*2.1. System Design*

The electrical schematic and the specifications of the key components of the e-bike charging station are shown in Figure 3 and Table 1, respectively. The core of the system is a 48 V DC nano-grid, which supports the power exchange between all the components. The PV generation consists of eight modules arranged in four parallel strings. A Victron BlueSolar 150/85 maximum power point tracking (MPPT) converter is used to process power from the PV panels to charge the 48 V battery bank [38]. Isolated DC-DC converters and a high-frequency DC/AC inverter are connected to the 48 V DC nano-grid for DC and wireless charging of e-bikes, respectively.

The connection between the DC nano-grid and the 50 Hz AC grid is realized by a Victron Multiplus 48/3000 hybrid bidirectional inverter [38], which is equipped with two AC outputs. One output is connected to the single-phase AC grid. The second output powers both the e-bike AC charger and the lighting and system monitoring unit, such that the e-bike charging station supports the off-grid operation. Lastly, a Lufft WS503-UMB weather station measures the incoming solar radiation, ambient temperature, and local wind speed for monitoring and research purposes [39]. The system control and monitoring are done using a central Raspberry Pi controller, and a website provides users with feedback and data for scientific research (http://solarpoweredbikes.tudelft.nl).

**Figure 3.** Schematic of the solar e-bike station with 48 V DC interconnection that facilities power exchange between the solar panels, e-bike chargers, and the AC grid.


**Table 1.** Specifications of the solar e-bike charging station.

The power balance equation of the system is:

$$P\_{PV} + P\_{grid} = P\_{batt} + P\_{load} + P\_{loss} \tag{1}$$

where *PPV* is the PV power, *Pgrid* is the power drawn from the grid, *Pbatt* is the charging power of the battery, *Pload* is the power consumption of the station including that of the e-bikes and the baseload, and *Ploss* is the total energy conversion losses.

## *2.2. Charging Demand*

The PV system and storage of the charging station is sized to charge up to five e-bikes and a single e-scooter per day throughout the year, even under low winter insolation. Table 2 shows the specifications of the e-bike, e-scooter, and small car considered for the system sizing [40]. Each e-bike and e-scooter battery has a capacity of 396 Wh and 1920 Wh, respectively. A total of five such e-bikes and one e-scooter, assuming that they arrive with fully drained batteries (the worst case), would translate to a load of 3900 Wh. A baseload of 90 W is consumed by the display, controller, lights, and the weather station, which translates to 2160 Wh. If the charging demand and baseload are combined, the net demand is approximately 6.06 kWh per day. This is the same as the load of one Renault Twizy small electric car. If the charging demand is >6 kWh due to a higher number of e-bikes, then they can be charged by using PV power if the daily PV yield is >6 kWh or using the AC grid power via the DC/AC inverter up to 3 kW\*24 h = 72 kWh per day.


**Table 2.** Specifications of the E-bike and E-scooter.

Three types of e-bike charging methods are developed for the charging station: AC, DC, and wireless charging. The benefit of the AC charging is that it can be universally used for charging all e-bikes, e-scooters, and light EVs by using a charging adapter. On the other hand, for safety reasons, the DC charging is limited to 100 W but has the benefit that the users can simply use a DC cable between the station and the e-bike battery for charging. This DC charging negates the need for an AC power adapter and, hence, provides convenience to the user while plugging in and preventing any possible theft of the adapter. In the case of the wireless charging, the motive is to do away with the user's need for cables altogether, which, therefore, increases the user convenience further [29,30].

## *2.3. Local Storage*

The battery plays a crucial role in providing the o ff-grid capability to the charging station. The station has four lead-acid gel batteries of 220 Ah capacity each. The batteries are series-connected to 48 V and provide a usable capacity of 9.5 kWh, when operated at a maximum depth of discharge of 90%. With ~6 kWh demand per day, it can provide close to 1.5 days of autonomy to the system. Lead-acid gel batteries were preferred due to two reasons. First, they have a much longer lifetime than standard lead-acid batteries, and second, the cost is much lower when compared to lithium-ion batteries.

