Control Engineering Practice 123 (2022) 105142

Contents lists available at ScienceDirect

Control Engineering Practice

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

A model predictive control scheme for regenerative braking in vehicles with

hybridized architectures via aftermarket kits
V. Mariani a ,∗, G. Rizzo b , F.A. Tiano b , L. Glielmo a
a Department of Engineering, University of Sannio, Piazza Roma 21, 82100 Benevento, Italy
b
Department of Industrial Engineering, University of Salerno, Via Giovanni Paolo II 132, 84084 Fisciano (SA), Italy

ARTICLE INFO ABSTRACT

Keywords: Regenerative braking is an established technique that can improve the driving range of vehicles with full-
Model predictive control electric or hybrid/hybridized powertrain, thus promoting their large-scale adoption. In this paper, the design
Regenerative braking of an MPC strategy for maximizing regenerative braking in a real vehicle that has been hybridized by means of
Hybridized electric vehicles
a kit is presented. The kit is entirely built by the partners of the European Horizon2020 LIFE-SAVE (Solar Aided
Through-the-road architectures
Vehicle Electrification – LIFE16 ENV/IT/000442) funded project, and features two in-wheel-motors placed on
Aftermarket kits
the rear axle, PV panels installed on the roof and the bonnet and a battery pack placed in the trunk. The
proposed MPC strategy will serve as a baseline for the implementation of an additional algorithm to the
actual version of the hybridized powertrain control such that the new kit will endow the vehicle also with
regenerative braking. Since the kit cannot access the vehicle’s ECU in order to avoid code infringements, the
MPC relies only on information that can be retrieved via the vehicle’s CAN. In addition to the maximization
of the energy recovery during braking, the proposed strategy considers aspects such as the in-wheel-motors
non-ideal efficiency and rated specifications, the constraints given by the brake operating region and possible
comfort requirements that have a positive impact on the driver’s experience. Furthermore, it also includes
a penalty term to force the controller to trade-off with possible high-efficiency operations. The simulation
outcomes show the effectiveness of the proposed MPC strategy, enabling the recovery up to approximately
18% of the vehicle’s kinetic energy, and suggest that it can be fruitfully used as a baseline for an additional
regenerative braking algorithm to be implemented onto the real vehicle.

1. Introduction In hybrid powertrains, polluting emissions and fuel consumptions can

be kept low by taking advantage of the different utilization ranges
Vehicles with electric powertrain are going to be widely used in the of ICEs and electric motors which operations have to be properly
future since their adoption can promote the reduction of polluting emis- orchestrated (de Jager et al., 2013). In general, what is called fuel
sions at a large scale (European Environment Agency, 2018; Pridmore autonomy, is still a main issue. In this regard, a flourishing literature
et al., 2018). Scientific research is putting extra effort to address several exists. Some authors propose new architectures/configurations of the
issues that prevent their actual diffusion, among which the limited hybrid powertrain Yuan et al. (2019), some prefer to improve drivers’
driving range is one of the most relevant (Baek & Chang, 2019; Bourass behavior through appropriate eco-driving policies (Maamria et al.,
et al., 2017; Sautermeister et al., 2018; Tang et al., 2019; Vatanparvar 2019; Padilla et al., 2018), in other papers the authors focus on the
et al., 2019). In general, advancements in battery technologies are speed profile (Chen et al., 2019; Uebel et al., 2018), on optimizing gear
strongly needed. As long as capacity, size, costs, operations, reliability shift (Li & Görges, 2019), on optimizing battery operations in the long
and safety requirements will not meet the automotive standards, the run (i.e., battery-degradation oriented) (Xie et al., 2020) or in the short
electric powertrains spread will be hindered. Meanwhile, the industry term (i.e., during driving) (Kim & Ahn, 2019), on optimizing power
has already started to sell cars with hybrid powertrains, where Internal consumption from auxiliaries such as the cooling system (Chen et al.,
Combustion Engines (ICEs) and electric motors jointly operate in order 2020) and even on the design stage, where decreasing the computa-
to provide the required power/torque, as an intermediate solution. tional effort allows to increase the amount of used data thus achieving

∗ Corresponding author.
Linkedin: valerio-mariani-ee (V. Mariani), Linkedin: gianfranco-rizzo-83596911 (G. Rizzo), Linkedin: francesco-tiano (F.A. Tiano), Linkedin:
luigi-glielmo-62158b10 (L. Glielmo).
E-mail addresses: vmariani@unisannio.it (V. Mariani), grizzo@unisa.it, ceo@eproinn.com (G. Rizzo), ftiano@unisa.it (F.A. Tiano), glielmo@unisannio.it
(L. Glielmo).

https://doi.org/10.1016/j.conengprac.2022.105142
Received 12 May 2021; Received in revised form 7 December 2021; Accepted 26 February 2022
Available online 22 March 2022
0967-0661/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Glossary 𝐹0𝑅 Maximum rear axle’s brake without wheels’

slip in case only the rear brake is engaged
4WD Four Wheel Drive
𝐹𝑃𝐹 Maximum front axle’s brake without
ABS Anti-lock Braking System
wheels’ slip in case both brakes are
BMS Battery Management System
engaged
ECM Equivalent Consumption Minimization
𝐹𝑃𝑅 Maximum rear axle’s brake without wheels’
EMF Electro-Magnetic Force slip in case both brakes are engaged
EN Electric Node 𝐹𝐵,𝐸 Total electric braking force
EV Electric Vehicle 𝐹𝐵,𝐻 Total hybrid braking force
HEV Hybrid Electric Vehicle 𝐹
𝐹𝐵,𝐻 Front axle’s hybrid braking force
ICE Internal Combustion Engine 𝑅
𝐹𝐵,𝐻 Rear axle’s hybrid braking force
IWM In-Wheel Motor
𝐹𝐵,𝑀 Total mechanical braking force
Li-Ion Lithyum-Ion 𝐹
𝐹𝐵,𝑀 Front axle’s mechanical braking force
MLD Mixed-Logic Dynamic 𝑅
𝐹𝐵,𝑀 Rear axle’s mechanical braking force
MPC Model Predictive Control
𝑔 Gravity acceleration
OBD On-Board Diagnosis
𝐽IWM Comfort cost function
PV Photo-Voltaic
𝑚 Discrete time-step
SoC State-of-Charge
𝑀𝑒 Vehicle’s equivalent mass
TTR Through-The-Road
𝑀ℎ Control/prediction horizon (in number of
VMU Vehicle Management Unit samples)
List of Symbols 𝑀𝑣 Vehicle’s mass
𝑃 Power (can be specified via
1 − 𝛼𝐻 Rear axle’s to total hybrid braking force
sub/superscripts according to the scope it
ratio
is applied to)
1 − 𝛼𝑀 Rear axle’s to total mechanical braking
𝑟𝑤 Wheels’ radius
force ratio
𝑇 Torque (can be specified via
𝛼𝐻 Front axle’s to total hybrid braking force
sub/superscripts according to the scope it
ratio
is applied to)
𝛼𝑀 Front axle’s to total mechanical braking
𝑇𝑠 Sampling time
force ratio
𝑣 Vehicle’s speed
𝛥𝐹̄𝐵𝐸 Maximum desired variation of the electric
𝑤IWM Weight of 𝐽IWM in the optimizer
brake
𝜂 Electric motor’s efficiency (can be specified
via sub/superscript according to the scope
it is applied to)
in combination with free-wheeling and a-priori known speed profiles;
𝛼̂ 𝐻 Maximum enabled front axle’s to total convex optimization and dynamic programming techniques are used in
hybrid braking force ratio Platt et al. (2018), Yang et al. (2019), respectively; the Pontryagin’s
𝜇 Tire–road grip coefficient (can be indexed) Minimum Principle is used in Lee et al. (2019), Zheng et al. (2018);
𝜔 Angular speed (can be specified via the Equivalent Consumption Minimization (ECM) approach is used
sub/superscripts according to the scope it in Liu et al. (2019), Rezaei et al. (2019); in Kamal and Adouane
is applied to) (2018) the authors propose a hierarchical control architecture where,
𝜌 Air density at each level, artificial neural networks and/or fuzzy logic paradigms
𝜎 Road’s slope are used; a heuristic approach is presented in Shabbir and Evangelou
𝐴 Vehicle’s front area (2016) while other authors suitably divide the problem into practical
𝑐𝑥 Drag coefficient subproblems and explicitly use traffic information enabled by con-
𝐸bat Energy recovery cost function nected vehicle technologies (Zulkefli & Sun, 2019). Overviews and
𝐹𝐹 Front axle’s braking force gauge comparative studies can also be easily found. An interesting review on
energy management systems for hybrid powertrains can be found in
𝐹𝑅 Rear axle’s braking force gauge
Biswas and Emadi (2019), while a more control-strategies oriented with
𝑓0 Rolling resistance coefficient
a specific focus on MPC can be found in Huang et al. (2017). In Lee
𝐹0𝐹 Maximum front axle’s brake without
et al. (2020) the authors present a comparative analysis between dy-
wheels’ slip in case only the front brake is
namic programming and reinforcement learning approaches for energy
engaged
management in hybrid powertrains, while, in Norouzi et al. (2021), a
survey on MPC methods, even if with focus restricted to ICEs control,
is presented.
An important aspect which in the literature is deeply investigated
feasible simulation times with more realistic and complete scenarios
and that is worth to be mentioned–in spite being less relevant with
for effectively improving the fuel economy (Leahey & Bauman, 2019).
respect to the scope of this paper–is the implementation of control al-
Many papers can be found that globally try to optimize the operations gorithms for real-time operations, sometimes claiming close-to-optimal
of the hybrid powertrain with a plenty of different approaches and performances but with very less computational effort (Baek & Chang,
mathematical frameworks. For example, optimal approaches with MPC 2019; Chen et al., 2019; Li & Görges, 2019; Rezaei et al., 2019; Wu
schemes and Mixed-Logic Dynamic (MLD) modeling can be found in et al., 2019; Xu & Shen, 2020; Yang et al., 2019; Zhang & Shen, 2016;
Lian et al. (2017); in Stroe et al. (2019) the authors apply the MPC Zhang et al., 2020; Zulkefli & Sun, 2019).

