharge cases (see
Section 4), the following gains may be determined for the presented
SHyKESS configuration; KP = 1.0 × 105 (kg m2/s) and KI = 2.0 × 105
(kg m2 s−2).
ωn2 / k
P1 =
s2 + ζ 2ωn s + ωn2 (10)

GC P1 P2
G1 =
1 + GC P1 P2 (11) Fig. 5. Block diagram for the control of ωSM in mode 2A and 2B.

6
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Fig. 6. SHyKESS transient response during constant load discharge (partial discharge of the flywheel and secondary store). Note that ts = 3000 s.

hydraulic machines, and power distribution characteristics; turnaround may be calculated for a 24 h period following loss of vacuum.
efficiency is, in the case of SHyKESS, loading history dependent. Before The observations made above underline the importance of operating
presenting an examination of turnaround efficiency values, approx- in a vacuum for high speed flywheels, however several important points
imate models are required that can indicate individual component, should be noted relating to the above calculation. Firstly, laboratory
namely the synchronous electric machine and hydraulic machine, ef- scale discs used to experimentally validate Eq. (14, 15) are typically
ficiency levels. much smaller than the flywheel sized for SHyKESS. In the work of
Flywheel systems can be highly efficient. The numerical analyses Etemad et al., for example, a disc of 92 mm in diameter was considered
presented in this section consider the flywheel storage component of spinning at a maximum speed of 10,472 rad/s [29]. While peripheral
SHyKESS to be 100% efficient, however a brief discussion of related speeds are comparable between Etemad's flywheel and the SHyKESS
losses is given here for completeness and to justify this assumption. It is example, Reynold's number values are several orders of magnitude
expected that a flywheel of the dimensions considered in the present greater for SHyKESS. Eq. (15) is empirically determined and its extra-
work will operate in a vacuum chamber, however the difficulties as- polation to the SHyKESS operating conditions must be acknowledged.
sociated with dynamic sealing make aerodynamic/frictional loss Further fundamental work is required in order to properly validate the
(windage) calculations relevant, with the total loss of operating vacuum losses reported above. Secondly, SHyKESS operating under atmospheric
representing a “worst case” operating condition. conditions would constitute a fault condition. If dynamic sealing can be
Windage losses, that is to say the losses related to aerodynamic achieved a vacuum chamber can be established around SHyKESS's
viscous friction, may be approximated for the wetted sides of a spinning flywheel. Hearn et al. note that the development of flywheel energy
disc using the equations of Daily and Nece [28,29] (see Eqs. (14) and stores for long term storage has been hampered due to, in part, fric-
(15), valid for mixed boundary turbulent flow conditions). A torque due tional losses resulting from windage and magnetic bearings, however
to windage, TW, can be correlated to the dimensionless coefficient CW an operating vacuum of 1 mTorr has been suggested to address these
through Eq. (14). CW may be calculated using Eq. (15), where the concerns [30]. If air density values are for corrected for a 1 mTorr
quantity s/RO is a ratio of the axial air gap and Re is Reynold's number chamber pressure, windage losses drop to 0.15 kW. Self discharge under
(given in Eq. (16)). Using typical values for ambient dry air and as- such conditions is arguably negligible, with a loss in stored energy over
suming a clearance of 0.01 m, losses due to windage at synchronous a 24 h period of 3.6 × 10−3 MWh. This figure may be alternatively
speed may be estimated as 3884 kW, or 19.42% of the SHyKESS system expressed as 0.018% of the total storage capacity of the sized SHyKESS
rated power. It is important to note that this value represents total flywheel. It is worth noting here that Liu et al. [31] have suggested that
windage losses due to viscous effects on both wetted end faces of the low pressure environment windage losses are proportional to the square
spinning disc; losses in the radial air gap are not evaluated. High of angular speed. Magnetic bearing lamination core losses may also be
leakage rates associated with this situation are clearly undesirable; expressed using the same relationship. Functions to define pro-
losses of 18.65 MWh, almost the total capacity of the flywheel store, portionality constants, which are heavily dependent upon flywheel

Fig. 7. SHyKESS transient response during a discharge/charge cycle, highlighting the re-locking of the DDU (moving from mode 2A to 1 during charging, point E)
and flywheel over speed protection (mode 3B, point F). Note that ts = 7000 s.

7
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Fig. 8. SHyKESS transient response during discharge with a step load change during mode 1 operation. Note that ts = 500 s.

