Evaluation of lightweight battery management system with field test of electric bus in campus transit system

Received Apr 10, 2020 Revised Jun 6, 2020 Accepted Jun 18, 2020 A battery management system is a crucial part of a battery-powered electric vehicle, which functions as a monitoring system, state estimation, and protection for the vehicle. Among these functions, the state estimation, i.e., state of charge and remaining battery life estimation, is widely researched in order to find an accuracy estimation methodology. Most of the recent researches are based on the study of the battery cell level and the complex algorithm. In practice, there is a statement that the method should be simple and robust. Therefore, this research work is focused on the study of lightweight methodology for state estimation based on the battery pack. The discrete Coulomb counting method and the data-driven approach, based on the Palmgren-Miner method, are proposed for the estimation of the state of charge and remaining battery life, respectively. The proposed methods are evaluated through a battery-powered electric bus under real scenario-based circumstances in the campus transit system. In addition, the battery life-cycle cost analysis is also investigated. The tested bus has currently been in operation in the transit system for more than one year.


INTRODUCTION
Moving toward a new direction of transportation, in terms of sustainability and low impact to the environment, electric vehicles (EVs) are the trend and drive the change in the transportation market in recent years. One of the key factors of the development in EVs is the impressive achievement in battery technology. Among many types of batteries, the Lithium-ion battery is dominant in EV application nowadays regarding high energy density, high power density, and longer battery life [1]. However, this battery type is sensitive. It requires an additional apparatus, i.e., a battery management system (BMS), to ensure the reliability of the battery system.
The BMS's functionalities of a Lithium-ion battery are a monitoring system, battery state estimation, battery cell balancing, and a protection system [2]. Among these functionalities, battery state estimation, i.e., battery state of charge (SoC) and battery state of health (SoH), are important parameters of battery packs, which is required by the driver [3,4]. To estimate the battery SoC, various calculation approaches are classified as follows. The direct measurement method-open-circuit voltage [5], the booking keeping method-coulomb counting [6], the adaptive method-Kalman filter [7], and the hybrid method-the combination of Coulomb counting and a Kalman filter [8]. Based on these four methods, [9] it has been investigated and summarized that the hybrid method is the most potent method for SoC estimation. Consequently, this hybrid method results in a complex algorithm. On the other hand,  In addition to Figure 1, the sensor positions are pointed out. The voltage is measured in every battery cell and the total voltage of the battery pack. There is one current sensor, which measures the current of the battery pack. There are two temperature sensors. The sensors are equipped at the second battery module of each battery string. All the measured data were logged with a sampling time at 1 sec, which is sent to the reference BMS. The reference BMS is in charge of data acquisition, battery protection, and battery SoC estimation. The value of battery SoC estimation from reference BMS is used as a benchmark for proposed SoC estimation, which is calculated by the Coulomb counting method. Remarks, the reference BMS is provided and configured by the battery supplier.
In the part of proposed BMS, the battery voltage, current, and temperature are received through reference BMS in a real-time via controller area network (CAN) communication protocol with the four decimal places data resolution. In the processing unit, the SoC estimation is examined in an online process. Meanwhile, battery life estimation is observed in an offline process. Regarding the monitoring system, there are two opportunities. The diagnosis computer is designed for engineering purposes, and the user display is for the bus driver. Both are able to observe the data in real-time.

PROPOSED MATHEMATICAL DESCRIPTION OF BATTERY SOC AND REMAINING BATTERY LIFE ESTIMATION
To determine the proposed battery SoC and the remaining battery life estimation, there are three discussed sections: i.e., the mathematical description of the SoC method, the determination of the initial SoC, and the battery damage model based on PM. Moreover, the complete working process of the SoC estimation and remaining battery life estimation is summarized.