2
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

A well-known concept used in order to increase the fuel autonomy to that investigated in this paper, as in Eckert et al. (2019), where
of full electric, in general, or hybrid, in particular, vehicles is the a hybridization aftermarket kit for the conversion of conventional
regenerative braking (Ko et al., 2015; Li et al., 2018; Mutoh et al., vehicles with ICEs to hybrid architectures is presented. However the kit
2007; Nian et al., 2014; Wyczalek & Wang, 1992). The electric motor is very different from that used in the vehicle targeted by this paper.
is operated as an electric generator with the input being the vehicle’s The electric motors are coupled with the rear wheels with a V belt
kinetic energy and the output being the resulting electric energy that pulley while in our case two IWMs replace the rear wheels. Further, the
then is stored into suitable battery packs (Suntharalingam, 2011). The control problem is set up within a Genetic Algorithm framework meant
concept of regenerative braking seems to be very powerful indeed, both for finding an optimal sizing of the kit and an off-line strategy to
as in vehicles with thermal powertrains, the kinetic energy is mostly operate the two installed electric motors.
wasted as heat because of the friction achieved via the mechanical
braking system while the rest is dissipated through other mechanical
and aerodynamical losses with different contribution depending also 1.1. Target scenario specific challenges
on the speed. In full electric/hybrid vehicles, proper control algorithms
are meant to steer the kinetic energy to suitable batteries, by reversely
The first challenge is that the MPC strategy must be simple enough.
operating the electric powertrain. Several problems arise when regener-
Indeed, it has to serve as a baseline for an implementation onto the
ative braking is used and the literature presents many papers that try to
control platform of an aftermarket kit which cannot rely upon full ve-
address them. For instance, safety related aspects are of major concern,
among the others (Xu et al., 2016). Beyond safety, a major issue is that hicle information due to legal infringements that this would otherwise
regenerative braking cannot be used at any speed for different reasons. imply: in practice only the vehicle speed, the ICE speed, the gas pedal
At ‘‘low speeds’’, the regenerative effectiveness degrades: the electric and the IWMs speed are available. The second challenge is that the
motor is capable of applying a resistive torque to slow the vehicle regenerative braking provided by the MPC strategy via the additional
down. However, in this case, the efficiency of the power conversion IWMs installed on the vehicle’s rear axle sums to the mechanical
stage plays the relevant role since the back Electro-Magnetic Force braking provided by the mechanical braking system which, on that
(EMF) at the motor windings becomes too little and cannot be increased axle, is not removed1 : in practice, from the controller perspective, the
even by the boost converter that is usually placed in between the motor mechanical braking on the rear axle acts as a disturbance and prevents
and the battery. This prevents the power conversion stage to act as a the electric braking to achieve its full theoretical range. The third
voltage generator with respect to the battery which, then, provides a challenge is that the controller has to handle two conflicting objectives,
positive outward current that causes its further depletion (Fajri et al., the first being the maximization of the energy recovery during a brake,
2016). Thus, at low speeds it is important to find the cut-off point and the second being high efficiency operations since this would enable
where electric braking should be disengaged (Heydari et al., 2019). At costs saving both at design stage, because the designers can opt for
‘‘high speeds’’, the braking effectiveness is strongly affected because of less expensive components and reduce size and weights, and during
the electric motor limits which reflects in too low torques that further operations because the related components’ stress is also reduced.
decrease as the speed increase (Mohan et al., 2003).
At ‘‘intermediate speeds’’, motor efficiency is of concern (Chu et al.,
2015; Williamson et al., 2006; Zhang et al., 2018). Besides energy 1.2. Contributions and novelties
saving, operating the electric motor at high efficiencies implies low heat
dissipation, which decreases the motor components’ and the vehicle’s The main contributions and novelties of this paper to the existing
thermal stress in general, and therefore results in a reduction of the
literature and state-of-the-art are twofold:
overall costs. In order to include the motor efficiency when this is
relevant, such as in regenerative braking for hybrid/electric vehicle 1. this paper is an extension of Grandone et al. (2016) where a
applications, the motor efficiency maps can be considered which are regenerative braking control strategy for a real hybridized solar
usually given as contour plots in the 𝑇 − 𝜔 plane (Jung et al., 2020). vehicle was developed. However, that work does not consider
In this regards, researches focus on several aspects ranging from esti- the non-ideal IWMs efficiency and comfort aspects which, on the
mation (Mohammadi & Lowther, 2017), measurement (Dépature et al., contrary, are handled by the MPC scheme proposed in this paper;
2017), modeling (Mahmoudi et al., 2017), optimization (Jung et al., 2. the developed MPC strategy shows to be an appropriate base-
2020) and exploitation for control (So et al., 2020) and design (Kra-
line for the implementation onto the control platform of the
sopoulos et al., 2018; Yuan & Wang, 2012). The key point when
same real vehicle of Grandone et al. (2016). Future works will
high efficiency operations need to be achieved is that they may not
be directed toward the adaptation of the presented strategy
correspond to (𝑇 , 𝜔) pairs complying with the driving requirements
toward the implementation on an embedded system with lim-
(e.g., required propulsion/braking torque from the driver).
ited computation and memory resources, and with real-time
Regarding hybridized vehicles, they belong to the class of hybrid
vehicles with additional peculiarities which complicate further the requirements.
scenario. In this paper we refer to hybridized vehicles as those con- Apart from Grandone et al. (2016) and Eckert et al. (2019), no
ventional vehicles where the electric powertrain (e.g., IWMs) is added
further literature exists on the specific topic at the best of the authors’
aftermarket. One of the main problems when the implementation of
knowledge.
control strategies is of concern in hybridized vehicles is that they must
The rest of paper is organized as follows: Section 2 reports the
rely on the old engine control strategy that cannot be adapted to the
new architecture (Grandone et al., 2016). Further, in addition to the description of the target vehicle that is achieved after the hybridization
unavailability of useful data from the underlying vehicle’s system, the through the kit; Section 3 develops the constraints deriving by the
control strategies cannot even be actuated directly and they require the feasible operating region and those deriving by other considerations,
driver to close the loop somehow. that are required by the MPC for the computation of the optimal strat-
The scientific literature regarding conventional vehicles with pow- egy; Section 4 reports the aspects pertaining to the electric powertrain
ertrain hybridized aftermarket looks not very much developed. Some that affect the hybrid braking strategy and develops corresponding
authors investigated the dual scenario where a built-in Electric Vehicle constraints and cost functions; Section 5 develops the MPC scheme and
(EV) is equipped aftermarket with a small ICE to extend the driving problem which is the main focus of this paper; Section 6 reports the
range (Park et al., 2016). Others investigated a similar application simulation results, and Section 7 concludes the paper.

3
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Fig. 2. Hybridized architecture under investigation. Dashed lines indicate data flows,
solid lines indicate energy flows, blocks with blue edges identify the aftermarket kit’s
components.

Fig. 1. Hybridized FIAT Grande Punto.

Table 1
Fiat Grande Punto relevant data.
Feature Value
ICE Power 75 kW
Fuel Diesel
Coefficient of drag 0.325
Frontal area 2.05 m2
Rolling radius 0.295 m
Rolling resistance coeff. 0.02
Base vehicle mass 1105 kg

Fig. 3. Simplified block scheme of the vehicle power system for the hybridized
2. Description of the vehicle system architecture under investigation (Tiano, 2020).

In this section we present the main facts regarding the real vehicle
where the MPC policy is meant to be implemented. The hybridized vehicle established as described is meant for real-life
A FIAT Grande Punto, with features reported in Table 1, has been mobility. However, as it is also mentioned in the literature analysis,
hybridized by means of the solar-hybrid retrofit kit HySolarKit (Rizzo one main problem is that any possible control loop cannot be closed
et al., 2011), developed at the University of Salerno under the European without avoiding legal infringements. Further, for the same reason,
Horizon2020 LIFE-SAVE2 funded project, and constructed with the help limited amount of data from the underlying vehicle’s information and
of the other partners (see Fig. 1). It includes two IWMs, one Lithyum-
control system can be retrieved, basically only through the OBD gate:
Ion (Li-Ion) battery pack, flexible Photo-Voltaic (PV) panels, an Electric
only the vehicle speed, the ICE speed, the gas pedal and the IWMs speed
Node (EN) and a Vehicle Management Unit (VMU): the IWMs are
are available. The regenerative braking controller has to be designed
installed on the vehicle’s rear axle in place of the two standard wheels,
in order to cope with such limits of the ‘‘plant’’ where one major
the battery pack is placed into the trunk, the PV panels are installed
concern is that a human-being is expected to close the loop. This thorny
on the vehicle’s bonnet and roof and the VMU is placed onboard. The
battery pack can be recharged either by regenerative braking or by aspect has many implications ranging from safety to comfort. If, on the
the PV panels or by the electric grid. The VMU receives data from one hand, the driving must be kept stable and the hybridized vehicle
the battery, for the State-of-Charge (SoC) estimation, and from the On- must conform with all the required standards, on the other hand, the
Board Diagnosis (OBD) gate and accordingly drives the IWMs through regenerative braking has to be practically attainable by the ensemble
the EN, by computing the power split. A block scheme reporting the driver-controller. The driver presses the brake pedal in order to achieve
described hybridized architecture is shown in Fig. 2. Further equipment a desired deceleration given his/her prior knowledge of the combined
is also required in order to implement the support power system: a effect of the expected brake intensity and the vehicle dynamics. Over-
solar charge controller, a Battery Management System (BMS) and a looking that the brake signal is not available through the OBD gate
Hybrid Electric Vehicle (HEV) power controller. The simplified block and assuming that the braking condition can be estimated precisely
scheme of the power system is reported in Fig. 3 (Tiano, 2020). In somehow3 , the regenerative braking controller will then actuate the
practice, a Four Wheel Drive (4WD) powertrain and a Through-The- IWMs so as to provide an additional opposite (to the actual direction)
Road (TTR) architecture (Rizzo et al., 2020) is achieved, which is a mechanical torque. If the regenerative braking action is too strong, the
particular implementation of a parallel hybrid architecture where the driver might experience impractical or, even worse, unsafe brake. Such
electric motor(s) and the ICE are engaged on different axles thus not important aspect has then to be taken into account by the controller
requiring any particular system to couple the corresponding torques: that has to actuate properly the IWMs on the rear axle so that the
the coupling takes place exactly through the road which, basically, resulting total braking force is acceptable besides not endangering the
connects the rear and front axles via the tire–road contact and related driving safety.
friction.

1 3
This does not affect the preexisting braking system, thus not invalidating In fact, the braking condition is achieved by the inversion of the vehicle’s
the vehicle’s safety. dynamic equation at real-time, in the actual version of the kit: the road’s slope
2
Solar Aided Vehicle Electrification – LIFE16 ENV/IT/000442 – https: and vehicle’s mass are achieved via suitable sensors, such as the acceleration;
//www.life-save.eu/. therefore, if the traction force is negative, then the vehicle is braking.

4
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

3. Operating region and limits for the braking

Regarding mechanical braking, the target vehicle implements a

typical braking force distribution: the mechanical braking force on the
𝐹
front axle 𝐹𝐵,𝑀 𝑅
and that on the rear axle 𝐹𝐵,𝑀 are a fraction 𝛼𝑀 and
1 − 𝛼𝑀 , respectively, of the total mechanical braking force 𝐹𝐵,𝑀 . That
is,
𝐹
𝐹𝐵,𝑀 = 𝛼𝑀 𝐹𝐵,𝑀 , (1a)
𝑅
( )
𝐹𝐵,𝑀 = 1 − 𝛼𝑀 𝐹𝐵,𝑀 (1b)

such that the ratio

𝐹
𝐹𝐵,𝑀 𝛼𝑀
𝑅
= , (2)
𝐹𝐵,𝑀 1 − 𝛼𝑀

is achieved.
Since the two IWMs are installed on the rear axle, the total braking
𝑅
force of the hybridized vehicle on the rear axle 𝐹𝐵,𝐻 is
𝑅 𝑅
𝐹𝐵,𝐻 = 𝐹𝐵,𝑀 + 𝐹𝐵,𝐸 , (3)

where 𝐹𝐵,𝐸 is the electric brake component given by the IWMs, while
on the front axle
𝐹 𝐹
𝐹𝐵,𝐻 = 𝐹𝐵,𝑀 (4)

