Variable structure control for an isolated boost converter used in fuel cell applications

Received Oct 24, 2018 Revised Jun 24, 2019 Accepted Jul 4, 2019 In recent years fuel cells have become prominent as an alternative source of energy to meet the society’s energy requirements. A control strategy derived from variable structure theory known as Sliding Mode Control (SMC) was proposed for an Isolated Boost topology which was mostly used in fuel cell systems. Converter operation and its detailed mathematical modelling are also presented. Then the converter with the control strategy suggested is simulated in MATLAB/SIMULINK and compared with other controllers. The results show that transient response of the converter is very fast and steady state error is reduced throughout the load change period with proposed control topology.


INTRODUCTION
As fossil fuel reserves are exhausting, environment is polluted due to gas emissions and as energy demands are increasing an emergency situation arises to search for an alternative source of energy. As a result various renewable energy sources have been explored and Fuel cells arise as the most efficient alternative energy sources because of its advantages like refuel ability, producing very low emissions, portable in size, little maintenance etc. In spite of these benefits the price of fuel cell is its main constraint [1,2]. Fuel cell is an electro chemical device which gives electricity by combining hydrogen and oxygen without any combustion producing water and heat. The output voltage of these fuel cells largely varies under variable load conditions and is also very low in magnitude. This low voltage must be raised to peak of the utility to interconnect fuel cells to grid. Therefore DC-DC converters which boost the voltage to the required level are necessary to connect fuel cells to utility loads.
Many models of DC-DC converters are mentioned in the previous survey papers which are acceptable for fuel cell applications. From the suggested topologies in the literature [3][4][5] Isolated Full Bridge topology was most efficient and acceptable for use in fuel cells. The main advantages of isolated full bridge topology are possibility of applying soft switching techniques, reasonable device voltage ratings, less transistor voltage and current stress, possibility of connecting devices in parallel to achieve desired power levels and high efficiency and galvanic isolation.
Under varying load and varying input voltage conditions DC-DC converters should provide regulated DC output voltage. Changes in time, temperature, pressure etc changes the values of the converter components. Applying of negative feedback in the form of a closed loop, regulation of DC voltage can be achieved. So the next task in designing a power conditioning system is developing a best controller meeting the requirements suitable for fuel cell applications. Basically there are two types of controllers, Voltage control and current control. Vast majority of DC-DC converters are implemented using current mode controllers because of its advantages [6][7][8][9]. The basic performance of current control is as indicated in Figure 1. Here the output stage is fed by the inductor current and the error voltage output from the error amplifier (known as control voltage) controls it. This has an extreme effect on a dynamic behaviour of the negative feedback control loop. Using this approach many techniques are implemented like linear peak current control (LPCM), non-linear carrier control (NLC), predictive switching modulator (PSM) control etc.
In the linear peak current control (LPCM) [10] peak switching current has to be sensed and therefore the sensing devices needed are more. In non-linear carrier current control (NLC) [11,12] three reset integrators are needed to generate the carrier waveform. The NLC controlled converter results in distorted input current waveform in discontinuous conduction mode of operation.
To overcome the problems in LPCM and NLC controllers another control known as Predictive switching modulator control (PSM) [13][14][15] is proposed where only two reset integrators are used when compared to NLC. Also PSM [16,17] control extends the range of continuous conduction mode as prediction of off state ripple current is possible which is added to on state actual current at the end of switching period. The transient reponse is also very slow with these controllers [18].
But the above methods of control require small signal representation of the system for an appropriate control to be developed and also the linear time invariant nature produces unpredicted action in the converter when running out of the normal operating point as small signal parameters strongly depend on it. Therefore to improve the transient reponse by considering the stability of the system in any case and its static and dynamic performances a control method should be developed. The control should also ensure not to accept disturbances in input voltage and effects of load variations. Generally the non-linear types of controllers are mostly used now days as they can meet the requirements of the control considering the uncertainties and non-linear state of the converter. This concept is obtained from the non-minimum phase character and variability in the structure of converters with unforeseeable non-linear change in the loads. As a result a robust control method was proposed based on variable structure theory taking into account the non-linear nature of the converters.
The introduction of variable structure system control popularly known as Sliding Mode control and its development was started in the early of 1950's in Soviet Union by Emelyanov and many co-researchers like Utkins and Itkis [19]. This method of control has many advantages as: excellent stability and Robustness for large input and output variations when compared to any other hysteresis control; as all the control loops are simultaneous, dynamic response of the system is very fast; execution of this control is very easy and not complicated; if N is the system state and N-1 is the state of system's response, then actually sliding mode have just N-1 independent state variables where ℎ variable is restricted by sliding surface. The proposed control was applied to IFBC and simulated in MATLAB/SIMULINK. The obtained results were compared with LPCM, NLC and PSM controllers in terms of transient response parameters.