The preferred charging mode is to charge the batteries directly from the solar panels on DC. In the case of insu fficient solar generation, the battery can be charged (and also discharged) from the AC grid using the DC-AC grid inverter. This bidirectional option can be useful for di fferent grid support services such as peak shaving, reactive power compensation, energy arbitrage, or emergency backup power.

#### **3. PV System Design**

The 2.6 kW PV system is the primary source of power for the e-bike charging station. The PV generation potential is estimated based on solar insolation, wind speed, ambient temperature, panel orientation, and shading due to the surrounding terrain. The meteorological data Cabauw Experimental Site for Atmospheric Research (CESAR) database is used for the PV system modelling, which has a resolution of 1 min [41].

#### *3.1. PV System Modelling*

Using the CESAR data, the incident solar irradiance and the cell temperature of the PV panel are estimated for di fferent azimuth and orientation using Equations (2)–(6) [42–47].

$$G\_{dir(\beta,Am)} = G\_{DNI} \cos(\theta\_i) \tag{2}$$

$$G\_{diff(\beta)} = G\_{DHI} \left( 1 + \cos \beta \right) / 2 \tag{3}$$

$$G\_{\rm gnd} = G\_{\rm GHI} \,\rho \Big(1 - \mu^{\rm sgf} \Big) \tag{4}$$

$$G = G\_{dir(\beta, Am)} + G\_{diff(\beta)} + G\_{alb} \tag{5}$$

$$T\_{\rm cell} = \frac{\Phi G + h\_c T\_{amb} + h\_{r,sky} T\_{sky} + h\_{r,gr} T\_{\rm gr}}{h\_c + h\_{r,sky} + h\_{r,gr}} \tag{6}$$

where *Gdir*, *Gdi f f* , and *Ggnd* are the direct, di ffused, and ground irradiance incident on the module with a tilt β and azimuth *Am*, *GDNI*, *GDNI* and *GGHI* are the direct normal, di ffuse horizontal, and global horizontal irradiance, θ*i* is the angle of incidence of the direct irradiance beam on the panel, Φ = 0.727 is the absorptivity, ρ = 0.2 is the albedo (measured using albedometer), μ*sv f* is the sky view factor, *hc*, is the coe fficient for convective heat transfer *hr*,*sky*, *hr*,*gr* are the coe fficient for radiative heat transfer to the sky, and to the ground, respectively, and *Tamb*, *Tsky*, *Tgr*, and *Tcell* are the ambient, sky, ground, and PV cell temperature, respectively. In this case, the heat transfer coe fficients are estimated using an iterative procedure, where *Tsky* = 0.0522 *T*2/3 *amb*. Lastly, the output power of the PV modules *PPV* can be calculated. 

$$P\_{PV} = P\_{STC} \left(\frac{G}{G\_{STC}}\right) \left[1 - \gamma \left(T\_{cell} - 25\right)\right] \tag{7}$$

where *PSTC* = 327 W is the module power at Standard Test Conditions (STC), *GSTC* = 800 <sup>W</sup>/m<sup>2</sup> is the irradiance under STC, η = 20.3% is the e fficiency of the modules, and γ = −0.3%/ ◦C is the module temperature coe fficient for the Sunpower X20-327-BLK modules.

#### *3.2. PV System Orientation*

Based on the meteorological data of 2013, the yield of the PV system is estimated for di fferent fixed orientations. For the whole year, considering a fixed orientation, a tilt of 28◦ and azimuth facing south was found to result in maximum annual yield [6]. Figure 4 shows the average daily yield of the south-facing PV system for each month, considering various fixed tilt angles between 0◦ and 90◦. It can be clearly seen how the tilt angle has a major influence on the monthly yield, especially in the summer months.

Figure 4 also shows the average daily yield if a two-axis tracking system is used. First, it can be seen that the tracking system has a much higher yield as it tracks both the azimuth and tilt when various tilt angles are considered in which the azimuth is fixed. Second, it can be seen that the increase in yield due to tracking is very small for the winter months as most of the irradiance is di ffused irradiance. Due to the higher cost of the tracker and the need for a fixed structure for the bike shelter, tracking systems were not further considered in this study.