geometry, dimensions, and surface roughness, are however difficult to been developed in the literature (for example, in the work of Kerkman
find in the literature. The model of Daily and Nece has therefore been et al. [34] and Jiao et al. [35]) these are considered outside the scope of
used to evaluate low pressure windage losses here. Finally, the structure the present work and a simplified semi-empirical model is assumed
of SHyKESS is such that it may continue to operate through the vacuum here. Scalable machine efficiency maps may be found in the work of
chamber fault outlined at the start of this discussion. The secondary Stipetic and Goss [36] and suggest that machine efficiencies (of the
energy store provides slip energy to account for the difference between scale used here, namely 20 MW) should vary between 85% at low load
the synchronous machine speed and the flywheel speed. If flywheel self- to 98% at rated load. For the present work, an alternator/motor model
discharge is high, the point at which the DDU is unlocked and slip is taken from the work of Notton et al. [37], wherein machine effi-
energy is introduced occurs sooner, but power may still be injected in ciencies are represented by simple polynomials. Eqs. (17) and (18) are
to/taken from the grid by SHyKESS. Given that the SHyKESS system is used here for generator and motor efficiency expressions, giving rise to
intended to operate in at least a partial vacuum, windage losses have ηG and ηM, respectively. These expressions make use of a power fraction
been neglected from the rest of the calculations presented here. argument, P̂SM , the ratio of instantaneous electrical power to rated
electrical power. Plots of generator and motor efficiencies (with respect
1
TW = CW ρAir ω2RO5 to power fraction) can be seen in Fig. 11(a).
2 (14)
P PˆSM
0.062 ηG = ⎛ Elec ⎞ =
⎜ ⎟
2
CW = ⎝ PMech ⎠ ˆ
P + 0.00915738 + 0.0797107PˆSM (17)
(s / RO )0.25Re0.25 (15) SM

ρAir ωRO2 P PˆSM

Re = ηM = ⎛ Mech ⎞ =
⎜ ⎟
2
μAir (16) ⎝ PElec ⎠ PˆSM + 0.01010391 + 0.00731429PˆSM (18)

Determining the fraction of mechanical work that is converted to Hydraulic pump/motor efficiencies (ηP and ηM, respectively) may be
electrical power in a synchronous generator is a task that has received a calculated using Eqs. (19) and (20), respectively. These expressions are
great deal attention in the literature (both in terms of experimental taken from the work of McCandlish and Dorey [38], who built upon the
evaluation and analytical modelling) and, to a degree, the exact char- work of Wilson by introducing linearly variable loss coefficient ex-
acteristics depend upon the choice of standard [32]. Loses in these pressions for flow (QP/M, where subscripts here distinguish pump and
machines are generally categorised into windage/friction, core (ex- motor modes of operation) and torque (TP/M) calculations, thereby
citer), field, stray (eddy current losses in end turn features, for ex- giving more realistic representations of volumetric and frictional losses
ample), and auxiliary system (lubrication/cooling pump requirements, in hydraulic machines, respectively. Expressions for QP/M and TP/M may
for example) related [33]. Although in depth analytical models have be seen in Eqs. (21) and ((22). These expressions incorporate machine

Fig. 9. SHyKESS transient response during discharge with a step load change during mode 2A operation. Note that ts = 1500 s.

8
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Fig. 10. SHyKESS transient response during discharge with a step load change during mode 2A operation (during initial transition settling time). Note that ts = 500 s.

operating conditions and design parameters, namely speed (ω, ex- be seen in Fig. 11(b) and (c), for the pump and motor modes, respec-
pressed in rad/s), displacement (D, expressed in cc/rev), working tively.
pressure difference between inlet and outlet (ΔP, expressed in bar), and
100QP π ΔP Dx
the machine displacement fraction (the fraction of actual displacement ηP =
3xDω TP 2π (19)
to rated machine displacement, assuming a variable displacement ma-
chine and identified by x). Considered working fluid properties are 3xDω TM 2π
dynamic viscosity (μ, here taken to be 0.783 Ns/m2 [23]) and bulk ηM =
100QM π ΔP Dx (20)
modulus (B, here taken as 1.71 × 109 N/m2 [23]), both of which are
assumed to be independent of machine operating condition (i.e., they 3xωD 30 × 106ΔPD ΔPωD 1+x
QP / M = ∓ Cs ∓ ⎛Vr + ⎞
are temperature independent). Note PAtmos is the atmospheric pressure. 100π πμ 1 × 106B ⎝ 2 ⎠ (21)
Loss coefficient (Cs, Cv , and Cf) expressions used here assume a radial
piston design machines and are summarised by Eqs. (23)–(25) [38]. ΔP Dx μωD ΔPD
TP / M = ± Cv ± Cf
Hydraulic machine efficiency maps produced by the above method can 2π 1 × 1092π 2π (22)

Fig. 11. Plots of assumed SHyKESS component efficiency models, showing (a) the synchronous machine, operating in as both a generator and a motor (ηG and ηM,
respectively), (b) the hydraulic machine (pump mode), and (c) the hydraulic machine (motor mode).