ISOLATED FULL BRIDGE BOOST CONVERTER TOPOLOGY (IFBC) 2.1. Basic converter operation
In this section complete working of the converter is presented. The converter circuit is represented by Figure 2 and its performance according to modes of operation is described by Figure 3.

SLIDING MODE CONTROL THEORY
This control is developed by utilizing concept of variable structure systems [21]. In these type of systems using a preset principle of control, system's actual composition is modified deliberately. The moments of modifying the structure can be resolved using current state of the system. This makes the Switching mode power supplies to fall into a specific category of VSS as its structure is modified regularly controlling operation of switches and diodes. In this process the structure of control system changes which indicates Variable structure control (VSC). As VSC is a high speed switching feedback control it results in a Sliding mode. Based on a principle each feedback path gains shift between two values depending on the state value at each instant. The basis of the principle used which is also called as switching control law is to bring the non-linear system's trajectory on the surface in successive times. This surface is known as Switching Surface. The feedback path has a gain value if the plant's trajectory state is "above" this surface and it has a unique gain value when trajectory falls "below" the surface. Actual switching regulations are described by the surface. Perfectly speaking, control retains the trajectory state of the plant on the surface in successive times once obstructed. The plant dynamics confined to the surface indicate the performance of the system under control. The configuration of VSC can be split into two modules. -In the first part a switching surface is chosen which makes the trajectory state of the plant confine to the surface and has required dynamics. -In the second part a switching control law is modelled which moves the plant's state to switching surface and restricts to the surface on obstruction.
To specify the second part design, a lyapunov approach is used. In this approach a generalised lyapunov function which defines the trajectory state movement towards the surface is formed in terms of surface. The "gains" of each switched control structure is chosen such that the derivative of the lyapunov function is negative definite ensuring the movement of the trajectory state to the surface. This control is also known as Sliding Mode control highlighting the part of sliding mode. The variable structure control system may be formulated without a sliding mode, but such a system does not have related advantages.

Modelling of sliding mode system
Now consider a group of systems comprising state model with non-linear state vector (t) and linear control vector (t) in the form as [22] Where t ϵ and ( , ) ϵ and furthermore every entry in ( , ) and ( , ) is supposed to be continuous with a bounded continuous derivative with respect to .
A switching surface is chosen in the first module which makes trajectory state of plant confine to the surface and has required dynamics. Therefore by following the sliding mode control theory [14], the state variables of all areas are sensed and by using appropriate gains the states are multiplied and added together for the construction of sliding function ψ ( , ). The hysteresis block sustains this function to zero thereby the sliding surface can be outlined as Where N is the order of the system (i.e. count of state variables) Considering second order system This is a linear combination of two state variables. The relation ψ = 0 indicates a line in phase plane which passes from origin (final equilibrium point of the system) and is called as sliding line.
Let us now specify the following control strategy Where represents an appropriate hysteresis band. Likewise the phase plane is split to two regions partitioned with sliding line and individual region is related to one of the two sub topologies represented by switch position . Looking at phase trajectories in closeness to sliding line, they are directed towards line only. Continuous commutations around the sliding line leads to motion which makes the system position drive to final equilibrium point. Based on the theory of a satisfactory small value of , two important resolutions are derived from the example.
-The system's progress is independent of its parameters when it is in sliding mode and depends only on chosen sliding line. Example in Figure 4 is system with first order dynamics and time constant equal to . -Suppose order of the original system is N then system control dynamics in sliding mode possess order N-1 as the state variables are restricted by ψ = 0 Here using the amplitude of the hysteresis band switching frequency is derived. The abilities of the control technique are now visible for application in switched mode power supplies. Utilizing the inherent non-linear nature of these converters unlike to substructures constituting the system and its reduction dynamic performances are produced.