**Figure 4.** Average daily yield of the south facing 2.6 kW PV system for each month considering various fixed tilt angles compared to the use of a 2-axis tracker.

#### *3.3. Optimal PV System Design*

In order to ge<sup>t</sup> the maximum annual yield, the PV system could have been oriented at a tilt of 28◦ and azimuth facing south. However, the motivation is to ensure that there is sufficient generation in December, which has the lowest solar irradiation in the year. For December, the optimal tilt for maximum monthly yield was 65◦. This is shown in monthly yield estimation for December in Figure 5a for different PV orientations.

**Figure 5.** (**a**) PV system yield (% of maximum) for different orientations considering December month. (**b**) The estimated monthly energy yield of 2.6 kW PV system tilted at 51◦ and facing south.

On the other hand, when the tilt is increased from 28◦ to 65◦, the annual yield is dramatically reduced by up to 20%. Hence, a tradeoff was made to set the tilt angle at 51◦, which results in <5% reduction in annual yield compared to 28◦ and 1% reduction in December yield compared to 65◦.

## *3.4. Shading Analysis*

Since the e-bike station is installed at the ground level, shading from nearby buildings (especially the tall electrical faculty building) has a significant impact on the PV system output, as shown in Figure 6a. To account for the shading due to the adjacent buildings, their 3D models were made using Sketchup. Figure 6b shows the shading caused by the tall Electrical Faculty building in front of the bike station in the afternoon. The plug-in LSS-Chronolux 3D is then used to estimate the sky view factor μ*sv f* = 0.61 and the per minute shading factor *Ss<sup>h</sup>* as well as the corresponding direct irradiance including the shading, *Gshdir*(β,*Am*):

$$\mathcal{G}^{sh}\_{\text{dir}(\boldsymbol{\beta}\_{\mathcal{A}}\boldsymbol{m})} = \mathcal{S}^{sh}\mathcal{G}\_{\text{dir}(\boldsymbol{\beta}\_{\mathcal{A}}\boldsymbol{m})} \tag{8}$$

#### *3.5. PV System Yield*

Using the 2013 CESAR data and the shading analysis, the PV yield of the system is estimated. The corresponding annual energy yield is 2012 kWh/year, which provides an average daily yield of 5.51 kWh/day, as shown in Figure 5b. Based on the charging demand (Section 2.2), the PV system is sized to a rated power of 2.61 kWp. This means that >90% of the total load demand can be supplied by the PV system on average. At the same time, we have to keep in mind that there is about six times the difference in energy yield between the winter and summer months, as shown in Figure 5b. This results in the load being met on about 180–200 days of the years. That is why a grid connection is required to provide the energy demand in winter, as seen in Equation (1). While the battery cannot provide seasonal storage, it can facilitate off-grid operation and help manage the diurnal solar insolation variation.

**Figure 6.** (**a**) Google Earth image showing the station location and nearby building. (**b**) Calculation of the shading factor due to nearby buildings using Sketchup.

The highly efficient Sunpower X20-327-BLK modules with a relatively high cost per watt peak are chosen since this reduces the total area occupied by the PV system. The tradeoff is that, in return, the lower area significantly reduces the quantity of steel required for the station structure, which has a relatively higher cost for both material and labour.

#### **4. AC E-Bike Charging**

The e-bike charging station provides a single 230 V 50 Hz schuko wall socket for AC charging of e-bikes through the use of a charging adapter. The benefit of the AC charging is that it can be universally used for all light EVs providing up to 16 A charging current, which corresponds to a power of 3.7 kW. This is more than sufficient to charge a small electric car like the Twizy as shown in Figure 2b, which requires a charging current of 10 A. By using the hybrid grid inverter, the AC charging is possible in both grid-connected and an off-grid mode.