9
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

ΔP ⎞
0.5
3ω ηT for an increase in tL. These effects may be readily observed in Fig. 12
Cs = 1.86 × 10−10 ⎛ ⎜ ⎛0.76 + 0.41
⎟ ⎞
250π ⎠ when AL is greater than 20 MW.
⎝ PAtmos ⎠ ⎝ (23)
When AL < 20 MW a drop off in ηT is observed, with a region of
x − 0.25 particularly low efficiency values for high tL (see the bottom right hand
Cv = 1.77 × 105 + (4.05 − 1.77) × 105 ⎛ ⎞
⎝ 0.75 − 0.25 ⎠ (24) corner of Fig. 12). Such behaviour can be explained by considering how
the hydraulic machine “moves” through the efficiency maps presented
3ω 3ω 2⎞ in Fig. 11(b) and (c) during a discharge/charge cycle. For a constant
Cf = 0.031 ⎛1 − 0.51 ⎛
⎜ ⎞ + 0.98 ⎛ ⎞⎟

⎝ ⎝ 250π ⎠ ⎝ 250π⎠ ⎠ (25) load condition (which is easier to consider for the present discussion
than the sinusoidal profile used in Fig. 12), hydraulic machine effi-
The turnaround efficiency of SHyKESS can be investigated using the ciency follows a linear path along a diagonal, the bearing of which is
efficiency models for the synchronous and hydraulic machines de- determined by the magnitude of the system load. Initially, slip speeds
scribed above, however some careful consideration is warranted in are small and only a small fraction of the load power is supplied by the
order to define an appropriate turnaround efficiency metric, re- hydraulic machine (as far as the present discussion is concerned, this
membering that this will almost certainly be dependent upon loading can be thought of as x in Fig. 11(b) and (c)). Returning attention to the
history path. A symmetric discharge/charge loading profile will be as- sinusoidal loading patterns, higher values of AL allow for the larger
sumed in the following discussions, with losses in the flywheel and values hydraulic machine efficiency to be exploited (the machine op-
secondary energy stores evaluated in order to estimate the system ef- erates closer to its rated value for a greater portion of the loading cycle).
ficiency. It is assumed that SHyKESS is initially locked (in mode 1, see For comparatively low values of AL however the hydraulic machine
Fig. 2) and is subjected to a sinusoidal loading profile (see Eq. (26)), power is consistently low compared to the machine's rated value, re-
where PL is the instantaneous grid power. Note that the same sign sulting in low machine efficiencies. Similar effects can be observed in
convention is used here as in other sections of the present work, the synchronous machine (i.e. when it operates at a low power frac-
whereby a positive power indicates a grid demand and necessitates that tion), thereby leading to relatively low values of ηT when the DDU
SHyKESS will be discharged. This allows for the definition of a loading unlocks at low AL values. At very low values of AL, ηT is almost com-
power amplitude (AL) and period (tL), which can be used to characterise pletely independent of tL as the depth of discharge/charge is not suffi-
a wide range of potential grid imbalance scenarios. Turnaround effi- cient to unlock the DDU over the loading period. It should be recalled
ciency for SHyKESS (ηT) is defined by Eq. (27), where ΔEFW is the that windage losses have not been consider in the generation of Fig. 12.
change in flywheel energy over the loading period and ΔEHY is the These additional losses, which at most amount to 660 kW, are likely to
change in energy in the secondary store. A “transaction” energy is de- have little influence at high AL values, where the flywheel store is
fined (ETran = (2ALtL)/π) that quantifies the total amount of energy quickly discharged and DDU unlocks relatively early in the simulation.
extracted from/injected into the grid. Eq. (27) is judged to be a re- At lower AL values windage is expected to have a significant effect as its
presentative measure of system efficiency (in this case) as it is antici- magnitude is similar to that of the peak transaction power. This ob-
pated that SHyKESS will typically operate at relatively low slip speeds servation further indicates why SHyKESS may not be suitable for use in
and around synchronous speed. Self discharge of the flywheel is as- low partial load conditions.
sumed to be negligible, allowing the system can remain in a charged In order to demonstrate the applicability of SHyKESS and its ability
and locked state for long periods of time. A detailed discussion of this to reinstate real inertia in grids where it has been removed, we here
point may be found in the conclusions section of this paper, however it present a simple case study for an isolated grid that has a baseline
should be noted that additional fundamental work is required to de- supply which is augmented by a photovoltaic (PV) power station
termine windage losses, for example, in systems on the scale of SHy- (which has little intrinsic real inertia). A fictional Californian grid is
KESS before enhanced efficiency maps can be created. SHyKESS turn- considered here in order to provide a suitable setting for a PV power
around efficiency is plotted in Eq. (27), with respect to loading station.
amplitude (AL) and period (tL). A PV cell model is implemented here that is based on the work
Durisch et al. A semi-empirical efficiency model was developed in this
t
PL (t ) = AL sin ⎛ 2π ⎞
⎜ ⎟ work and verified against observations made in Jordan (Al Quawairah)
⎝ tL ⎠ (26)
[39] over a three year period [40], with further comparisons made
|ΔEFW | + |ΔEHY | against Californian and Alpine data (making its application in the
ηT = 1 −
ETran (27)

Several interesting features can be noted from Fig. 12 which illu-

minate certain subtle aspects of SHyKESS operation. For very low va-
lues of tL turnaround efficiencies are typically high (over 0.8). At low
depths of discharge/charge SHyKESS efficiency is almost entirely de-
termined by the synchronous machine; the DDU remains locked (mode
1) and all power is extracted from/injection into the flywheel. Note a
lower threshold value of 0.85 is applied to the electric machine in the
current work to avoid difficulties associated with electrical machines
operating at low power fractions. Similarly, a lower efficiency threshold
of 0.3 is applied to the hydraulic machine. It is easy to imagine that as
AL increases in magnitude the time at which the DDU will unlock
(moving from mode 1 into mode 2 operation) will reduce. As tl in-
creases therefore (for a given value of AL), a great proportion of power
must be supplied through the DDU (that is to say, from the secondary
store) during the loading cycle. Although, generally speaking, operating
the hydraulic machine at higher speeds and closer to rated power re-
sults in larger component efficiencies, the fact remains the energy
transactions in this range of operation experience losses in both the Fig. 12. The derived SHyKESS efficiency map (for sinusoidal supply/demand
synchronous machine and hydraulic machine, leading to reductions in loading conditions).