holds.
Similarly to (1), considering the combined mechanical-electric brak-
𝐹 𝐹 , Fig. 4. Operating region of the vehicle braking system in the braking force plane when
ing, the braking force on the front and on the rear axles 𝐹𝐵,𝐻 and 𝐹𝐵,𝐻 the tire–road grip coefficient is 𝜇1 (gray shaded area). The reader is referred to the
respectively, are a fraction 𝛼𝐻 and 1 − 𝛼𝐻 , respectively, of the total article’s body for the explanation of the items reported in the picture.
braking force 𝐹𝐵,𝐻 :
𝐹
𝐹𝐵,𝐻 = 𝛼𝐻 𝐹𝐵,𝐻 , (5a)
( ) constant braking force loci 𝐹̄1 and 𝐹̄2 and corresponding decelerations
𝑅
𝐹𝐵,𝐻 = 1 − 𝛼𝐻 𝐹𝐵,𝐻 . (5b) 𝑎̄1 and 𝑎̄2 , respectively; the point 𝑏1 identifies a particular mechanical
braking force combination on the corresponding locus for a given brake
Thus, the different ratio
pedal angle, the point 𝑏2 indicates the hybrid braking that is instead
𝐹
𝐹𝐵,𝐻 𝛼𝐻 achieved in the hybridized vehicle and the point 𝑏3 is the mechanical
= , (6)
𝑅
𝐹𝐵,𝐻 1 − 𝛼𝐻 braking force that the driver would experience on the factory vehicle,
instead.
is achieved.
In principle, a certain braking force corresponds to a point in the
The installation of two in-wheels motors on the rear axle in place
plane 𝐹 𝐹 − 𝐹 𝑅 of Fig. 4. However, in practice not all the points can
of the factory wheels, enables to change the braking force distribution
be achieved because of the tire–road grip limits related to the front
and intensity during the brake. Thus, it is important to identify what
and rear axles, which are reported as the black solid lines. They can
are the limits that the electric braking is subject to in order to ensure
be easily derived in terms of the maximum brake 𝐹0𝐹 without wheels’
feasible conditions. As we will show, actually an operating region can
slip on the front axle in case only the front brake is engaged, the
be identified.
maximum brake 𝐹0𝑅 without wheels’ slip on the rear axle in case
only the rear brake is engaged and the maximum brakes, 𝐹𝑝𝐹 and 𝐹𝑝𝑅 ,
3.1. Limits for the hybrid braking
without wheels’ slip on the front and rear axles, respectively, in case
both brakes are engaged. As it can be seen, the grip limits intersect
The limits for the braking of the hybridized vehicle can be sketched
exactly in (𝐹𝑝𝑅 , 𝐹𝑝𝐹 ) thus identifying a region that encloses all the front
in a two-dimensional plane by defining a suitable reference frame, as
and rear axles braking force combinations such that no tire–road slip
depicted in Fig. 4 (Guiggiani, 2014). The two axes 𝐹 𝐹 and 𝐹 𝑅 identify
happens. Usually, when such limits are hit, the ABS is correspondingly
the braking forces on the front and rear axles, respectively. The solid
engaged, which dynamically adapts the brake in order to avoid the
red line indicates the mechanical breaking force locus 𝐹𝐵,𝑀 ; the black
wheels to slip. However, in the hybridized vehicle it is clear that such
lines denote the grip limit on the front and rear axles and define an area
limit should not be reached because of any control action actuating
that encloses all the front and rear axles breaking force combinations
the electric braking, therefore resulting in an upper bound to the total
such that no wheel slip takes place/the Anti-lock Braking System (ABS)
braking force on the rear axle. Further, other additional limits exist
is not engaged during a brake when the tire–road grip coefficient is
such that a smaller area for the braking is actually identified.
𝜇1 ; 𝐹0𝐹 denotes the maximum brake without wheels’ slip on the front
In the factory vehicle, the mechanical braking force that can be
axle in case only the front brake is engaged when the tire–road grip
actuated spans the solid red line of Fig. 4, whose equation,
coefficient is 𝜇1 ; 𝐹0𝑅 denotes the maximum brake without wheels’ slip
on the rear axle in case only the rear brake is engaged when the 𝐹 𝛼𝑀
𝐹𝐵,𝑀 = 𝐹𝑅 , (7)
tire–road grip coefficient is 𝜇1 ; (𝐹𝑝𝑅 , 𝐹𝑝𝐹 ) denotes the maximum brakes 1 − 𝛼𝑀 𝐵,𝑀
without wheels’ slip on both axles in case both brakes are engaged derives easily from (2); the solid red line is also called mechanical
when the tire–road grip coefficient is 𝜇1 ; the dashed gray lines denote braking force locus. Regarding the hybrid braking, from (6) the similar
the grip limits on the front and rear axles when the tire–road grip equation
coefficient is 𝜇2 < 𝜇1 ; the solid blue line is the hybrid braking force 𝛼𝐻
𝐹
locus 𝐹𝐵,𝐻 for an acceptable ratio 𝛼̂ 𝐻 ; the dotted lines denote the two 𝐹𝐵,𝐻 = 𝐹𝑅 (8)
1 − 𝛼𝐻 𝐵,𝐻

5
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

easily derives. Since 𝛼𝐻 is decided in part by the controller, (8) iden- 3.3. Constraints for the hybrid brake feasibility
tifies a line that in principle could span the entire first quadrant.
However, a simple intuitive and mathematical explanation exists for
which the blue line lies within the mechanical braking force locus and In order to implement control strategies that take into account
the 𝐹 𝑅 axis: the electric braking affects only the rear axle, then the the operating region, the quantities 𝐹0𝐹 , 𝐹0𝑅 , 𝐹𝑝𝐹 and 𝐹𝑝𝑅 , have to be
braking force on the front axle in the hybridized vehicle is the same of computed for a given tire–road grip coefficient 𝜇. To this aim, the
the braking force on the same axle in the factory vehicle. Thus, from (7) reader can refer to (Guiggiani, 2014) for the details. However, for sake
𝐹
and (8) with the condition 𝐹𝐵,𝐻 𝐹 , it follows
= 𝐹𝐵,𝑀 of completeness, we briefly report them, where the relevant parameters
𝑅
𝐹𝐵,𝑀 are also highlighted in Fig. 5 for convenience.
𝛼𝐻 1 − 𝛼𝑀
𝑅
= . (9) In particular, for a vehicle of mass 𝑀𝑣 and being 𝑔 the gravity
𝐹𝐵,𝐻 1 − 𝛼𝐻 𝛼𝑀
acceleration, 𝐹0𝐹 , 𝐹0𝑅 , and 𝐹𝑝𝐹 , 𝐹𝑝𝑅 are determined as
𝑅 ∕𝐹 𝑅
Since 𝐹𝐵,𝑀 ≤ 1 (for obvious reasons and from (3)), (9) implies
𝐵,𝐻 ( ) 𝑏
that the slope 𝛼 ∕) 1 − 𝛼𝐻 of the blue line is not greater than the 𝐹0𝐹 =𝜇𝑀𝑣 𝑔 , (14a)
( 𝐻 𝑎 + 𝑏 − 𝜇ℎ
slope 𝛼𝑀 ∕ 1 − 𝛼𝑀 of the mechanical braking force locus. However, 𝑎
the slope of (8) is the result of a trade-off. From the one side, it cannot 𝐹0𝑅 =𝜇𝑀𝑣 𝑔 , (14b)
𝑎 + 𝑏 − 𝜇ℎ
be too small since small ranges of the mechanical braking force locus
would be mapped to too wide ranges of the hybrid braking force. Thus and
a driver would be prevented to ‘‘fine tune’’ the desired braking force 𝑏 + 𝜇ℎ
set-point through small increments of the brake pedal while at the 𝐹𝑝𝐹 =𝜇𝑀𝑣 𝑔 , (15a)
𝑎+𝑏
same time he/she would be allowed to reach very soon the tire–road 𝑎 + 𝜇ℎ
grip limit even for small brake pedal angles. On the other hand, small 𝐹𝑝𝑅 =𝜇𝑀𝑣 𝑔 , (15b)
𝑎+𝑏
slopes are desirable since they enable higher regeneration because they
imply a greater amount of electric braking. Assuming a reasonable 𝛼̂ 𝐻 respectively, such that, for a generic braking force with front and rear
components 𝐹gen𝐹 , 𝐹 𝑅 , respectively, the area enclosed within the grip
is determined, the solid blue line in Fig. 4 identifies the locus spanned gen
by the corresponding hybrid braking force and the gray shaded area is limits is identified by the inequalities
the operating region for the braking in the hybridized vehicle.
𝐹 𝑅
From the point of view of the driver’s experience during the hybrid 𝐹gen ≤𝑚1 𝐹gen + 𝑏1 , (16a)
braking, a few words can be spent. For convenience, in Fig. 4 the dotted 𝐹 𝑅
𝐹gen ≤ − 𝑚2 𝐹gen + 𝑏2 , (16b)
line highlights two constant braking force loci at 𝐹̄1 and 𝐹̄2 , that clearly
correspond to the constant decelerations 𝑎̄1 and 𝑎̄2 , respectively. In 𝐹
𝐹gen ≥0, (16c)
principle, when the driver presses the brake pedal, he/she means to
𝑅
set a desired deceleration. Assuming the driver wants to achieve 𝑎̄1 , in 𝐹gen ≤𝐹0𝑅 , (16d)
case of only mechanical braking, he/she would press the brake pedal at 𝑅
𝐹gen ≥0, (16e)
an angle corresponding to 𝑏1 . Because of the additional braking force
on the rear axle due to the electric braking, the point 𝑏2 is instead where
achieved, which in turn corresponds to the deceleration set-point 𝑎̄2 .
In other words, the driver achieves the deceleration that he/she would 𝐹𝑝𝐹 − 𝐹0𝐹
𝑚1 = , (17a)
get if he/she presses the brake pedal at an angle corresponding to 𝑏3 in 𝐹𝑝𝑅
the factory vehicle.
𝑏1 =𝐹0𝐹 , (17b)
3.2. Controlling the hybrid brake 𝐹𝑝𝐹
𝑚2 = , (17c)
The solid blue line can be described also in terms of the mechanical 𝐹0𝑅 − 𝐹𝑝𝑅
braking locus and the electric braking force, which can give a nice 1 + 𝐹𝑝𝑅
understanding of the driver experience. Then, by using (1b) and (5b) 𝑏2 = 𝐹𝑝𝐹 . (17d)
𝐹0𝑅 − 𝐹𝑝𝑅
in (3), we get
( ) ( ) Regarding the area enclosed within the cone determined by the
1 − 𝛼𝐻 𝐹𝐵,𝐻 = 1 − 𝛼𝑀 𝐹𝐵,𝑀 + 𝐹𝐵,𝐸 , (10)
mechanical braking force locus and by the hybrid braking force locus,
which clearly highlights the driver’s input through 𝐹𝐵,𝑀 = 𝑘𝑝 𝑃% , where it is identified by the inequalities
𝑃% is the brake pedal angle in percentage of the available range and
𝑘𝑝 is the slope of the pedal-to-mechanical brake linear characteristics 𝐹
𝐹gen ≤𝛼𝑀 𝐹𝐵,𝑀 , (18a)
that we assume instantaneous, and clearly highlights the control action
𝐹
through 𝐹𝐵,𝐸 . 𝐹gen ≥𝛼̂ 𝐻 𝐹𝐵,𝐻 , (18b)
Further, by using (1a) and (5a) in (8), it follows
and the operating region for the braking is therefore given by (16) and
𝛼𝐻 𝐹𝐵,𝐻 = 𝛼𝑀 𝐹𝐵,𝑀 , (11) (18), with (17).
which, plugged into (10), gives In practice, the hybrid brake 𝐹𝐵,𝐻 will be constrained to stay within
𝛼𝑀 the operating region by enforcing (16) and (18) (with (17)) with
𝛼𝐻 = . (12) 𝐹 =𝛼 𝐹 𝑅
( )
𝐹 conditions 𝐹gen 𝐻 𝐵,𝐻 and 𝐹gen = 1 − 𝛼𝐻 𝐹𝐵,𝐻 , and by taking into
1 + 𝐹 𝐵,𝐸
𝐵,𝑀
account (10), (12) and (13), thus resulting in the constraints
Eq. (12) allows two useful considerations. The first is that 𝛼𝐻 ( )
depends on 𝐹𝐵,𝐸 once the driver’s brake input 𝐹𝐵,𝑀 is provided, and 𝛼𝐻 (1 + 𝑚1 ) − 𝑚1 𝐹𝐵,𝐻 ≤𝑏1 , (19a)
the second is that ( )
𝛼𝐻 (1 − 𝑚2 ) + 𝑚2 𝐹𝐵,𝐻 ≤𝑏2 , (19b)
𝛼𝐻 ≤ 𝛼𝑀 , (13) ( )
1 − 𝛼𝐻 𝐹𝐵,𝐻 ≤𝐹0𝑅 , (19c)
since 𝐹𝐵,𝐸 , 𝐹𝐵,𝑀 ≥ 0.
𝐹𝐵,𝐻 ≥0, (19d)
Eq. (10), (12) and (13) will have to be considered by any control
algorithm for implementing a feasible braking strategy. 𝛼𝐻 ∈[𝛼̂ 𝐻 , 𝛼𝑀 ]. (19e)

6
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

While this choice is a reasonable trade-off with the need of limit-

ing the computational effort required to run the algorithm, future
investigations will be directed toward the inclusion also of the effi-
ciency of the whole conversion stage in order to improve the achieved
outcomes.