#### **5. DC E-Bike Charging**

The DC chargers for the e-bikes should provide galvanically-isolated DC power that is controllable in both output voltage level and maximum charging current, based on the e-bike battery. A dual interleaved quasi-resonant flyback converter with digital current-mode control from Involar is chosen for this task [14,48]. As shown in Figure 7a, the primary side of the flyback is operated directly from the 48 V DC nano-grid. The secondary side of the flyback is connected to the DC cable that the user has to plug into the charging inlet of the e-bike battery. The input current ripple is reduced by half due to the interleaving. The magnetically coupled inductors are wound to have minimum parasitic leakage inductance to reduce the Electromagnetic interference (EMI) and the voltage stress on the MOSFET.

#### *5.1. Current Mode Control*

Depending on the required output power to charge the e-bike battery, the current level in the primary circuit of each flyback is regulated using the current mode control. Each flyback employs two control loops that control the gate of the MOSFET, represented in Figure 7a by the green block, and shown in detail in Figure 8. The inner control is a digitally implemented current mode controller implementing quasi-resonant switching. In this case, the internal current control (curved red line) regulates the amount of energy that is allowed to flow out of the charger and, in that way, sets a limit on the maximum continuous output current, typically 1 A or 2 A. The outer voltage feedback (curved blue line) controls the maximum output voltage of the charger and, accordingly, turns the charger on/off. The maximum output voltage is limited to the nominal voltage expected by e-bike's battery managemen<sup>t</sup> system, which is typically 24, 36, and 48 volts. A Type-2 control is implemented and is

shown on the left side of the control IC in Figure 8, including the current mode controller and gate driver [49].

**Figure 7.** (**a**) Flyback converter for e-bike battery charging with a digital current mode control and charging current monitoring and logging. (**b**) Simulation results show the primary (red) and secondary (blue) current as well as the charge voltage (Vout = 36 Volt) and charge current (Ich = 2.4 Ampere) for the 48-volt DC grid, and operating in a continuous conduction mode.

**Figure 8.** Current mode control of the flyback converter using an outer voltage loop and an inner current loop.

#### *5.2. Design and Simulation of the Flyback DC Charger*

The flyback converter is designed based on the design procedure for a quasi-resonant operation [10]. Figure 7b shows the simulation of the e-bike charger in Caspoc, where only a single leg flyback is shown for simplicity [50,51]. The multi-level modelling method is employed, which allows the modelling of both the power electronics circuit and the hybrid control consisting of the inner current and outer voltage control loop [52]. The simulation takes into account the delays of the digital feedback compensation as well as the delays caused by switching of the MOSFET. Voltage overshoot caused by the non-coupled parasitic winding inductance is taken into account in the simulation, but not shown here in Figure 7b for clarity.

Figure 7b shows the primary and secondary currents for a duty cycle of 50% in a continuous conduction mode for a switching frequency fs = 100 kHz and turns the ratio of Np:Ns = 48:36. The scope clearly shows the difference in the voltage level on the primary side of 48 V, and secondary side of 36 V, and the charging current of 2.4 A, which is measured on the second scope.

#### *5.3. Safety and Monitoring*

As long as no e-bike is connected, the mechanical on/off switch on the flyback's primary side (Figure 6a) is in the off position. As soon as an e-bike battery is connected to the flyback's secondary side, the voltage is monitored, and the switch is set to the 'on' position. This subsequently powers the flyback converter and its internal control. If either the connection with the e-bike's battery is removed or the charging current drops below the minimum charging current threshold, the outer control turns off the flyback converter by setting the On/Off control to its off position.

#### *5.4. Hardware Realization and Losses*

Figure 9a shows the hardware of the flyback DC charger [48]. On the primary side, IRFS4321PBF MOSFETs with low on-state resistance Rds(on) = 15 mΩ and, on the secondary side, MBR10150 schottky diodes with negligible reverse recovery losses are used. The primary current is sensed using a low ohmic sense resistor placed between the MOSFET source and ground, which is filtered using a low pass first-order filter. In this way, the maximum power is limited by the maximum primary current. The output voltage is measured and, after filtering, sampling is input to the current mode controller. The feedback amplifier with a Type-2 compensation network is replaced by its digital equivalent 2-Pole, 2-Zero compensator [2p2z].