10
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

present case study appropriate). The developed efficiency model is Notton's interpretation [43,37] of Kasten and Young's revised air mass
summarised by Eq. (28), where the parameters hPV (the Ross coefficient tables [44] (see Eq. (32)).
[39]), pPV, qPV, mPV, rPV, sPV, and uPV are manufacturer/model specific. 1
In the present work, a Siemens CIS ST40 copper-indium-diselenide MAir =
cos(ψz ) + 0.50572(96.07995 − ψz )−1.6364 (32)
(CuInSe2) is adopted, for which the parameters presented in Table 1
may be assumed (note individual modules have an area, A, of Finally, inverter efficiency for the PV cell is based on a model used
0.423 m2). Note that in Eq. (28), Gβ is the global solar irradiation, θPV is extensively by Notton et al. [43,37] (see Eqs. (33)–(35)), where
the panel temperature, and MAir is a (normalised) air mass. For re- η10 = 0.87 and η100 = 0.945. Note that P̂ is the ratio of the inverter
ference, a typical working efficiency of the selected PV module (where output power to the rated inverter power.
600 W/m2 ≤ Gβ ≤ 800 W/m2, θpv = 25 °C, and MAir = 1.5) is approxi-
Pˆ
mately 12% [39]. Several site specific meteorological variables (namely ηINV =
global solar irradiation, Gβ, air temperature, θAir), and normalised air (P + P0 + kPˆ )2
ˆ (33)
mass, MAir, must be defined in order for Eq. (28) to be implemented.
1 ⎛ 10 1
Some attention will now be given to appropriate models that described P0 = ⎜ − − 9⎞⎟
these parameters in California. 99 ⎝ η1 0 η100 ⎠ (34)
mPV
Gβ Gβ ⎞ θPV M 1 ⎞
⎛
ηPV = pPV ⎜qPV + ⎜⎛
⎞⎛
+ sPV Air k = ⎜⎛ ⎟ − P0 − 1
⎟
⎟ ⎜1 + rPV θ
⎝
Gβ,0 G
⎝ β,0 ⎠ ⎠⎝ PV,0 MAir,0 ⎝ η100 ⎠ (35)
uPV
M ⎞ Using the PV model (and related equations) expressed above, a PV
+ ⎜⎛ Air ⎟⎞ ⎟
M (28) cell's response in California may be approximated. A plot of the daily
⎝ Air,0 ⎠ ⎠
variations in air temperature (θAir), extraterrestrial solar irradiance
Solar irradiance is modelled here using the work of Sung et al., (IETI), global solar irradiance (Gβ), and PV cell output power (after the
which was validated against data taken from Anatolia-Rancho Cordova, inverter, PPV) can be seen in Fig. 13(a). which was generated assuming
California [41]. This model allows for both long and short term weather the California representative parameters reported in the present work.
induced variations, see Eqs. (29) and (30). In these equations, IETI is the A scaled demand profile (PDemand) has been assumed to be re-
maximum extraterrestrial solar irradiation, dn is the day within the year presentative of a Californian grid (see Fig. 13(b)). Assuming a baseline
(samplings in mid July are considered here), ψz is the zenith angle (in load (PBase, here taken to be the first 15 terms in a Fourier decom-
degrees), t is a specific time in the day (expressed in hours), trise is the position of the demand signal, representing temporal frequencies be-
sunrise time, and tset in the sun set time. The following model para- tween 1.6534 × 10−6 Hz and 2.4802 × 10−5 Hz) and superimposing
meters are assumed to be representative of California, M = 0.76, the PV power station generation (PPV,Total), temporally varying supply
a1 = 0.1, b1 = 7, c1 = 1, a2 = 0.5, b2 = 40, c2 = 2, and dn = 180 [41]. (PGen) and demand (PDemand) fluctuations may be generated (see
c1 Fig. 14(a)). Inspection of these profiles has identified three loading
π (t − trise ) ⎞ ⎛ ⎛ π (t − trise ) ⎞ events which are interest to SHyKESS (i.e. cases where there are short
Gβ = Smax sin ⎛ ⎜
⎜1 − a1 sin b1
⎟ ⎜ ⎟