4.1. IWMs Efficiency

The IWMs efficiency can be modeled as

[ ( ) ( ) ( ) ]
max 𝜔 − 𝜔me 2 𝑇 − 𝑇me 2 𝑃 − 𝑃me 2
Fig. 5. Vehicle’s chassis main parameters for braking limits computation. Also the 𝜂IWM = 𝜂IWM 1 − 𝑘1 − 𝑘 2 − 𝑘 3 ,
relevant forces are highlighted. CG stands for Center of Gravity. 𝜔max 𝑇 max 𝑃 max
(21)

where 𝜔, 𝑇 and 𝑃 = 𝑇 𝜔 are the IWMs speed, torque and power, respec-
max is the IWMs achievable maximum efficiency; 𝜔max , 𝑇 max ,
tively; 𝜂IWM
𝑃 max are the IWMs maximum speed, torque and power, respectively;
𝑇me , 𝜔me , 𝑃me are the IWMs speed, torque and power at the maximum
max , respectively; and 𝑘 , 𝑘 and 𝑘 are three weights that
efficiency 𝜂IWM 1 2 3
depend upon the IWMs geometry, size and other physical parameters
according to which the power losses distribute across the corresponding
terms in (21) (Mahmoudi et al., 2017).

4.2. IWMs Limitations

In addition to the (obviously) non unitary and operating conditions

dependent efficiency, the IWMs have additional limitations in the
maximum torque and power they can provide and the maximum speed
they can achieve. These limitations will be considered through the
Fig. 6. Typical torque-speed (𝑇 , 𝜔) plots w.r.t. corresponding nominal values 𝑇𝑛 and
constraints
𝜔𝑛 , and some efficiency contours 𝜂0 > ⋯ > 𝜂𝑖 , for an electric motor.
𝑟𝑤 𝐹𝐵,𝐸 ≤𝑇 max , (22a)
𝑟𝑤 𝐹𝐵,𝐸 𝜔 ≤𝑃 max , (22b)
3.4. Comfort constraints for the electric brake 𝜔 ≤𝜔max , (22c)

where 𝑟𝑤 is the wheel’s radius, in the proposed MPC algorithm so

The electric brake adds up to the mechanical brake such that in
as to achieve feasible operations. In practice, (22a) is the maximum
the hybridized vehicle a new front-axle/rear-axle brake distribution 𝛼̂ 𝐻
torque limit that the IWMs are subject to for speeds within the constant-
is achieved. In principle, in case only the maximization of the energy
torque (e.g., below 𝜔∕𝜔𝑛 = 1 in Fig. 6) region while (22b) is the
recovery is of concern without further limitations, any optimal strategy
maximum power limit that the IWMs are subject to for speeds within
should always choose the maximum regeneration possible which does
the constant-power (e.g., above 𝜔∕𝜔𝑛 = 1 in Fig. 6) region. The stricter
not violate the feasible braking region limits. However, in case of the
constraint between (22a) and (22b) will be automatically active since
target application, the driver is required to close the loop, that is, to
it will depend on the wheels’ speed 𝜔.
adapt the brake according to the control actuation. If, on the one hand,
𝛼̂ 𝐻 is appropriately set such that the driving is not endangered and any
possible risky lack of sensitiveness is avoided, the driver’s comfort can 5. Optimal braking modulation for regeneration
be affected. For this reason, the constraint
The optimal braking modulation for regeneration is pursued by
| |
|𝐹𝐵,𝐸 (𝑡 + 𝑇𝑠 ) − 𝐹𝐵,𝐸 (𝑡)| ≤ 𝛥𝐹̄𝐵,𝐸 , (20) means of an MPC scheme which takes into account the limits provided
| |
by the operating region of Fig. 4 and those pertaining the IWMs,
where 𝑇𝑠 > 0 is the sampling time and 𝛥𝐹̄𝐵,𝐸 is the desired bound on
along with the corresponding efficiency. Additionally, the developed
𝛥𝐹𝐵,𝐸 , will be also considered.
algorithm includes also limitations on the electric force variation such
that possible requirements on the smoothness of the brake are also taken
4. Electric powertrain for efficient operations and regenerative
into account.
braking
The considered vehicle’s dynamics are the longitudinal, given by
( )
The electric powertrain of the hybridized vehicle consists of two 1 1
̇ =−
𝑣(𝑡) 𝑐𝑥 𝐴𝜌𝑣2 (𝑡) + 𝑓0 𝑀𝑣 𝑔 cos 𝜎 + 𝑀𝑣 𝑔 sin 𝜎 + 𝐹𝐵,𝐻 (𝑡) , (23)
IWMs installed on the vehicle’s rear axle and is connected to the 𝑀𝑒 2
onboard battery through the EN and a power conversion stage, as with initial conditions 𝑣(0) = 𝑣0 and where 𝑣 is the vehicle’s speed,
depicted in Fig. 2. The EN with the power conversion stage enables 𝑀𝑒 is the equivalent vehicle’s mass which takes into account for the
bidirectional power flows from the battery to the IWMs and back, and vehicle’s extra-inertia caused by the rotational masses, 𝑐𝑥 is the drag
is controlled by the VMU. Fig. 6 depicts the well-known, typical torque- coefficient, 𝐴 is the vehicle’s front area which contributes to the aerody-
speed characteristics of an electric motor along with the level curves of namic losses, 𝜌 is the air density, 𝑓0 is the rolling resistance coefficient,
the related efficiency. 𝑀𝑣 is the vehicle’s mass, 𝑔 is the gravity acceleration, 𝜎 is the road’s
In this paper only the IWMs total efficiency is included since it slope that it is considered constant across all the maneuver and 𝐹𝐵,𝐻 (𝑡)
majorly affects the overall performances of the proposed MPC strategy. is the hybrid brake.

7
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

During a brake, the energy that can be recovered to the battery is 5.1.1. Costs for the MPC scheme
𝑡 𝑡 The costs composing the objective function are the discrete-time
𝐸bat = 𝜂IWM (𝜏)𝑇 (𝜏) 𝜔(𝜏) d𝜏 = 𝜂IWM (𝜏)𝐹𝐵,𝐸 (𝜏) 𝑣(𝜏) d𝜏, (24) version of (24) and (25), which at discrete-time 𝑚 and across the
∫0 ∫0
horizon 𝑀ℎ rewrite respectively as
since 𝑇 (𝑡) = 𝑟𝑤 𝐹𝐵,𝐸 (𝑡) and 𝑟𝑤 𝜔(𝑡) = 𝑣(𝑡), and where 𝜂IWM (𝑡) is the IWM’s
( )⊤ ( 𝑀ℎ )⊤ ( 𝑀ℎ ) 𝑀ℎ
efficiency given by (21) and computed across a curve 𝜔(𝑡) 𝑇 (𝑡) 𝑡 , 𝐸bat (𝑚) = 𝐅𝐵,𝐸,𝑚 diag 𝜼IWM,𝑚 𝐯𝑚 𝑇𝑠 , (28)
and 𝐹𝐵,𝐸 (𝑡) is the electric brake force. Eq. (24) will be used as a cost in
the MPC algorithm, however a penalty term is also introduced such that and
{[
the optimizer is pushed to find a trade-off with possible high efficiency ( 𝑀ℎ ) ( 𝑀ℎ )]⊤
operations for the IWMs. High efficiency operations can be desirable 𝐽IWM (𝑚) = 𝜼IWM,𝑚 − 𝟏𝑀ℎ 𝜂̄IWM ⊙ 𝜼IWM,𝑚 − 𝟏𝑀ℎ 𝜂̄IWM
in order to achieve operating conditions with smaller stress, which can
[( ]} 21
be beneficial in terms of an increased reliability of the components. 𝑀ℎ 𝑀 ) ( 𝑀ℎ ) ( 𝑀ℎ 𝑀ℎ ) ( 𝑀ℎ 𝑀ℎ )
𝝎𝑚+1 − 𝝎𝑚 ℎ ⊙ 𝝎𝑀
𝑚+1
− 𝝎 𝑚 + 𝐓𝑚+1
− 𝐓𝑚 ⊙ 𝐓𝑚+1
− 𝐓𝑚 ,
Of course, the trade-off implies the energy recovered to the battery
to be smaller when the penalty is used. This concept will be further (29)
investigated in future works where also the efficiency of the power
conversion stage and of the battery operations will be included. where
In order to introduce the objective of attaining higher efficiency [ ( 𝝎𝑀ℎ − 𝟏 𝜔 ) ( 𝝎𝑀ℎ − 𝟏 𝜔 )
𝑀ℎ max 𝑚 𝑀ℎ me 𝑚 𝑀ℎ me
operations, we consider a penalty term given by the line integral of 𝜼IWM,𝑚 = 𝜂IWM 𝟏𝑀ℎ − 𝑘1 ⊙
𝜔max 𝜔max
the scalar field identified by (21) centered at the desired 𝜂̄IWM along a
( )⊤ ( 𝐓𝑀ℎ − 𝟏 𝑇 ) ( 𝐓𝑀ℎ − 𝟏 𝑇 )
piecewise smooth curve (𝑡) = 𝜆𝜔 𝜔(𝑡) 𝜆𝑇 𝑇 (𝑡) 𝑡 : 𝑚 𝑀ℎ me 𝑚 𝑀ℎ me
− 𝑘2 ⊙
𝑇 max 𝑇 max
| |
𝑡
| ( ) | || ||
𝐽IWM = |𝜂IWM − 𝜂̄IWM | d𝑠 = |𝜂 𝑠(𝜏) − 𝜂̄IWM | ||𝑠(𝜏) ̇ || d𝜏 ( 𝐏𝑀ℎ − 𝟏 𝑃 ) ( 𝐏𝑀ℎ − 𝟏 𝑃 )]
∫ | | ∫0 | IWM | || || 𝑚 𝑀ℎ me 𝑚 𝑀ℎ me
− 𝑘3 ⊙ , (30)
𝑡 ( ) ( )1 𝑃 max 𝑃 max
| | 2
= |𝜂 𝜔(𝜏), 𝑇 (𝜏) − 𝜂̄IWM | 𝜆2𝜔 𝜔̇ 2 (𝜏) + 𝜆2𝑇 𝑇̇ 2 (𝜏) d𝜏, (25)
∫0 | IWM | is obtained similarly by discretizing (21), and where 𝟏𝑀ℎ is a column
where | | denotes the absolute value, ‖ ‖ denotes the standard vector vector with unit entries of dimension 𝑀ℎ .
norm, 𝑠(𝜏) is in principle an arbitrary parametrization of (𝜏) that we The controller will aim at maximizing (28) and minimizing (30).
have chosen to be Cartesian and 𝜆𝜔 , 𝜆𝑇 are two normalization constants
that, in what follows, will be assumed unitary.
5.1.2. Dynamic constraints for the MPC scheme
5.1. MPC scheme for optimal braking modulation The vehicle’s dynamic constraint is the discrete-time version of (23),
thus following
The MPC algorithm is designed at discrete time 𝑚 with sampling (
𝑀ℎ 𝑀 𝑇 1 𝑀 𝑀
time 𝑇𝑠 , such that the continuous time 𝑡 can be easily recovered by 𝐯𝑚+1 = 𝐯𝑚 ℎ − 𝑠 𝑐 𝐴𝜌 𝐯𝑚 ℎ ⊙ 𝐯𝑚 ℎ + 𝟏𝑀ℎ 𝑓0 𝑀𝑣 𝑔 cos 𝜎
𝑡 = 𝑚𝑇𝑠 . The scenario under investigation considers the vehicle moving 𝑀𝑒 2 𝑥
)
at constant speed 𝑣0 > 0 with the brake maneuver starting at 𝑚 = 0. 𝑀ℎ
+ 𝟏𝑀ℎ 𝑀𝑣 𝑔 sin 𝜎 + 𝐅𝐵,𝐻,𝑚 . (31)
However, it does not ends when the vehicle stops, rather when the
brake signal is off or, in principle, when other conditions happen. In
this paper, the brake pedal signal is considered as a condition for the 5.1.3. Operating region constraints for the MPC scheme
controller to be engaged/disengaged: when the driver starts the brake
The link between (31) and (28) is achieved by recasting (10) in
maneuver, he/she presses the brake pedal and basically set a decelera-
vector form as
tion set-point. The electro-hydraulic actuation of the mechanical brake
can be considered not slower than that of the electric braking with ( 𝑀ℎ ) 𝑀ℎ ( ) 𝑀ℎ
𝟏𝑀ℎ − 𝜶 𝐻,𝑚 ⊙ 𝐅𝐵,𝐻,𝑚 = 𝟏𝑀ℎ 1 − 𝛼𝑀 𝐹𝐵,𝑀 (𝑚) + 𝐅𝐵,𝐸,𝑚 , (32)
control loops, therefore on short time-scales the deceleration set-point
is achieved without regenerative braking, in practice. From the driver’s 𝑀ℎ
where 𝜶 𝐻,𝑚 can be retrieved by means of (12), i.e., as
perspective this is an unwanted deceleration that adds to the set-point
he/she had in mind at the beginning of the maneuver. Other trigger ( 𝑀ℎ
𝐅𝐵,𝐸,𝑚 )
𝑀ℎ
conditions, such as, e.g., the cut-off speed, are neglected because they 𝜶 𝐻,𝑚 ⊙ 𝟏𝑀ℎ + = 𝟏𝑀ℎ 𝛼𝑀 . (33)
𝐹𝐵,𝑀
are assumed to be handled by external algorithms that are out of the
scope of the present paper. The constraints regarding the braking operating region depicted in
Let us now define the vectors Fig. 4 can be also included by recasting (19) at discrete time and then
𝑀ℎ [ ] in vector form as
𝐅𝐵,𝐸,𝑚 = 𝐹𝐵,𝐸 (𝑚) 𝐹𝐵,𝐸 (𝑚 + 𝑀ℎ − 1) ⊤ , (26a)
𝑀ℎ [ ]⊤ ( 𝑀ℎ ) 𝑀ℎ
𝐅𝐵,𝐻,𝑚 = 𝐹𝐵,𝐻 (𝑚) 𝐹𝐵,𝐻 (𝑚 + 𝑀ℎ − 1) , (26b) 𝜶 𝐻,𝑚 (1 + 𝑚1 ) − 𝟏𝑀ℎ 𝑚1 ⊙ 𝐅𝐵,𝐻,𝑚 ≤𝑏1 , (34a)
𝑀ℎ [ ]⊤ ( 𝑀ℎ ) 𝑀ℎ
𝜶 𝐻,𝑚 = 𝛼𝐻 (𝑚) 𝛼𝐻 (𝑚 + 𝑀ℎ − 1) , (26c) 𝜶 𝐻,𝑚 (1 − 𝑚2 ) + 𝟏𝑀ℎ 𝑚2 ⊙ 𝐅𝐵,𝐻,𝑚 ≤𝑏2 , (34b)
𝑀ℎ [ ]⊤ ( 𝑀ℎ ) 𝑀ℎ 𝑅
𝐯𝑚 = 𝑣(𝑚) 𝑣(𝑚 + 𝑀ℎ − 1) , (26d) 𝟏𝑀ℎ − 𝜶 𝐻,𝑚 ⊙ 𝐅𝐵,𝐻,𝑚 ≤𝐹0 , (34c)
𝑀
ℎ
at a given time-step 𝑚 and across a prediction horizon 𝑀ℎ ahead in 𝐅𝐵,𝐻,𝑚 ≥0, (34d)
the future. In what follows we will also use 𝑀ℎ
𝜶 𝐻,𝑚 ∈[𝛼̂ 𝐻 , 𝛼𝑀 ], (34e)
𝑀 ℎ 𝑀
𝑟𝑤 𝐅𝐵,𝐸,𝑚 =𝐓𝑚 ℎ , (27a)
along with conditions (17).
𝑀 𝑀
𝐯𝑚 ℎ =𝑟𝑤 𝝎𝑚 ℎ , (27a) In (32), it is important to remark that by the assumptions regarding
𝑀 𝑀 𝑀 the time-scales of the mechanical braking system dynamics and those
𝐏𝑚 ℎ =𝐓𝑚 ℎ ⊙ 𝝎𝑚 ℎ
ℎ𝑀 𝑀 of the electric braking system, at each MPC iteration and across all the
=𝐅𝐵,𝐸,𝑚 ⊙ 𝐯𝑚 ℎ , (27c)
horizon 𝑀ℎ , the mechanical braking is considered constant to its initial
where ⊙ is the Hadamard (element-wise) product. value 𝐹𝐵,𝑀 (𝑚) at time-step 𝑚.