**Figure 9.** (**a**) Interleaved flyback converter for DC e-bike charging. (**b**) The cable used for DC charging. (**c**) Measured charging current of two e-bikes in which each is charged using the flyback based DC e-bike charger.

For the converter losses, the worst-case operating point is at the lowest output voltage of 24 V and maximum power of 100 W, which corresponds to a current of 4.2 A. For the dual interleaved flyback converter, each converter delivers 50 W and 2.1 A current. The MPR10150 diode has a maximum forward voltage *Vf* = 0.6V at 2.1A, conducts only 50% of the time, and, therefore, has *Pd* = 50% (2.1A)(0.6V) = 0.62 W losses. The average input current for each converter and an input voltage of 48V is roughly *Iin* = 1A. Considering a nominal 50% duty cycle, this will give a peak input current Ipk = 4 A and an RMS value of *Iin(rms)* = 1.7 A. This results in a maximum conduction loss of *(Iin(rms))* 2 *Rds(on)* = 44 mW. As a rule of thumb, the switching losses are of the same magnitude of maximum and twice the conduction losses, which gives a total of conduction and switching losses of Psw = 3(44 mW) = 132 mW. Cooling via the PCB is, hence, su fficient in this case.

Figure 9b shows the custom-designed cable that is used to connect the e-bike to the output of the flyback converter in the charging station. The right-side plug is magnetic and, hence, easily attaches to the station, while the left side plug attaches to the e-bike battery. If required, a custom-designed adapter is made for the e-bike side plug based on the manufacturer-specific connector on the user's e-bike. This points to the need for standardization in DC e-bike charging and plugs similar to what was achieved with CCS and CHAdeMO in electric car charging [53].

Figure 9c shows the charging current of two e-bikes in which each was charged using the flyback based DC e-bike charger. For charger 1, the battery gets charged according to the Constant Current, Constant Voltage (CC-CV) principle beginning at a current of 1.3 A, and then slowly reducing to zero. For charger 2, the battery charging begins in the CV region due to the relatively high state of charge (SOC) of the e-bike battery. It can be observed how the e-bike is disconnected at 120 min and then connected again after about an hour.

#### *5.5. Battery Connection Communication*

The internal digital control employs an automated voltage level detection to allow the direct connection to either a 24 V, 36 V, or 48 V battery. An automated output voltage level detection of the battery voltage is achieved through the outer control loop that compares the output voltage to the reference value. The outer control loop subsequently regulates the amplitude of the primary current pulse through the inner loop. The Raspberry Pi is used to control the flyback operation and to monitor the output voltage and charging current. Connected to the Internet, it exchanges and logs information with a server operating in the cloud.

#### **6. Wireless E-Bike Charging**

Wireless charging provides the most convenient and safe experience for the e-bike user. The cyclists do not need to bring along cables and power adapter because the charging process is facilitated through the bike kickstand, as shown in Figure 10. The developed charging system is started once the bike is parked on the appointed parking spot, which is a 30 × 30 cm tile underneath the solar charging station. On the tile, the users have the possibility to park their bike in any position, which makes the wireless charging convenient. On top of that, the wireless charging has intrinsic galvanic insolation, which does not require the users to touch any conductive parts such as cables and connectors that might become dangerous, especially in wet weather conditions.

#### *6.1. Wireless Power Transfer via Resonant Circuit*

Figures 10, 11a and Table 3 shows the schematic, circuit diagram, and specifications of the wireless power transfer system for the e-bike that works based on inductive power transfer through magnetic resonance. In the developed system, the transmitter coil is located under the charging tile, and it is formed by a U-shaped ferromagnetic core with a winding located at the center. On the other hand, the receiver coil consists of the double kickstand of the e-bike. This coil has a similar magnetic circuit as the transmitter coil but, in this case, the ferromagnetic core is closer to a V-shape to resemble the structure of commercial double kickstands.