⎝ tset − trise ⎠ ⎝ ⎝ tset − trise ⎠ term energy imbalances which may be extended into the long term).
c2 These are summarised in Fig. 14(b)–(d) (in subsequent discussions
π (t − trise ) ⎞ ⎞
− a2 sin ⎛b2 ⎜ ⎟
⎟ these will be referred to as events 1, 2, and 3, respectively). Transient
⎝ tset − trise ⎠ ⎠ (29) SHyKESS responses are plotted for the three loading events (assuming
the same 50 MW 20 MW Hr design discussed previously) in Figs. 15–17
2πdn ⎞ ⎞
Smax = M max(IETI) = M max ⎜⎛1362 ⎛1 + 0.033 cos ⎛ ⎜ cos(ψz ) ⎞⎟ ⎟
(note the same normalisation convention is used as in Section 4). It
⎝ ⎝ ⎝ 365 ⎠ ⎠ ⎠ should be noted that, in all cases, SHyKESS initially presents a high
(30) inertia to the grid in order to limit the rate of change of frequency and,
after the DDU unlocks, the simple PID controller is sufficient to return
PV cell temperature (θPV) is assumed to be governed by
the frequency back to it nominal operating value (this is the case even
θPV = θAir + hPVGβ, where hPV is the Ross coefficient (see Table 1). An
when presented with loading patterns that are representative of real
approximation of daily air temperature variations is now required and
grids). These results indicate that SHyKESS is suitable for controlling
is here based on the work of Cesaraccio et al. (particularly related to the
grid frequency over a wide range of time scales, with a significant
Calipatria weather station in California) [42]. The TM model splits
amount of power being provided by a flywheel in all cases.
diurnal temperature variations into three regions, based on time of day
relative to sun rise and sun set (each controlled by a function, see Eq.
6. Conclusions and future work
(31)). In this model, θ̂Air is the peak air temperature during the day
(occurring at the time t θ̂Air ), θAir,n is the air temperature at trise (θAir,p is
Replacing system inertia in a decarbonised energy market has been
the air temperature at trise the following day; it is assumed here that
identified by many parties as an area that needs investment. A hybrid
θAir,p = θAir,n), and θAir,o is the air temperature at tset. A summary of
energy storage system (SHyKESS) is presented here that incorporates 2
assumed parameters used for air temperature modelling (and solar ir-
radiance modelling) can be found in Table 2 (note that these are largely
Table 1
taken from the work of Cesaraccio et al. [42]). Note that, in the three
Assumed Durisch model PV cell parameters for a
phases of Eq. (31), the following substitutions are used; α = θˆAir − θAir, n ,
Siemens CIS ST40 CuInSe2 module [39].
R = θˆAir − θAir, o , and b = (θAir, p − θAir, o)/( trise − tset ) .
Parameter Value
t−t
θAir, n + α sin ⎡ ⎛ t −rise ⎞ π ⎤, if trise < t ≤ t θˆAir 2
hPV (°C m /W) 0.032
⎢ ˆAir trise 2 ⎥
⎣ ⎝ θ ⎠ ⎦ pPV 0.1855
θAir (t ) = π t−tˆ π qPV −0.3288
θAir, o + R sin ⎡ 2 + ⎛ t − θtAir ⎞ 2 ⎤, if t θˆAir < t ≤ tset mPV 0.2612
⎢
⎣ ⎝ set θˆAir ⎠ ⎥⎦ rPV −0.10039
θAir, o + b t − tset , if tset < t ≤ trise (31) sPV −0.9678
uPV 0.9864
Normalised air mass (MAir) is correlated to ψz (zenith angle) through