8
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Fig. 7. Block diagram of the proposed MPC scheme (reported within the gray shaded
area) interacting with the simulated target vehicle. The actual time-step is 𝑚, the
sampling time is 𝑇𝑠 and the prediction horizon is 𝑀ℎ . The vehicle’s dynamic model is
the same as that of the MPC scheme, i.e., no model-mismatch is assumed.

5.1.4. Comfort constraints for the MPC scheme Fig. 8. Possible onboard logic in charge of deciding whether to regenerate, to compute
the power split or not to do anything.
The comfort constraints (20) for the MPC scheme are easily recast
as
( 𝑀ℎ 𝑀ℎ )| |
𝐅𝐵,𝐸,𝑚+1 − 𝐅𝐵,𝐸,𝑚 ≤ 𝟏𝑀ℎ 𝛥𝐹̄𝐵,𝐸 , (35) • no model-mismatch between that used as a constraint in the op-
where the superscript | | indicates the element-wise absolute value. timizer and that used for simulating the vehicle dynamics under
the control action;
5.1.5. IWM limits constraints for the MPC scheme • known driver’s model and perfect knowledge of the required
The constraints (22) deriving from the IWMs limits rewrite straight- mechanical braking force 𝐹𝐵,𝑀 (𝑚);
forwardly as • known vehicle and environment’s (e.g., 𝜇 = 0.7) parameters;
𝑀 • no impact of the battery and power conversion stage limitations
ℎ
𝑟𝑤 𝐅𝐵,𝐸,𝑚 ≤𝟏𝑀ℎ 𝑇 max , (36a)
(e.g., no cut-off speed).
𝑀ℎ 𝑀 max
𝑟𝑤 𝐅𝐵,𝐸,𝑚 ⊙ 𝝎𝑚 ℎ ≤𝟏𝑀ℎ 𝑃 , (36b)
However, in a realistic application, the grip coefficient will have to
𝑀 max be estimated through appropriate algorithms at real-time, and the
𝝎𝑚 ℎ ≤𝟏𝑀ℎ 𝜔 . (36c)
limitations of the battery and the power conversion stage should be
5.1.6. MPC algorithm appropriately considered, while parameters such as, e.g., 𝑎, 𝑏, ℎ, 𝑀𝑣 ,
In conclusion, the MPC controller computes 𝑀𝑒 , and 𝜎, could be reasonably derived either by the vehicle’s factory
specifications or by appropriate sensors.
( 𝑀ℎ )∗ The algorithm is implemented in Julia (The Julia Project, 2021a)
𝐅𝐵,𝐸,𝑚 = arg max 𝐸bat (𝑚) − 𝑤IWM 𝐽IWM (𝑚)
𝑀ℎ
with JuMP, a modeling language and collection of supporting packages
𝐅𝐵,𝐸,𝑚
for mathematical optimization in Julia, (The Julia Project, 2021b)
subject to Eq. (31), Eq. (34)-Eq. (36) (37) and Ipopt (COIN-OR, 2021; Wächter & Biegler, 2006) and runs on an
𝑀ℎ
𝐅𝐵,𝐸,𝑚 ≥0 2.6 GHz Intel Core i7 with 16 GB RAM. The choice of open-source scien-
𝑀
𝒗𝑚 ℎ ≥ 0, tific computing languages, software packages and optimizers complies
with the recommendations of the EU regarding open-science.
with conditions (17), (27), (30), (32) and (33), and where 𝑤IWM > 0 is a
In case of a possible implementation of the scheme onto the kit,
suitable weight to be properly tuned according to the desired blending
the required typical execution times will be of the order of hundred
of the two objectives.
of milliseconds, as also reported in Rizzo et al. (2021). Of course,
The problem (37) is solved at each time-step 𝑚 and provides 𝑀ℎ
the scheme will need to be implemented by taking into account all
optimal inputs to the hybrid brake system. However, only the first
∗ (𝑚) is implemented while the others are discarded. Then, the technological aspects implied by and the limitations of the kit’s
element 𝐹𝐵,𝐸
embedded control platform.
new measurements from the vehicle are obtained and the optimization
is iterated at time-step 𝑚 + 1, thus implementing a feedback policy. For the vehicle of Fig. 1 with 𝑎, 𝑏 and ℎ as per Table 2, 𝐹0𝐹 , 𝐹𝑝𝐹 , 𝐹0𝑅 ,
𝑅
𝐹𝑝 , 𝑚1 , 𝑏1 , 𝑚2 and 𝑏2 are reported in Table 3. Across all scenarios the
Fig. 7 reports a block diagram of the proposed MPC scheme inter-
acting with the target vehicle. An appropriate logic will coordinate the sampling time is 𝑇𝑠 = 50 ms, with typical execution times not greater
operations of a possible implementation onto the aftermarket kit with than about 8 ms for a single MPC iteration, and the brake maneuver is
the VMU’s, which in the current release is in charge of computing the
𝑚 𝑚
power split during traction, according to the flow-chart in Fig. 8. The max
𝐹𝐵,𝑀 (𝑚) = 𝐹𝐵,𝑀 min − 𝜆
(1 − 𝑒− 𝜆 ) + 𝐹𝐵,𝑀 𝑒 , (38)
power split is computed via an ECM strategy, by minimizing a suitable
where max ,
𝐹𝐵,𝑀 min
𝐹𝐵,𝑀and 𝜆 will be defined according to the particular
cost function at real-time. More on this can be found in Tiano (2020).
scenario under consideration.
6. Simulation results and comparisons The simulation scenarios consider problems (37) where the different
aspects presented in this paper (comfort, IWMs limitations and effi-
In this section we present the simulations carried out by running ciency) are introduced once at a time so as to show their implications
the problem (37) in different scenarios with parameters summarized in on the regeneration operations and to highlight the effectiveness of
Table 2. There, the IWM specs are for a single unit while, for plotting the proposed scheme. Also, the simulations are achieved in various
the simulation outcomes, the joint operations of the two units will be conditions in order to highlight how these different aspects are affected.
considered. Further, all the simulations are carried out by assuming: Table 4 summarizes the various scenarios.