Modelling of switching surface
To model sliding mode control for the simple second order systems reviewed in previous section, choice of parameter is enough. Choice of this parameter should satisfy the following three constraints. Disregarding the starting point in phase plane system trajectories are needed to cross the sliding line which is known as hitting condition. The system trajectories which are near to the sliding line in both the regions are managed to run towards the line itself and is known as existence condition. The movement of system on sliding line (i.e. shifting towards equilibrium point) is regarded as stability condition.

Existence condition
In this condition the phase trajectories are needed to move towards sliding surface by little portion about the surface itself [23]. This is reached by interpreting a suitable Lyapunov function V ( , , ψ). Generally it is appropriate to select a Lyapunov function of the form V ( , , ψ) = 0.5 2 ( ). To deduce the required gains so that the system state drives to the surface ψ(t) = 0, one can select such that So, when the distances to the surface and velocity of its change ̇ are of opposite signs i.e. when lim →+0̇> 0 and lim →+0̇< 0 the sliding mode exists on a discontinuity surface.

Stability condition
Modelling of switching surface is estimated based on how the system understands its operation in sliding mode. The parameters of switching surface decide the system operation. If in any case attaining model of switching surface require logical reasoning describing the movement of state trajectory in sliding mode a method called Equivalent control is necessary for this description.

Equivalent control
This comprises an equivalent input and as the system stimulates, it provides system movement on sliding surface as long as the initial state is on. Let us assume that at time 1 plant's state trajectory obstructs switching surface and a sliding mode exists. This indicates that, for all t ≥ 1 , ( ( ), ) = 0 and hence ̇( ( ), ) = 0. Utilizing chain rule [23], the equivalent control for the system that can be interpreted as the input fulfils the condition ̇= +̇= + ( , ) + ( , ) = 0 Supposing that the product matrix ( , ) is non-singular for all t and , one can calculate as Thus given ( ( 1 ) This shows dynamics of equivalent system on sliding surface. When some form of tracing or control is needed by the control system then the operating term is present. For example, when Where r(t) behaves as a "reference" signal.

Actual interpretation of equivalent control
A real control is that it always encompasses a slow component for which a high rate component is added. Therefore the control structure is decomposed as In which is valid only on sliding surface and satisfies actuality of sliding mode. Now is described as Where The slow component indicates performance of the plant which is a dynamic device, but the plant's reaction to high rate component can be neglected. On the other side here there is need for interchange of actual control in the movement (12) with continuous function ( , ) that does not possess any high rate component. Thus the slow component of actual control which is the average control value is made equal therefore by using first order linear filter it can be estimated once the time constant is small in contrast with slow component, besides large enough so that high rate component can be filtered and can be roughly made equal with boundary layer width.

OUTPUT VOLTAGE REGULATION OF IFBC WITH SLIDING MODE CONTROL
By forcing the states to zero the regulation problem is resolved. In constructing the switching function, the nominal linear system representation is considered only when the matched uncertainty is present. Implementation IFBC with sliding mode control is depicted by Figure 5.
Here ( ) is nothing but switch state or when average model is considered [24] it is the switch duty cycle and , are inductor current and output voltage of converter respectively. In sliding mode control ( ) is described by the (4).