Proposed battery SoC estimation method
As mentioned before, the SoC estimation method in this application requires a lightweight algorithm and operated under discontinuous time. After investigation, the discrete Coulomb counting method is proposed in this paper. To understand the mathematical description of this method, the relationship between SoC and the discharge capacity (it) in Figure 2 needs to be discussed. In the figure, a linear function is used for describing the between SoC and discharge capacity, when the time interval (t) is considered as a small difference. As a result, a small-time interval of SoC (SoCk) and discharge capacity (itk) based on discrete-time are changed in linear characteristics. The change of SoC in a function of discharge capacity during the time interval (SoCk) and the small change of discharge capacity (itk) are included in the effect of battery chemistry and electrical non-idealities. Therefore, the proposed discrete battery SoC estimation equation based on Coulomb counting is given as where SoCk and SoCk-1 are batteries state at different instants of time, k is an index of discrete-time, ik is the battery current for both charging and discharging processes, t is calculated step-time interval, and  is the battery round trip efficiency. As a remark, the battery round trip efficiency in (1) is able to adjust for the correction of SoC estimation error due to battery aging. In a short conclusion, the proposed discrete battery SoC estimation algorithm described is based on a linear function. This linear character is perfectly harmonized with the Lithium-ion battery character. However, in some battery types, the linear function cannot be used for relating the battery SoC and discharge capacity.

Initial SoC determination
The initial battery SoC is a key parameter to define a preliminary state of battery SoC. If the determination of initial SoC does not correct, the SoC estimation process will result in an error as well [26]. Typically, the initial SoC can be determined by three methods [6], i.e., charging voltage method, discharging voltage method, and open-circuit voltage method. However, these three methods are typically used in a laboratory investigation.
In practicality, the initial SoC should be directly analyzed when the battery is fully charged. Because at this point, the charging current will be zero. Thus, it can be assured that the initial SoC, or rather a battery SoC, is 100%. In addition to this method, the open-circuit voltage method is also applied. When the charging current is zero, the open-circuit voltage is measured and compared with the information in Figure 3. Regarding the driving cycle of the electric bus in the field test, as mentioned before, the bus driver will charge the battery during the 1-hour breaks. Within this period, the battery will be fully charged. Therefore, the proposed initial SoC determination is flawlessly matched with the schedule of the bus driver.

Proposed battery damage model based on the PM method
Since the charge cycle and the discharge cycle of the electric bus are non-uniform, then the datadriven approach is taken into account in order to estimate the remaining battery life. To accomplish this, the accumulation of system fatigue analysis based on PM is considered. Normally, the PM is used for the analysis of mechanical systems fatigue. However, this method can be effectively applied for the battery as well. Applying the PM method for remaining battery life estimation, the accumulation of battery damage (Dcu) is formulated based on the relationship between battery life cycle (L) and battery DoD. As a result, it can be written as (2).
where n is the driving cycle of the electric vehicle.
To assign the battery life end of life, this paper assumes that the battery will end at 80% of rated capacity. According to this percentage, the battery capacity linearly fades away regarding the accumulation of damage [27]. Since the damage accumulation based on PM is a linear function, the remaining battery life (Qmax) can be modeled in the mathematic equation, as in (3).
where df is the damage factor. Finding the Qmax, the relationship between battery life cycle and battery DoD is needed. From the information in Table 2, the relationship in the term of the mathematic equation has to be modeled. For the Lithium-ion battery, an exponential function [28] or a power function [29] can be used for the relationship expression. To select the best curve fitting, the R-squared index is analyzed by the curve fitting tool in MATLAB. The R-squared is a regression analysis based on the statistical measurement of how close the data are to the fitted curve. The best fit curve will result in a high value of R-squared. The R-squared results of the exponential function and the power function are 99.89% and 98.29%, respectively. Therefore, the exponential function is selected. To find the coefficient of the exponential function, the information in Table 2 is processed through the curve fitting tool in MATLAB. As a result, the mathematic equation of the battery life cycle is written in (4) The exponential equations in (4)-(6) are plotted and displayed in Figure 4. It can be obvious that the battery DoD and the temperature are the parameters to indicate the battery life cycle. Additionally, if the temperature does not match with the model in (4)-(6), the interpolation technique can be applied.