9
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Table 2 Table 4
(
Vehicle’s data and environment’s parameters for simulations and comparison. ∗ (2𝜋∕ Scenarios summary.
)
60)𝑟𝑤 𝜔rpm = 5∕18 𝑣km−1 . Scenario no. Conditions Value
Feature Item Value 1
Vehicle’s Body Maneuver Duration 3s
Frontal area 𝐴 2.05 m2 Prediction horizon 𝑀ℎ 0.5 s
Wheels radius 𝑟𝑤 0.295 m Initial vehicle’s speed 𝑣(0) 100 kmh−1
Base vehicle mass 𝑀𝑣 1105 kg Mechanical Brake Time Constant 𝜆 5 (250 ms)
max
Equivalent vehicle mass 𝑀𝑒 1270 kg Maximum Mechanical Brake 𝐹𝐵,𝑀 8 kN
min
Mechanical brake distribution 𝛼𝑀 0.65 Minimum Mechanical Brake 𝐹𝐵,𝑀 1 kN
Hybrid brake distribution limit 𝛼̂ 𝐻 0.35 Constraints Considered (31), (34), (35)
Center of gravity (CG) height ℎ 0.5 m Constraints Neglected (36)
max
Front axle distance from CG 𝑎 1.13 m IWMs Efficiency 𝜂IWM 𝜂IWM
Rear axle distance from CG 𝑏 1.38 m Penalty Term Weight 𝑤IWM 0
Electric Brake Bound 𝛥𝐹̄𝐵,𝐸 {100, 300, 500, ∞} N
IWM’s Specs
max 2
Max efficiency 𝜂IWM 0.9
Max speed 𝜔max 1500 rpm∗ Maneuver Duration 3s
Max torque 𝑇 max 381.97 N m Prediction horizon 𝑀ℎ 0.5 s
Max power 𝑃 max 10 kW Initial vehicle’s speed 𝑣(0) 100 kmh−1
Speed at max efficiency 𝜔max
me
750 rpm∗ Mechanical Brake Time Constant 𝜆 5 (250 ms)
max max
Torque at max efficiency 𝑇me 230 N m Maximum Mechanical Brake 𝐹𝐵,𝑀 8 kN
Power at max efficiency max
𝑃me 9 kW min
Minimum Mechanical Brake 𝐹𝐵,𝑀 1 kN
Speed related loss coefficient 𝑘1 0.5 Constraints Considered (31), (34), (36)
Torque related loss coefficient 𝑘2 0.5 Constraints Neglected (35)
Power related loss coefficient 𝑘3 0.2 IWMs Efficiency 𝜂IWM (30)
Environment’s Params Desired IWMs Efficiency 𝜂̄IWM 0.9
Penalty Term Weight 𝑤IWM {0, 1000, 10000}
Rolling resistance coefficient 𝑓0 0.02 Electric Brake Bound 𝛥𝐹̄𝐵,𝐸 ∞
Coefficient of drag 𝑐𝑥 0.325
Grip coefficient (dry road) 𝜇 0.7 3
Air density 𝜌 1.2 kg∕m3 Maneuver Duration 1s
Gravity acceleration 𝑔 9.8 m∕s2 Prediction horizon 𝑀ℎ 0.5 s
Initial vehicle’s speed 𝑣(0) 50 kmh−1
Mechanical Brake Time Constant 𝜆 5 (250 ms)
Table 3 max
Maximum Mechanical Brake 𝐹𝐵,𝑀 8 kN
Parameters for the determination of the grip limits for the target vehicle. Minimum Mechanical Brake 𝐹𝐵,𝑀 min
1 kN
𝐹0𝐹 𝐹𝑝𝐹 𝐹0𝑅 𝐹𝑝𝑅 Constraints Considered (31), (34), (36)
Constraints Neglected (35)
8000 N 8750 N 4500 N 3250 N
IWMs Efficiency 𝜂IWM (30)
Desired IWMs Efficiency 𝜂̄IWM 0.9
𝑚1 𝑏1 𝑚2 𝑏2 Penalty Term Weight 𝑤IWM {0, 1000, 10000}
1270 N 0.23 N 8000 N 31 500 N Electric Brake Bound 𝛥𝐹̄𝐵,𝐸 ∞
4
Maneuver Duration 3s
Prediction horizon 𝑀ℎ 0.5 s
6.1. (Scenario 1) regeneration with comfort constraints, constant efficiency Initial vehicle’s speed 𝑣(0) {75, 125} kmh−1
and no IWM limitations Mechanical Brake Time Constant 𝜆 10 (500 ms)
max
Maximum Mechanical Brake 𝐹𝐵,𝑀 5 kN
min
Minimum Mechanical Brake 𝐹𝐵,𝑀 500 N
In the first scenario, the problem (37) is solved by neglecting the
Constraints Considered all
IWMs limitations (36) and by considering maximum IWMs efficiency Constraints Neglected none
max , as also 𝑤
operations, i.e. 𝜂IWM = 𝜂IWM IWM = 0 is set so as to show how IWMs Efficiency 𝜂IWM (30)
regeneration is affected by the introduction of the comfort constraints. Desired IWMs Efficiency 𝜂̄IWM 0.9
Fig. 9 shows the resulting optimal hybrid brake 𝐹𝐵,𝐻 ∗ for 𝛥𝐹̄𝐵,𝐸 = Penalty Term Weight 𝑤IWM 1000
{100, 300, 500, ∞} N, respectively indicated by the magenta triangles, Electric Brake Bound 𝛥𝐹̄𝐵,𝐸 {100, ∞} N
the cyan squares, the black pentagons and the blue circles. Instead,
the red diamonds indicate the mechanical brake input. As expected, in
case of no bounds on the maximum variation of the electric brake, the ∗
Fig. 10 shows the resulting optimal hybrid brake 𝐹𝐵,𝐻 for 𝑤IWM =
MPC controller chooses the maximum regeneration possible without
{0, 1000, 10000}, respectively indicated by the magenta triangles, the
violating the constraints on the minimum slope of the hybrid brake
cyan squares and the blue circles, while the red diamonds indicate the
and on the grip limits4 . For decreasing bounds, the optimal strategy
mechanical brake input. Basically, there is no substantial difference
selects hybrid brakes such that the trajectories tend to overlap with the
mechanical brake when no regeneration is implied. among the corresponding optimal hybrid brakes and remarkably the
hybrid brakes do not achieve the hybrid brake locus at 𝛼̂ 𝑀 even when
6.2. (Scenario 2) regeneration with no comfort constraints, non-constant 𝑤IWM = 0, that is when the MPC algorithm just accounts for non-ideal
efficiency and IWM limitations efficiency operations due to (30).
A better insight can be obtained by looking at the torque-speed
In the second scenario, the problem (37) is solved by considering no trajectories in the 𝜔 − 𝑇 plane depicted in Fig. 11, where also the
bounds on the maximum electric brake variation while including the efficiency contours are plotted. In practice, at high speeds (i.e., at
IWMs non-ideal efficiency (30) and limitations (36) for several 𝑤IWM . the beginning of the brake) the IWMs maximum power limits the
achievable electric brake intensity while, when the vehicle slows down,
4
Of course, in a possible implementation onto the real kit, a small tolerance the grip limits do not allow to increase it. Fig. 11 also shows how
would be used in order for the optimal strategy not to lie exactly onto the grip the penalty term affects the operations in term of efficiency according
limits, that in a real application would engage the ABS. to the particular choice for 𝑤IWM . In particular, increasing the weight

10
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Fig. 9. (Scenario 1) Optimal hybrid brake trajectories in the braking operating region
for several bounds on the maximum electric brake variation. The blue circles identify Fig. 10. (Scenario 2) Optimal hybrid brake trajectories in the braking operating
the trajectory when no bound is enforced; the black pentagons identify the trajectory region for several weights of the IWMs penalty term. The blue circles identify the
when the max variation enforced is 500 N; the cyan squares identify the trajectory when trajectory when no penalty is introduced and the MPC algorithm just accounts for
the max variation enforced is 300 N; the magenta triangles identify the trajectory when non-ideal efficiency operations due to (30); the cyan squares identify the trajectory
the max variation enforced is 100 N; the red diamonds identify the mechanical brake when 𝑤IWM = 1000; the magenta triangles identify the trajectory when 𝑤IWM = 10000;
input. the red diamonds identify the mechanical brake input. No substantial differences are
noticed.

forces the controller to choose opposing torques such that the trajectory
majorly lies within higher efficiency regions. However, this comes at a
price, as it can be easily deduced. The energy recovered to the battery
decreases as higher efficiency operations are attained. For instance, in
case 𝑤IWM = 0, the recovered energy amounts at 11.84 W h, in case
𝑤IWM = 1000, the recovered energy amounts at 11.69 W h and, in case
𝑤IWM = 10000, the recovered energy amounts at 11.11 W h. Surprisingly,
the small differences among the recovered energies, suggest that in
the long run the benefits of higher efficiency operations might be
appealing since they imply a negligible impact on the driving range that
can be achieved thanks to the regeneration. Further, Fig. 11 suggests
that for lower initial speeds, the limitations due to the IWMs features
can enable higher regenerations, however at the price of lower effi-
ciency operations, as it can be argued by the position of the efficiency
contours.

6.3. (Scenario 3) regeneration with no comfort constraints, non-constant

efficiency and IWM limitations, at lower speeds

The third scenario differs from the second in the initial vehicle’s
speed and the maneuver duration which are 50 km h−1 and 1 s vs. Fig. 11. (Scenario 2) Torque-speed (𝑇 , 𝜔) and efficiency contours for the IWMs of the
100 km h−1 and 3 s, respectively, and aims at highlighting the impact target vehicle. The blue circles identify the trajectory when no penalty is introduced
and the MPC algorithm just accounts for non-ideal efficiency operations due to (30); the
of the vehicle’s speed on the optimal strategy achieved by the MPC
cyan squares identify the trajectory when 𝑤IWM = 1000; the magenta triangles identify
algorithm. the trajectory when 𝑤IWM = 10000; the red diamonds identify the mechanical brake
How Fig. 12 shows in comparison to Fig. 10, at lower speeds when input. As 𝑤IWM is increased higher efficiency operations are achieved.
𝑤IWM = 0 the MPC algorithm selects higher electric brakes enabling

11
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Fig. 13. (Scenario 3 is Scenario 2 at lower initial speed) Torque-speed (𝑇 , 𝜔) and

efficiency contours for the IWMs of the target vehicle. The blue circles identify the
trajectory when no penalty is introduced and the MPC algorithm just accounts for
non-ideal efficiency operations due to (30); the cyan squares identify the trajectory
when 𝑤IWM = 1000; the magenta triangles identify the trajectory when 𝑤IWM = 10000;
the red diamonds identify the mechanical brake input. As 𝑤IWM is increased higher
efficiency operations are achieved.

(which correspond to the points closer to the origin) the higher re-
generation is achieved by the strategy with no bounds on the electric
brake variation while, as the vehicle’s speed decreases, the intensity of
the electric brake tends to the same value for the strategy with same
initial speeds. This can be easily explained with the evolution of the
Fig. 12. (Scenario 3 is Scenario 2 at lower initial speed) Optimal hybrid brake mechanical brake which approaches its maximum value as smoothly
trajectories in the braking operating region for several weights of the IWMs penalty
as to make the bound on the electric brake variation ineffective.
term. The blue circles identify the trajectory when no penalty is introduced and the
MPC algorithm just accounts for non-ideal efficiency operations due to (30); the cyan From the point of view of the efficient operations, Fig. 15 provide a
squares identify the trajectory when 𝑤IWM = 1000; the magenta triangles identify the very useful insight in that, even though the strategies with same initial
trajectory when 𝑤IWM = 10000; the red diamonds identify the mechanical brake input. speeds but different bounds on the electric brake variation look similar,
Substantial differences are noticed between the trajectory at 𝑤IWM = 0 and the others.
they are very different. In particular, in case the bound is present it
plays a relevant role especially at the beginning of the brake and for
𝑣(0) = 75 km h−1 because it forces the controller to provide a higher
the hybrid brake to keep closer to the hybrid brake locus. However, number of points for the initial steep variation of the torque, thus
this effect vanishes as 𝑤IWM is increased. implying the IWMs to operate for a longer time at higher efficiencies
Also in this case, Fig. 13 provides further insights. For the initial (black pentagons vs. cyan squares). A similar consideration holds also
vehicle’s speed considered, the IWMs can provide the maximum con- for 𝑣(0) = 125 km h−1 , however in this case the additional points of
stant torque for regeneration and the controller chooses the electric the trajectory with bounded electric brake variation do not lie within
brakes such that this limit is active when 𝑤IWM = 0. However, this strat- higher efficiency regions. On the contrary, the strategy ‘‘waste’’ a
egy attains operations without considering other choices with higher number of decisions selecting braking torques that are not the best
efficiency. At the end of the maneuver the grip limits actually play choices in term of efficiency. This fact allows to interestingly conclude
the relevant role (see Fig. 12). As 𝑤IWM is increased, the penalty on that bounds on the electric brake variation can be used beyond comfort
low efficiency operations steers the MPC decision to different options. purposes, i.e., to force high efficiency operations according to the IWMs
Actually, the trajectory for 𝑤IWM = 10000 is that one which majorly lies speed and for a given efficiency map.
in the higher efficiency regions.