11
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Table 2 faults could be cleared or system protection could take effect. The
A summary of parameters used the solar ir- flexibility that a system such as SHyKESS would offer during both
radiance and air temperature modelling normal operating and fault conditions is an interesting feature of hybrid
(representative of California). energy stores that incorporate synchronous machines. Future work will
Parameter Value look to explore this area in detail.
The modelling procedure implemented here has used a first order
31.0
θ̂Air (°C) approximation of a synchronous machine. A constant field strength is
θAir,n (°C) 13.8
assumed, allowing the machine to be approximated by a constant
θAir,o (°C) 25.0
trise 06:00
stiffness rotational spring. While this has been judged to be sufficient
t θ̂Air 13:00 for the present work (the main objective in here is to introduce the
tset 20:30 SHyKESS concept and illustrate how such a system can operate in
several modes), this issue must be addressed in future work.
Furthermore, refined efficiency models for the electric and hydraulic
energy stores (facilitating both fast and slow energy storage) which machines should be investigated, particularly with reference to start up
attempts to address this issue. The use of flywheel energy storage is conditions.
attractive due to the inherently high levels of inertia, however fly- Numerous novel flywheel designs have been proposed which could
wheels also benefit from high cycle and operational lives, high turn- form part of a SHyKESS design and will be investigated. For example,
around efficiencies, high energy densities and low environmental im- integrated flywheel, motor/generator and magnetic bearing systems in
pacts [15,45]. The proposed system belongs to the electricity in/ the work of Kailasan et al. [46] benefited from reduced shaft lengths.
electricity out class of stores and is built on a hollow shaft in such a way Additional flexibility in the design is present through the choice of
that torque is not explicitly grounded to earth. As such, losses in the flywheel and transmission shaft materials [47] and shape of the fly-
system may be minimised and the support structures reduced (i.e. re- wheel (for example, the use of constant stress flywheel cross sections
action torques through component casings are minimised). An example with or without an outer retaining ring [47]), which are generally
50 MW system with 20 MWh of capacity has been sized and simulated, characterised by the shape factor K [48]. Of particular note will be the
with appropriate gain levels found for a PID controller that dictates the development of low leakage rotating unions that can accommodate the
power contributions from both energy stores in certain modes of op- intended working pressures (200 bar) of the compressed fluid store.
eration. Simulations considered both idealised step loading histories Flywheel sizing in the present work was conducted using an elastic-
and grid representative examples and, in all cases, synchronous ma- perfectly-plastic material (approximating EN24 steel) and assuming a
chine speeds remained within industry accepted tolerances; this could Tresca yield criterion. The validity of this solution was questioned in
be realised with in a system with only normal (i.e. thermal power plant) the work of Gamer [49] due to non-continuous displacement field so-
levels of power electronics. No frequency control power electronics are lutions in partial plasticity cases and negative radial plastic strains
required. Large inertia constants may be calculated (1440 s for the (associated with tensile radial stresses) for solid disc applications (are
developed system) and, during certain mode of operation, there is no the centre of the disc is considered). In order to address these points,
ambiguity as to whether this inertia is “seen” by the grid. Assuming Gamer developed a solution using a Prager hardening rule (linear ki-
steel prices of £2000/tonne, unit energy storage costs of approximately nematic hardening) which still implemented the Tresca criterion. More
111.5£/kW hr are achievable with this system. Almost no cyclic de- recently, solutions using a von Mises criterion have been found [50].
gradation would be expected and control (when required) can be both The exact implications of these refinements on the flywheel design is
simple and localised (the system is intrinsically un-hackable and is re- presently unclear and future work should look to quantify the potential
sistant to cyber attack). Furthermore, the inclusion of a synchronous for design enhancements.
machine in SHyKESS offers an advantage over power electronics con- By fluctuating rotational speed, tensile stresses experienced in fly-
trolled synthetic inertia systems. Power electronics are often sized wheels will also fluctuate, meaning that fatigue becomes a legitimate
based on a nominal power and are not suited to operating in overload life limiting failure mechanism. This observation is particularly true in
conditions for extended periods of time (typically on the millisecond the case of SHyKESS, where the extended operational range made
time scale). Synchronous machines on the other hand can typically be possible by the DDU offers the potential for a larger alternating stresses.
operated at overrated power levels for several seconds before thermal A brief examination of this failure mechanism is given here using the
and structural concerns limit continued operation. It is foreseeable that, Basquin model, which is summarised by Eq. (36). In Eq. (36), α and β
under fault conditions, high current levels would be required such that are material/surface condition dependent parameters which relate the

Fig. 13. Plots of (a) generated diurnal solar panel response (PPV based on a Siemens CIS ST40 CuInSe2 in California, USA, in July) per unit area and related parameters
and (b) the assumed demand profile.

12
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Fig. 14. Plots of (a) the generated power supply and demand profiles (with constitutive baseline and PV components plotted independently), (b) identified case study
event 1, (c) identified case study event 2, and (d) identified case study event 3.

Fig. 15. SHyKESS transient response for event 1 (see Fig. 14).

number of cycles to failure Nf to the “endurance” alternating stress Δσ loading pattern, wherein the flywheel operates between synchronous
for fully reversed conditions. Values for α and β are here taken as 9.84 speed and a 50% charged state. Using Eqs. (7)–(9) stresses in the hoop
and 4.56 × 1030, respectively, being representative of a smooth surface direction may be calculated for these conditions, yielding a result that
component made from medium carbon steel (similar to the EN24 ma- SA = 295 MPa and SM = 885 MPa. Substituting these values into Eq.
terial referenced above) [51]. Eq. (36) is only valid for fully reversed, or (37) indicates a required endurance stress of 529 MPa. From Eq. (36),
0 mean stress, loading conditions, therefore the well known Goodman these conditions suggest a failure life of approximately 7200 cycles. The
relationship must be used to account for mean stress effects. The above is a first order calculation which extends methods that are
Goodman relationship may be expressed by Eq. (37), where SA and SM usually reserved for high cycle fatigue calculations into scenarios where
are the alternating and mean stresses for a particular loading condition, low cycle methods may be more appropriate. What is revealed however
respectively, σe is the endurance limit for a given number of cycles to are the potentially significant fatigue effects which can limit the life of
failure, here calculated by Eq. (36), and σUTS is the ultimate tensile flywheel components, particularity if designs which look to utilise
strength of the material, here taken to be 2000 MPa. Attention is re- material effectively are sought. Future work will explore these limita-
stricted in the present discussion to an extended discharge of flywheel tions more closely and will look to account for variable loading cycles

13
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Fig. 16. SHyKESS transient response for event 2 (see Fig. 14).

Fig. 17. SHyKESS transient response for event 3 (see Fig. 14).

through damage summation methods such as Miners rule. 2003.