Working process of proposed onboard BMS
According to the SoC estimation method, the working process of the onboard BMS for the electric shuttle bus can be concluded in an algorithm flow chart, as displayed in Figure 5. The algorithm is divided into an initial SoC determination part and the remaining battery life estimation part. The initial algorithm part is the first SoC estimation process. The first step is measuring terminal voltage, current, and temperature of the battery, and the SoC value was estimated by reading historical data of the used battery is retrieved from associated memory. The initial SoC has to be determined every time the bus driver starts the bus. The following step is the initial part. It contains a process that monitoring display and SoC estimation. The discharge current can be estimated using the proposed SoC algorithm, which is an online process. It is active when the battery is operated. Conversely, battery life estimation is an offline process. It is active when the bus driver finishes the driving cycle or, rather, when the battery is charged. This process is determined by the battery damage model based on the PM method. Note that the proposed working process is developed for the battery pack. Then, the monitored parameters for this process are voltage, current, and temperature of the battery pack.

LIFE-CYCLE COST ANALYSIS
In this research, the life-cycle cost of the battery in the electric bus transit is based on the economic modeling analysis. There are four main life-cycle cost types in the analysis, the capacity costs, the replacement costs, the operating and maintenance costs, and the end-of-life cost [30]. The annualized life-cycle costs (clc) of a battery in the electric bus transit are formulated as follows: where ccap weights the capacity costs, which are associated with the acquisition and installation of the battery system, crep is the replacement costs due to degeneration, co&m is the operating and maintenance costs during the battery's lifetime, and ceol computed the disposal and recycling costs at the end-of-life of the battery. The capital costs (ccap) in (7) can comprise acquisition and installation costs for a battery system. Thus, the total capital costs are calculated as follows: where cbattery is the battery cost, and ccharger is the charger cost. The technology replacement costs refer to the essential replacement of battery due to degeneration. The calculation of annualized technology replacement costs (crep) can be formulated as follows: where fr is the replacement function, which has a value of 0 or 1 indicating battery replacement, Ncycle is the battery life as a number of deep cycles, drate is a yearly discount rate, and T is the time period for the life cycle analysis. The end-of-life battery life in terms of the total driving cycle is calculated based on the proposed battery life estimation method, according to the result in Section 5.3. Note that the estimated costs of a battery are dependent on the experience rates and the archetypal sigmoid function (S-curve) of energy storage technologies under a hundred percent market share assumed for each technology.

FIELD TEST RESULTS AND ALGORITHM VALIDATION
In this section, the proposed algorithms in section 3 are applied and validated in the field test. The SoC estimation results and battery life evaluation for the battery pack and battery life-cycle costs analysis are shown in detail. The electric bus and Lithium-ion battery pack in the field test are shown in Figure 6(a). The communication interface converts the battery state information voltage, current, and temperature into the CAN bus signal to be displayed on the user's screen and monitoring system, as shown in Figure 6(b). The test condition is under a real scenario in the campus transits system. Every day, the operation service time is from 7:00 am to 10:00 pm, 15 hours per day. Each driving cycle takes around 3 hours. The bus operator has a 1-hour break after each driving cycle. The tested bus is operated three driving cycles per day. The battery of the electric bus is charged three times per day.

Validation of the SoC estimation method
The field test results of the bus service operate, and the validation of proposed SoC estimation are shown in Figure 7. Figure 7(a) to Figure 7(c) shows the velocity, terminal battery voltage, and high and low temperature of the battery cell. From the results in Figure 7(a), one route of the service takes around 33 min. The average speed is approximately 14.9 km/h. In Figure 7(b), the terminal battery voltage is in the operating range 60-87.6 V. The cell temperature in the battery pack is maintained between 25-26 o C, under maximum temperature (35 o C) operation, as shown in Figure 7(c). It can be concluded that the Lithium-ion battery is functioning properly. Figure 7(d) to Figure 7(f) shows the battery current and battery SoC during the field test. The maximum current is around 255 A, which is lower than the maximum instantaneous discharge current (400 A), as shown in Figure 7(d). The proposed SoC estimation result is shown in Figure 7(e). It can be seen that the result of the SoC of the proposed method is shown along with the SoC from the reference BMS. The corresponding error of battery SoC is shown in Figure 7(f). The percentage difference between the reference and the proposed estimated SoC results remains below 1. Referring to the reference, the SoC estimation method achieves good results. The reference BMS for the SoC validation is provided by the battery manufacturer. Regarding the sampling time, the data logger is randomly delivered data between 1 sec and 2 sec. Even with this inconsistency, the proposed SoC algorithm is able to handle it. The sampling time can be substituted directly in the calculation formulation. This is one of the reasons why the proposed SoC estimation method functioned properly. Currently, the battery round trip efficiency in SoC estimation is set to 0.88 constant value, in which this value is received from the battery manufacturer.