6.5. (Scenario 4) time domain performances and recovered energy

6.4. (Scenario 4) full problem, different initial speeds and comfort con-
straints
The time domain performances of the complete strategy (Scenario
The fourth scenario shows the MPC controller strategy when the 4) is reported in this section. In Fig. 16 the time domain velocities
full problem (37) is solved by comparing the outcomes for different and decelerations of each case of Scenario 4 are reported in solid and
initial speeds and comfort constraints and a gradual brake input with dashed lines, respectively. In particular, the cyan and the black plots
lower intensity with respect to the previous scenarios. Fig. 14 shows correspond to the velocities (solid lines) and decelerations (dashed
the hybrid brake achieved in the brake operating region, where the lines) for 𝑣(0) = 75 km h−1 and 𝛥𝐹̄𝐵,𝐸 = {100, ∞} N respectively, and the
magenta triangles and the cyan squares identify the hybrid brake at blue and the magenta plots correspond to the velocities (solid lines)
𝑣(0) = 125 km h−1 and 𝑣(0) = 75 km h−1 , respectively, with no bounds on and decelerations (dashed lines) for 𝑣(0) = 125 km h−1 and 𝛥𝐹̄𝐵,𝐸 =
the electric brake variation, and the blue circles and black pentagons {100, ∞} N, respectively.
identify the hybrid brake at 𝑣(0) = 125 km h−1 and 𝑣(0) = 75 km h−1 , The plots of the velocity almost overlap when referring to the same
respectively, for 𝛥𝐹̄𝐵,𝐸 = 100 N. As it can be noticed, at higher speeds initial speed which shows that the impact of the bound on the electric

12
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Fig. 15. (Scenario 4) Torque-speed (𝑇 , 𝜔) and efficiency contours for the IWMs of
the target vehicle. The cyan squares and the black pentagons identify the trajectories
for 𝑣(0) = 75 km h−1 and 𝛥𝐹̄𝐵,𝐸 = {100, ∞} N respectively; the blue circles and the
magenta triangles identify the trajectories for 𝑣(0) = 125 km h−1 and 𝛥𝐹̄𝐵,𝐸 = {100, ∞} N,
respectively. As it can be noticed trajectories with bounds on 𝐹𝐵,𝐸 imply a higher
number of points about initial speeds, i.e., at the beginning of the brake where steep
variation of the electric brake would be required. Further, for lower initial speeds,
limiting 𝐹𝐵,𝐸 produces as a side effect that a higher number of points within the region
at higher efficiency are achieved with respect to the trajectory where the bound is not
enforced (black pentagons vs. cyan squares).

Fig. 14. (Scenario 4) Optimal hybrid brake trajectories in the braking operating region
for different initial speeds and bounds on the electric brake variation. The cyan
squares and the black pentagons identify the trajectories for 𝑣(0) = 75 km h−1 and
𝛥𝐹̄𝐵,𝐸 = {100, ∞} N respectively; the blue circles and the magenta triangles identify
the trajectories for 𝑣(0) = 125 km h−1 and 𝛥𝐹̄𝐵,𝐸 = {100, ∞} N, respectively; and the red
diamonds identify the mechanical brake input. As the speed decreases, the trajectories
with bounds on the electric brake variation tend to overlap to those with corresponding
same initial speed but no bound.

brake is almost negligible. This is also confirmed by the decelera-

tion plots which shows that the dependence of the decelerations with
respect to the vehicle speed is very small. In spite of a very small
difference among the decelerations for the various cases, displaying
at the initial time instants, all the decelerations tend to approach a
very similar constant value. Basically, the bounds on the electric brake
variation do not affect the brake effectiveness in terms of deceleration
even though they can improve the comfort. Of course, this depends also Fig. 16. Time domain velocities (solid lines, left 𝑦 axis) and decelerations (dashed
on the chosen brake maneuver (38). lines, right 𝑦 axis) for the cases identified by Scenario 4. The cyan and the black
plots correspond to the velocities and decelerations for 𝑣(0) = 75 km h−1 and 𝛥𝐹̄𝐵,𝐸 =
Since the brake effectiveness is practically almost insensitive to {100, ∞} N respectively, and the blue and the magenta plots correspond to the velocities
the bounds on the electric brake variation, as expected, there are no and decelerations for 𝑣(0) = 125 km h−1 and 𝛥𝐹̄𝐵,𝐸 = {100, ∞} N, respectively.
many differences in terms of energy recovered between the strategies
with same initial speed and different bounds 𝛥𝐹̄𝐵,𝐸 as Table 5 shows.
Table 5
Therein, the total kinetic energy that can be recovered during the Comparison of the strategies corresponding to the cases identified by Scenario 4, in
braking is indicated by 𝛥𝐸𝑘 and is given by the difference between terms of energy recovered w.r.t. the available amount.
the vehicle’s kinetic energy at the beginning of the braking and that 𝑣(0) 𝑣(𝑡 = 3 s) 𝛥𝐹̄𝐵,𝐸 𝛥𝐸𝑘 𝐸bat 𝐸bat ∕𝐸𝑘
[kmh−1 ] [kmh−1 ] [N] [W h] [W h] [%]
at the end of the braking. As it can be noticed, in percentage the best
performances are achieved at lower initial speeds (𝑣(0) = 75 km h−1 ) 125 79.45 ∞ 110.28 14.66 13.30
75 29.20 ∞ 56.55 10.53 18.62
with 𝛥𝐹̄𝐵,𝐸 = ∞. The relevant factor at higher speeds is the limitations 125 79.62 100 109.95 14.11 12.83
of the IWMs in terms of maximum opposing torque they can provide, 75 29.65 100 56.20 9.69 17.24
indeed, while at lower speeds the relevant factor is the grip limit, as
also already discussed in Section 6.4 and highlighted in Figs. 14 and
15.

13
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

7. Conclusion Dépature, C., Lhomme, W., Bouscayrol, A., Boulon, L., Sicard, P., & Jokela, T. (2017).
Characterisation of the electric drive of EV: on-road versus off-road method. IET
Electrical Systems in Transportation, 7(3), 215–222.
In this paper an MPC scheme for regenerative braking in vehicles Eckert, J. J., de Alkmin e Silva, L. C., dos Santos Costa, E., Santiciolli, F. M.,
with hybridized architectures via aftermarket kits has been presented. Corrêa, F. C., & Dedini, F. G. (2019). Optimization of electric propulsion system
In particular, the developed control scheme targets a real applica- for a hybridized vehicle. Mechanics Based Design of Structures and Machines, 47(2),
tion that is under development within the European Horizon2020 175–200. http://dx.doi.org/10.1080/15397734.2018.1520129.
European Environment Agency (2018). EEA report confirms: electric cars are bet-
LIFE-SAVE5 funded project, for which a working hybridized vehicle is
ter for climate and air quality. https://www.eea.europa.eu/highlights/eea-report-
already established and features hybrid traction. The proposed MPC confirms-electric-cars.
scheme includes several constraints for the achievement of feasible Fajri, P., Lee, S., Prabhala, V. A. K., & Ferdowsi, M. (2016). Modeling and integration
operations and two cost functions. The first cost function addresses the of electric vehicle regenerative and friction braking for motor/dynamometer test
bench emulation. IEEE Transactions on Vehicular Technology, 65(6), 4264–4273.
energy recovered to the battery which has to be maximized during
Grandone, M., Naddeo, M., Marra, D., & Rizzo, G. (2016). Development of a regenera-
braking while the second cost function is a penalty term in case a tive braking control strategy for hybridized solar vehicle. In IFAC-PapersOnLine: 8th
trade-off with high efficiency operations of the IWMs are required. IFAC symposium on advances in automotive control (vol. 49, no. 1) (pp. 497–504).
Indeed, some times higher efficiency operations can be useful because Guiggiani, M. (2014). The science of vehicle dynamics: Handling, braking, and ride of road
and race cars. Springer Netherlands.
they enable to save costs at design stage and during operations. The
Heydari, S., Fajri, P., Rasheduzzaman, M., & Sabzehgar, R. (2019). Maximizing
achieved strategy shows a promising potential to be effectively im- regenerative braking energy recovery of electric vehicles through dynamic low-
plemented onto the new releases of the kit’s management platform, speed cutoff point detection. IEEE Transactions on Transportation Electrification, 5(1),
because it handles consistently the constraints and the objectives with 262–270.
reasonable computation times. Future possible research efforts could be Huang, Y., Wang, H., Khajepour, A., He, H., & Ji, J. (2017). Model predictive control
power management strategies for HEVs: A review. Journal of Power Sources, 341,
directed exactly toward the implementation of an MPC algorithm onto 91–106.
embedded systems with limited computation and memory resources Jung, H., Kim, D., Jung, S., & Lee, D. (2020). Optimization method to maximize
with real-time requirements, using the scheme presented in this paper efficiency map of a drive motor with electrical winding changeover technique for
as a baseline, the inclusion of the efficiency of the power conversion hybrid ev. IEEE Transactions on Applied Superconductivity, 30(4), 1–5.
Kamal, E., & Adouane, L. (2018). Intelligent energy management strategy based on
stages, from the IWMs to the battery, the possible limits deriving by artificial neural fuzzy for hybrid vehicle. IEEE Transactions on Intelligent Vehicles,
the battery specifications (also thermal) and its state-of-health or state- 3(1), 112–125.
of-charge, an improved model of the mechanical braking (i.e., of the Kim, J., & Ahn, C. (2019). Rapid optimization of battery charging- discharging
driver) to be included into the controller so as to take into account profiles using soc-soc rate domain for cruising hybrid vehicles. IEEE Access, 7,
87866–87872.
for the possible driver behavior or an approach to handle unknown
Ko, J., Ko, S., Son, H., Yoo, B., Cheon, J., & Kim, H. (2015). Development
environment conditions affecting, e.g., the grip limits. of brake system and regenerative braking cooperative control algorithm for
automatic-transmission-based hybrid electric vehicles. IEEE Transactions on Vehicular
Funding Technology, 64(2), 431–440.
Krasopoulos, C. T., Beniakar, M. E., & Kladas, A. G. (2018). Multicriteria pm motor
design based on anfis evaluation of ev driving cycle efficiency. IEEE Transactions
This study is supported by a grant from the European Union (LIFE- on Transportation Electrification, 4(2), 525–535.
SAVE Solar Aided Vehicle Electrification, LIFE16 ENV/IT/000442). Leahey, N., & Bauman, J. (2019). A fast plant-controller optimization process for mild
hybrid vehicles. IEEE Transactions on Transportation Electrification, 5(2), 444–455.
Lee, W., Jeoung, H., Park, D., & Kim, N. (2019). An adaptive concept of PMP-
Declaration of competing interest based control for saving operating costs of extended-range electric vehicles. IEEE
Transactions on Vehicular Technology, 68(12), 11505–11512.
The authors declare that they have no known competing finan- Lee, H., Song, C., Kim, N., & Cha, S. W. (2020). Comparative analysis of energy
management strategies for HEV: dynamic programming and reinforcement learning.
cial interests or personal relationships that could have appeared to
IEEE Access, 8, 67112–67123.
influence the work reported in this paper. Li, W., Du, H., & Li, W. (2018). Driver intention based coordinate control of
regenerative and plugging braking for electric vehicles with in-wheel PMSMs. IET
References Intelligent Transport Systems, 12(10), 1300–1311.
Li, G., & Görges, D. (2019). Fuel-efficient gear shift and power split strategy for
parallel HEVs based on heuristic dynamic programming and neural networks. IEEE
Baek, D., & Chang, N. (2019). Runtime power management of battery electric vehicles
Transactions on Vehicular Technology, 68(10), 9519–9528.
for extended range with consideration of driving time. IEEE Transactions on Very
Lian, J., Liu, S., Li, L., Liu, X., Zhou, Y., Yang, F., & Yuan, L. (2017). A mixed
Large Scale Integration (VLSI) Systems, 27(3), 549–559.
logical dynamical-model predictive control (MLD-MPC) energy management control
Biswas, A., & Emadi, A. (2019). Energy management systems for electrified powertrains: strategy for plug-in hybrid electric vehicles (PHEVs). Energies, 10(1), 74–91.
state-of-the-art review and future trends. IEEE Transactions on Vehicular Technology, Liu, Y., Li, J., Lei, Z., Li, W., Qin, D., & Chen, Z. (2019). An adaptive equivalent
68(7), 6453–6467. consumption minimization strategy for plug-in hybrid electric vehicles based on
Bourass, A., Cherkaoui, S., & Khoukhi, L. (2017). Secure optimal itinerary planning for energy balance principle. IEEE Access, 7, 67589–67601.
electric vehicles in the smart grid. IEEE Transactions on Industrial Informatics, 13(6), Maamria, D., Gillet, K., Colin, G., Chamaillard, Y., & Nouillant, C. (2019). Optimal
3236–3245. predictive eco-driving cycles for conventional, electric, and hybrid electric cars.
Chen, B., Evangelou, S. A., & Lot, R. (2019). Hybrid electric vehicle two-step fuel IEEE Transactions on Vehicular Technology, 68(7), 6320–6330.
efficiency optimization with decoupled energy management and speed control. IEEE Mahmoudi, A., Soong, W. L., Pellegrino, G., & Armando, E. (2017). Loss function
Transactions on Vehicular Technology, 68(12), 11492–11504. modeling of efficiency maps of electrical machines. IEEE Transactions on Industry
Chen, B., Li, X., Evangelou, S. A., & Lot, R. (2020). Joint propulsion and cooling energy Applications, 53(5), 4221–4231.
management of hybrid electric vehicles by optimal control. IEEE Transactions on Mohammadi, M. H., & Lowther, D. A. (2017). A computational study of efficiency
Vehicular Technology, 69(5), 4894–4906. map calculation for synchronous AC motor drives including cross-coupling and
Chu, W. Q., Zhu, Z. Q., Zhang, J., Liu, X., Stone, D. A., & Foster, M. P. (2015). saturation effects. IEEE Transactions on Magnetics, 53(6), 1–4.
Investigation on operational envelops and efficiency maps of electrically excited Mohan, N., Undeland, T. M., & Robbins, P. (2003). Power electronics–converters,
machines for electrical vehicle applications. IEEE Transactions on Magnetics, 51(4), applications, and design. John Wiley & Sons.
1–10. Mutoh, N., Hayano, Y., Yahagi, H., & Takita, K. (2007). Electric braking control
COIN-OR (2021). Ipopt (interior point optimizer), an open source software package for methods for electric vehicles with independently driven front and rear wheels.
large-scale nonlinear optimization. https://coin-or.github.io/Ipopt/. IEEE Transactions on Industrial Electronics, 54(2), 1168–1176.
de Jager, B., van Keulen, T., & Kessels, J. (2013). Optimal control of hybrid vehicles. Nian, X., Peng, F., & Zhang, H. (2014). Regenerative braking system of electric vehicle
Springer-Verlag London. driven by brushless dc motor. IEEE Transactions on Industrial Electronics, 61(10),
5798–5808.
Norouzi, A., Heidarifar, H., Shahbakhti, M., Koch, C. R., & Borhan, H. (2021). Model
5
Solar Aided Vehicle Electrification – LIFE16 ENV/IT/000442 – https: predictive control of internal combustion engines: A review and future directions.
//www.life-save.eu/. Energies, 14(19).