[7] M. Dreidy, H. Mokhlis, S. Mekhilef, Inertia response and frequency control tech-
Δσ αNf = β (36) niques for renewable energy sources: a review, Renew. Sustain. Energy Rev. 69
(2017) 144–155.
S [8] M. Deepak, R.J. Abraham, F.M. Gonzalez-Longatt, D.M. Greenwood, H.S. Rajamani,
SA = σe ⎛1 − M ⎞
⎜ ⎟
A novel approach to frequency support in a wind integrated power system, Renew.
⎝ σUTS ⎠ (37) Energy 108 (2017) 194–206.
[9] J. Zhu, G.P.A.A.J.R. Campbell, D. Booth, C.G. Bright, Inertia emulation control
As discussed in Section 3, the flywheel considered in the present strategy for VSC-HVDC transmission systems, IEEE Trans. Power Syst. 28 (2013)
work is expected to be made up of several plates which are held axially. 1277–1287.
This helps to avoid several major manufacturing and logistic chal- [10] Y. Pipelzadeh, B. Chaudhuri, T.C. Green, Inertial response from remote offshore
wind farms connected through VSC-HVDC links: a communication-less scheme,
lenges, however it does necessitate the inclusion of additional features 2012 IEEE Power and Energy Society General Meeting (2012).
in the flywheel which may act as locations of stress concentration. [11] G.N. Prodromidis, F.A. Coutelieris, Experimental and theoretical investigation of
Future work will look to clarify the nature of these features and de- flywheel-based energy storage in off-grid power plants using renewables, J. Energy
Eng. 142 (2016) 04014055-1-8.
termine the likelihood of premature failure due to their inclusion. [12] G.N. Prodromidis, F.A. Coutelieris, Simulations of economical and technical feasi-
bility of battery and flywheel hybrid energy storage systems in autonomous pro-
Acknowledgements jects, Renew. Energy 39 (2012) 149–153.
[13] R. Sebastián, R. Peña Alzola, Control and simulation of a flywheel energy storage
for a wind diesel power system, Electr. Power Energy Syst. 64 (2015) 1049–1056.
The presented work is funded by the EPSRC (EP/R001251/1) and [14] C. Carrillo, A. Feijóo, J. Cidrás, Comparative study of flywheel systems in an iso-
incorporates collaborations with the Energy Technologies Institute lated wind plant, Renew. Energy 34 (2009) 890–898.
[15] M.E. Amiryar, K.R. Pullen, A review of flywheel energy storage system technologies
(ETI), Power Continuity Systems and Quartzelec Ltd.
and their applications, Appl. Sci. 7 (2017) 1–21.
[16] R. Sebastián, R. Peña Alzola, Flywheel energy storage systems: review and simu-
References lation for an isolated wind power system, Renew. Sustain. Energy Rev. 16 (2012)
6803–6813.
[17] S.O. Amrouche, D. Rekioua, T. Rekioua, S. Bacha, Overview of energy storage in
[1] B. Sørensen, Energy Intermittency, CRC Press, 2014. renewable energy systems, Int. J. Hydrog. Energy 41 (2016) 20914–20927.
[2] P. Jain, Wind Energy Engineering, McGraw Hill, 2010. [18] T. Bocklisch, Hybrid energy storage systems for renewable energy applications,
[3] F. Díaz-González, A. Sumper, O. Gomis-Bellmunt, Energy Storage in Power Systems, Energy Proc. 73 (2015) 103–111.
Wiley, 2016. [19] G. Shankar, V. Mukherjee, Load frequency control of an autonomous hybrid power
[4] A. Keyhani, M.N. Marwali, M. Dai, Integration of Green and Renewable Energy in system by quasi-oppositional harmony search algorithm, Int. J. Electr. Power
Electric Power Systems, Wiley, 2010. Energy Syst. 78 (2016) 715–734.
[5] F. Bassi, L. Caciolli, G. Giannuzzi, N. Corsi, A. Giorgi, Use of hidden inertia from [20] I. Hamzaoui, F. Bouchafaa, A. Talha, Advanced control for wind energy conversion
wind generation for frequency support in power grids, AEIT International Annual systems with flywheel storage dedicated to improving the quality of energy, Int. J.
Conference (AEIT) (2016). Hydrog. Energy 41 (2016) 20832–20846.
[6] D.P. Kothari, I.J. Nagrath, Modern Power System Analysis, McGraw-Hill Education, [21] U. Government, The Electricity Safety, Quality and Continuity Regulations, (2002)

14
J.P. Rouse et al. Journal of Energy Storage 20 (2018) 1–15

Available from: http://www.legislation.gov.uk/uksi/2002/2665/pdfs/uksi_ context, Sol. Energy 157 (2017) 20–34.