Average model of IFBC
As seen in section 2 operation of IFBC has two main intervals and their identical circuit diagrams are as shown in Figure 6.
This interval continues for (1-d) T. The state equation for output in both the operations is same as (20). On averaging the state equations acquired during half of the switching cycle, a relation which has the features of both the modes is obtained Here d is the controlling parameter of the system and is same as

Control model
The main intention of IFBC control is regulation of output voltage to a reference value . The modelling of sliding mode controller begins with selection of sliding surface. Only when the current increases constantly the direct surface inclines to zero. For solving two loop control problem, generally a cascade control structure is employed. Output voltage loop produces the reference current from the voltage error and inner current loop restricts inductor current through sliding mode.
Controlling of output voltage of the converter satisfies stability and presence of sliding mode. As sliding mode is extremely a non-linear method, it is very tough to obtain the gain of voltage loop. Moreover, as SMC can be implemented only to regulate current, voltage loop will quickly respond for any high frequency phenomena and uncertainties in reference current. Therefore to enhance controller capability, sliding surface based control mode that includes output voltage is suggested. For the IFBC converter the system cannot be represented in canonical form as the derivative of output voltage becomes a discontinuous variable. Therefore as state variables the inductor current and output voltage errors are selected Here the inductor current reference relies upon operating point (input voltage and output power) of the converter.

Existence condition
The relations of the IFBC converter system in terms of 1 and 2 are written as ̇= +̂+ (25) Where ̇= [ 1 2 ] Suppose for a short time let the signal is available and constant. The phase trajectories of the system are as shown in Figure 7 and the sliding line equation is as On selecting the control law (4) it is simply observed that both the existence and reaching conditions are fulfilled (the past one at least in a small region surrounding the origin). The sliding mode existence region is specified by the following inequalities from (4) The relations 1 ( ) = 0 and 2 ( ) = 0 describe two lines in phase plane and are as shown in Figure 8. One must make sure that both the intersections of the line 1 ( ) = 0 with the 2 axis and of 2 ( ) = 0 with 1 axis should be positive as can be seen to determine an existence region for sliding mode that involves origin (that shows the steady state point). This limitation from the above inequality determines the gain < To make sure the presence of sliding mode at least in a region about the steady state operating point, the coefficient of sliding surface is selected in such a way that it fulfils the (29). As the latter is based on the converter operating point (output power and input voltage), the presumption about availability of current reference signal is not possible practically. Therefore to obtain this inductor current reference signal that controls the output voltage of converter , generally a low-pass filter (or PI controller) is employed. Obviously all the features of sliding control are considerably affected by this approach. At first owing to new state variable 3 , the order of the system is increased by one. Therefore the whole system can be characterised by selecting the state variables as below

Stability analysis
The system stability is possible only if the system dynamics in sliding rule provide path to the desired equilibrium point. It is suggested that only when the sliding regime is attained the dynamics of new state variables 1 , 2 and 3 are obtained. For the system state to slide along the surface the equivalent average control should be implemented with the defined state space model (31), sliding surface (32) and ̇( ) = 0 and is given by Thus the (34) and (35) represent the modelling relations for sliding mode control of IFBC converter that forces output voltage to zero under steady state.

Switching frequency
Generally a practical system cannot be switched at infinite frequency, so a hysteresis band can be employed throughout the sliding line so that the switching frequency is settled at required value. Now by examining the Figure 9 the switching frequency can be evaluated

SIMULATION AND RESULTS
The Sliding mode control is modelled as stated in previous sections and is applied to IFBC converter. IFBC with SMC is simulated in MATLAB Simulink. The modeled sliding mode control block of IFBC and its sliding surface in MATLAB are as shown in Figure 10 and Figure 11 respectively. The converter specifications were given in Table 1. The output voltage of the converter is shown in Figure 12 with load change at 0.5sec and Figure 13 shows comparison of sliding mode control with PSM, NLC and LPCM controllers. Table 2 gives detailed performance analysis of IFBC with SMC, PSM, NLC and LPCM controllers.

CONCLUSION
In this paper an Isolated Full Bridge Boost Converter mostly used for fuel cell applications was presented and implemented in closed loop configuration. A robust control technique for controlling output voltage of converter known as Sliding Mode Control (SMC) was proposed and presented in detail. SMC was applied to the IFBC topology and the results obtained are compared with other controllers. From the comparison results it can be shown that faster transient response of the converter is achieved i.e the rise time and settling time has been improved and when a disturbance i.e load change has occurred the response of output voltage to reach the actual value is very fast with Sliding Mode Control. Inspite of these many advantages, the only disadvantage observed is the peak overshoot at load change which can be limited by designing any other advance controllers like H-infinity or ANFIS.