14
V. Mariani, G. Rizzo, F.A. Tiano et al. Control Engineering Practice 123 (2022) 105142

Padilla, G. P., Weiland, S., & Donkers, M. C. F. (2018). A global optimal solution to Vatanparvar, K., Faezi, S., Burago, I., Levorato, M., & Al Faruque, M. A. (2019).
the eco-driving problem. IEEE Control Systems Letters, 2(4), 599–604. Extended range electric vehicle with driving behavior estimation in energy
Park, J., Murphey, Y. L., & Abul Masrur, M. (2016). Intelligent energy management management. IEEE Transactions on Smart Grid, 10(3), 2959–2968.
and optimization in a hybridized all-terrain vehicle with simple on–off control of Wächter, A., & Biegler, L. T. (2006). On the implementation of an interior-point
the internal combustion engine. IEEE Transactions on Vehicular Technology, 65(6), filter line-search algorithm for large-scale nonlinear programming. Mathematical
4584–4596. http://dx.doi.org/10.1109/TVT.2015.2466655. Programming, 106(1), 25–57.
Platt, J., Moehle, N., Fox, J. D., & Dally, W. (2018). Optimal operation of a plug-in Williamson, S., Lukic, M., & Emadi, A. (2006). Comprehensive drive train efficiency
hybrid vehicle. IEEE Transactions on Vehicular Technology, 67(11), 10366–10377. analysis of hybrid electric and fuel cell vehicles based on motor-controller efficiency
Pridmore, A., Hampshire, K., German, R., & Fons, J. (2018). Electric vehicles from modeling. IEEE Transactions on Power Electronics, 21(3), 730–740.
life cycle and circular economy perspectives: Technical report, European Environment Wu, J., Ruan, J., Zhang, N., & Walker, P. D. (2019). An optimized real-time energy
Agency, URL: https://www.eea.europa.eu/publications/electric-vehicles-from-life- management strategy for the power-split hybrid electric vehicles. IEEE Transactions
cycle. on Control Systems Technology, 27(3), 1194–1202.
Rezaei, A., Burl, J. B., Zhou, B., & Rezaei, M. (2019). A new real-time optimal energy Wyczalek, F. A., & Wang, T. C. (1992). Regenerative braking concepts for electric vehicles
management strategy for parallel hybrid electric vehicles. IEEE Transactions on - A primer: SAE technical paper, Detroit, MI, USA: SAE International, Society of
Control Systems Technology, 27(2), 830–837. Automotive Engineers.
Rizzo, G., Naghinajad, S., Tiano, F. A., & Marino, M. (2020). A survey on through-the- Xie, S., Qi, S., & Lang, K. (2020). A data-driven power management strategy for plug-
road hybrid electric vehicles. Electronics, 9(5), URL: https://www.mdpi.com/2079- in hybrid electric vehicles including optimal battery depth of discharging. IEEE
9292/9/5/879. Transactions on Industrial Informatics, 16(5), 3387–3396.
Rizzo, G., Pianese, C., Arsie, I., & Sorrentino, M. (2011). Kit for transforming a Xu, F., & Shen, T. (2020). Look-ahead prediction-based real-time optimal energy
conventional motor vehicle into a solar hybrid vehicle, and relevant motorvehi- management for connected HEVs. IEEE Transactions on Vehicular Technology, 69(3),
cle obtained by the kit, wo2011125084. URL: https://patents.google.com/patent/ 2537–2551.
WO2011125084A1. Xu, G., Xu, K., Zheng, C., Zhang, X., & Zahid, T. (2016). Fully electrified regenerative
Rizzo, G., Tiano, F. A., Mariani, V., & Marino, M. (2021). Optimal modulation of braking control for deep energy recovery and maintaining safety of electric vehicles.
regenerative braking in through-the-road hybridized vehicles. Energies, 14(20). IEEE Transactions on Vehicular Technology, 65(3), 1186–1198.
Sautermeister, S., Falk, M., Bäker, B., Gauterin, F., & Vaillant, M. (2018). Influence of Yang, Y., Pei, H., Hu, X., Liu, Y., Hou, C., & Cao, D. (2019). Fuel economy optimization
measurement and prediction uncertainties on range estimation for electric vehicles. of power split hybrid vehicles: A rapid dynamic programming approach. Energy,
IEEE Transactions on Intelligent Transportation Systems, 19(8), 2615–2626. 166, 929–938.
Shabbir, W., & Evangelou, S. A. (2016). Exclusive operation strategy for the supervisory Yuan, X., & Wang, J. (2012). Torque distribution strategy for a front- and rear-
control of series hybrid electric vehicles. IEEE Transactions on Control Systems wheel-driven electric vehicle. IEEE Transactions on Vehicular Technology, 61(8),
Technology, 24(6), 2190–2198. 3365–3374.
So, K. M., Gruber, P., Tavernini, D., Karci, A. E. H., Sorniotti, A., & Motaln, T. (2020). Yuan, Y., Zhou, W., & Shi, L. (2019). An investigation on the control strategies and
On the optimal speed profile for electric vehicles. IEEE Access, 8, 78504–78518. fuel economy of a novel plug-in hybrid electric vehicle system. IEEE Transactions
Stroe, N., Olaru, S., Colin, G., Ben-Cherif, K., & Chamaillard, Y. (2019). Predictive on Vehicular Technology, 68(6), 5271–5280.
control framework for HEV: energy management and free-wheeling analysis. IEEE Zhang, X., Göhlich, D., & Li, J. (2018). Energy-efficient toque allocation design of
Transactions on Intelligent Vehicles, 4(2), 220–231. traction and regenerative braking for distributed drive electric vehicles. IEEE
Suntharalingam, P. (2011). Kinetic energy recovery and power management for hybrid Transactions on Vehicular Technology, 67(1), 285–295.
electric vehicles (Ph.D. thesis), Bedford, U.K.: Dept. Eng. Appl. Sci., Cranfield Univ. Zhang, J., & Shen, T. (2016). Real-time fuel economy optimization with
Tang, X., Bi, S., & Zhang, Y. A. (2019). Distributed routing and charging scheduling nonlinear MPC for PHEVs. IEEE Transactions on Control Systems Technology, 24(6),
optimization for internet of electric vehicles. IEEE Internet of Things Journal, 6(1), 2167–2175.
136–148. Zhang, J., Shen, T., & Kako, J. (2020). Short-term optimal energy management
The Julia Project (2021a). The Julia programming language. https://julialang.org/. of power-split hybrid electric vehicles under velocity tracking control. IEEE
The Julia Project (2021b). A modeling language and collection of supporting packages Transactions on Vehicular Technology, 69(1), 182–193.
for mathematical optimization in Julia. https://jump.dev/. Zheng, C., Li, W., & Liang, Q. (2018). An energy management strategy of hybrid energy
Tiano, F. A. (2020). Energy and environmental analysis for a through-the-road solar hybrid storage systems for electric vehicle applications. IEEE Transactions on Sustainable
electric vehicle including control strategies and life cycle assessment (Ph.D. thesis), Energy, 9(4), 1880–1888.
Salerno, Italy: Dept. Ind. Eng., University of Salerno. Zulkefli, M. A. M., & Sun, Z. (2019). Fast numerical powertrain optimization strategy
Uebel, S., Murgovski, N., Tempelhahn, C., & Bäker, B. (2018). Optimal energy for connected hybrid electric vehicles. IEEE Transactions on Vehicular Technology,
management and velocity control of hybrid electric vehicles. IEEE Transactions on 68(9), 8629–8641.
Vehicular Technology, 67(1), 327–337.

15