20022665_en.pdf .. [38] D. McCandlish, R.E. Dorey, The mathematical modelling of hydrostatic pumps and
[22] S. Hameer, J.L. van Niekerk, A review of large-scale electrical energy storage, Int. J. motors, Proc. Inst. Mech. Eng. B: J. Eng. Manuf. 198B (1984) 165–174.
Energy Res. 39 (2015) 1179–1195. [39] W. Durisch, B. Bitnar, J.C. Mayor, H. Kiess, K.H. Lam, J. Close, Efficiency model for
[23] T. Hunt, N. Vaughan, Hydraulic Handbook, Elsevier, 1996. photovoltaic modules and demonstration of its application to energy yield estima-
[24] D.W.A. Rees, The Mechanics of Solids and Structures, Imperial College Press, 2016. tion, Sol. Energy Mater. Sol. Cells 91 (2007) 79–84.
[25] D.W.A. Rees, Elastic–plastic stresses in rotating discs by von Mises and Tresca, J. [40] W. Durisch, J. Keller, W. Bulgheroni, L. Keller, H. Fricker, Solar irradiation mea-
Appl. Math. Mech. 79 (1999) 281–288. surements in Jordan and comparisons with Californian and Alpine data, Appl.
[26] RoyMech. Available from: http://www.roymech.co.uk/Useful_Tables/Matter/ Energy 52 (1995) 111–124.
Strength_st.htm. [41] T. Sung, S.Y. Yoon, K.C. Kim, A mathematical model of hourly solar radiation in
[27] G. Klempner, I. Kerszenbaum, Handbook of Large Turbo-Generator Operation and varying weather conditions for a dynamic simulation of the solar organic Rankine
Maintenance, 2nd ed., Wiley, 2008. cycle, Energies 8 (2015) 7058–7069.
[28] J.W. Daily, R.E. Nece, Chamber dimension effects on induced flow and frictional [42] C. Cesaraccio, D. Spano, P. Duce, R.L. Snyder, An improved model for determining
resistance of enclosed rotating disks, Trans. ASME J. Basic Eng. 82 (1960) 217–232. degree-day values from daily temperature data, Int. J. Biometeorol. 45 (2001)
[29] M.R. Etemad, K. Pullen, C.B. Besant, N. Baines, Evaluation of windage losses for 161–169.
high-speed disc machinery, Proc. Inst. Mech. Eng. 206 (1992) 149–157. [43] G. Notton, V. Lazarov, L. Stoyanov, Optimal sizing of a grid-connected PV system
[30] C.S. Hearn, M.C. Lewis, S.B. Pratap, R.E. Hebner, F.M. Uriarte, D. Chen, et al., for various PV module technologies and inclinations, inverter efficiency char-
Utilization of optimal control law to size grid-level flywheel energy storage, IEEE acteristics and locations, Renew. Energy 35 (2010) 541–554.
Trans. Sustain. Energy 4 (2013) 611–618. [44] F. Kasten, T. Young, Revised optical air mass tables and approximation formula,
[31] H.P. Liu, M.D. Werst, J.J. Hahne, D. Bogard, Investigation of windage splits in an Appl. Opt. 28 (1989) 4735–4738.
enclosed test fixture having a high speed composite rotor in low air pressure en- [45] G.S.M. Mousavi, F. Faraji, A. Majazi, K. Al-Haddad, A comprehensive review of
vironments, 12th Symposium on Electromagnetic Launch Technology (2004). flywheel energy storage system technology, Renew. Sustain. Energy Rev. 67 (2017)
[32] A. Sumper, A. Baggini, Electrical Energy Efficiency: Technologies and Applications, 477–490.
Wiley, 2012. [46] A. Kailasan, T. Dimond, P. Allaire, D. Sheffler, Design and analysis of a unique
[33] J.J. Cathey, Electric Machines: Analysis and Design Applying MATLAB, McGraw- energy storage flywheel system – an integrated flywheel, motor/generator, and
Hill Inc., 2001. magnetic bearing configuration, J. Eng. Gas Turbines Power 137 (2015) 042505-
[34] R.J. Kerkman, P.C. Krause, T.A. Lipo, Simulation of a synchronous machine with an 1-9.
open phase, Electr. Mach. Power Syst. 1 (1977) 245–254. [47] R.A. Huggins, Energy Storage, Springer, 2010.
[35] N. Jiao, W. Liu, J. Pang, Z. Zhang, Y. Jiang, Integrated model of brushless wound- [48] B. Bolund, H. Bernhoff, M. Leijon, Flywheel energy and power storage systems,
rotor synchronous starter-generator based on improved parametric average-value Renew. Sustain. Energy Rev. 11 (2007) 235–258.
model of rotating rectifier, Int. J. Electr. Power Energy Syst. 96 (2018) 106–113. [49] U. Gamer, Tresca's yield condition and the rotating disk, J. Appl. Mech. 50 (1983)
[36] S. Stipetic, J. Goss, Calculation of efficiency maps using scalable saturated flux- 676–678.
linkage and loss model of a synchronous motor, 22nd International Conference on [50] N. Aleksandrova, Exact deformation analysis of a solid rotating elastic-perfectly-
Electrical Machines, ICEM 2016 (2016). plastic disk, Int. J. Mech. Sci. 88 (2014) 55–60.
[37] G. Notton, D. Mistrushi, L. Stoyanov, P. Berberi, Operation of a photovoltaic-wind [51] C. Li, W. Dai, F. Duan, Y. Zhang, D. He, Fatigue life estimation of medium-carbon
plant with a hydro pumping-storage for electricity peak-shaving in an island steel with different surface roughness, Appl. Sci. (2007) 7.

15