Once the bike is parked over the tile, the two coils become coupled, which is equivalent to saying that the magnetic circuit becomes closed, and the charging process is ready to start. In a traditional transformer, the coils are strongly coupled because they are wounded around the same ferromagnetic core, and the airgap's order of magnitude is not comparable to the core's dimensions. On the other hand, when the two coils become coupled, the proposed magnetic circuit has two airgaps of about 5 mm, which are less than one order of magnitude smaller than the cross-sectional area of the core. Therefore, in this case, the equivalent transformer has a high leakage inductance and, consequently, is loosely coupled.

**Figure 10.** Block diagram of the wireless power transfer system for the e-bike.

**Figure 11.** (**a**) Simplified circuit of the e-bike wireless charging system. (**b**) Waveforms of the primary and secondary voltages and currents.


**Table 3.** Values of the components of the e-bike wireless charging system.

The capacitors *C*1 and *C*2 compensate the coils' reactive power and form a resonant circuit together with the inductor coils *L*1 and *L*2, which is tuned to the resonant frequency *f*0. The estimation of *C*1 and *C*2 and the Kirchhoff voltage law equations of the circuit schematic can be found in Equations (9) and (10).

$$a\_0 = 2\pi f\_0 \mathbb{C}\_1 = \frac{1}{a\_0^2 L\_1} \mathbb{C}\_2 = \frac{1}{a\_0^2 L\_2} \tag{9}$$

$$\begin{array}{l} \frac{4}{\pi}V\_{in} = \left(j\omega L\_1 + \frac{1}{j\omega \mathcal{C}\_1}\right)I\_1 + j\omega MI\_2\\ 0 = \left(\frac{8}{\pi^2}R\_L + j\omega L\_2 + \frac{1}{j\omega \mathcal{C}\_2}\right)I\_2 + j\omega MI\_1 \end{array} \tag{10}$$

As shown in Figure 10, the wireless charging system draws power directly from the 48 V DC nano-grid. The DC input voltage is then inverted such that a high-frequency voltage supplies the transmitter resonant circuit, which is also called the primary circuit, and the current produces a magnetic field that links to the secondary coil. The high-frequency voltage induced in the secondary circuit is then rectified, and its value is regulated via a DC/DC converter to be supplied correctly to the battery.

#### *6.2. Variable Frequency for Misalignment Tolerance*

Since the cyclist has flexibility in parking the bike, the coupling between the two coils can vary. Moreover, depending on the state of charge of the battery, the loading condition also changes during the charging process. On top of that, the value of the circuit components can change because of an increase in temperature or degradation with time. All these factors make the resonant frequency of the system vary from the designed value. To overcome this problem, the inverter's operating frequency is set by the inner control loop that tracks the natural resonant frequency of the system. For this reason, it is called auto-resonant frequency control. It sets the soft switching of the inverter by predicting the primary current zero-crossing depending on the current slope. This control circuit is simple, analog, fast, and it automatically adapts the operating frequency to the optimum value in a few periods.

#### *6.3. Communication with the Bike and Foreign Object Identification*

The communication between the e-bike and the charging station allows the start-up, shut-down, and foreign object detection. It is realized through backscatter modulation in the power line through amplitude shift keying (ASK) modulation. In this way, the information can be reflected by one side to the other by modulating the voltage of the resonant circuit between a low and high value with a frequency lower than the operating one (about 1 kHz). To send information via ASK from the e-bike to the station, a resistor in series with a switch connected across the output DC voltage is used. When the switch is in an open or close position, ASK assumes a high or low value (due to a voltage drop across the resistor), respectively. These binary values can be organized in sets of bits that form the messages of the communication system. Once a message is received, it is demodulated and interpreted for execution.

The wireless charging system has to be able to detect and stop the charging process if a foreign object is placed on the surface of the transmitter coil. This is because the magnetic field can potentially heat the object through eddy currents, and could lead to fire or injury. This is because either the charging process is not started or the ongoing charging process is stopped if a foreign object is detected. The first scenario is avoided because the charging begins only if proper communication messages are sent by the e-bike. On the other hand, e fficiency measurements of the power transfer can indicate if a foreign object is receiving part of the transferred power, which results in a lower e fficiency level than expected.