Remaining battery life estimation
As mentioned, this paper proposes a data-driven approach with the accumulative damage based on the PM method to estimate the remaining battery life. Resulting in accumulative damage, battery temperature, and battery DoD all being input parameters, see (2), and (4)- (6). and the consequent faded battery capacity is estimated through (3). As an experiment, Figure 8 shows the data collection of the remaining battery life estimation in which the electric bus has been operating in the transit system for one year, about 1,000 cycles. It can be seen that the faded battery capacity data from the field test is marked with the dotted line.
Based on the data, the degree of one polynomial is applied to fit the estimation curve, and the faded battery capacity equation (Qn) is rolled out in (10). With this field test data, the end of battery life is at 2,888 cycles. In this research, the linear assumption is used for the estimation, when the end of life is set to 80% of initial battery capacity. Based on the current driving cycle, the battery will be ended after 2,228 cycles. According to this driving cycle, it can be implied for 2.6 years. Thus, the validation of (10) or battery life estimation and the knee point of battery degradation aging need probably 3 years for the investigation. However, it is noteworthy to mention that the linear assumption is a good choice for battery life estimation when the observation is started with a new battery.

Life-cycle cost results
This section presents an application of the life-cycle cost analysis. The objective of the study is to compare the annual cost of the Lithium-ion battery with a battery management system and the original Lead-acid battery. The cost parameters for this evaluation are defined based on the cost in Table 3. This information is provided by the operator of the transit system.
The operating and maintenance costs include electricity consumption, general repairs, and regular maintenance costs. The replacement costs include the periodic replacement of batteries and the maintenance of the charging devices. The disposal and recycling costs are based on the buyer's market. All costs are considered in the current price data of the USD, based on the average price in 2018. The yearly discount rate was assumed to be 3% of the initial costs. Finally, the maximum calendar life was estimated to be 2.5 years and 0.677 years (8 months) for the Lithium-ion and the Lead-acid batteries, respectively. After the calculation, according to (7), Figure 9 presents the life-cycle cost analysis for a battery-powered transit bus, using driving cycles from the field test. Figure 9(a) shows the development of the total costs in individual years of life-cycle during the use of the Lead-acid battery and the Lithium-ion battery for a concession period longer than five years. Considering the maximum calendar life, the results show that the Lithium-ion battery has the highest capital cost over the Lead-acid battery in the first year. However, the total costs in other years were decreased. It can be concluded that the replacement cost of the battery, naturally, has a significant effect on the life-cycle costs. For the purpose of the battery life-cycle costs, a relation of 5 years was used, based on which, it is possible to declare that the life-cycle costs are $3,182.38/year and $5,371.83/year for the Lithium-ion battery and Lead-acid battery, respectively. The life-cycle costs of a Lead-acid battery showed that the total costs are higher than the Lithium-ion battery, as shown in Figure 9(b). The difference in the battery life-cycle costs between analyzed alternatives is the sum of $2,189.45/year. Based on the parameter assumption, the payback has been set to 2 years.

CONCLUSION
This research work is aimed to develop and evaluate the lightweight algorithms for BMS. The proposed algorithms are the discrete Coulomb counting method, and the accumulative damage should be based on PM, for the estimation of battery state of charge and remaining battery life, respectively. For the economic aspect, the life cycle cost analysis is also taken into account in this paper. This information will answer the common question from the transit system operator.
The discrete Coulomb counting method is formulated based on linear function, which is proper for the character of the Lithium-ion battery. Furthermore, the battery round trip efficiency is also included in the formulation in order to correct the error, which can be caused by battery aging. The remaining battery life is estimated based on the relationship between battery life and battery DoD. Then, the accumulative damage is collected, once the battery is charged. The proposed PM method is developed based on the data-driven approach to overcome the complexity of the model-based approach, for which the internal battery parameters have to be measured. In addition, the PM method is appropriate for the non-uniform charge and discharge cycle of the electric vehicle. To complete the BMS, the protection system is obviously implemented. The observed parameters are cut-off voltage, maximum voltage, and temperature.
The battery-powered electric bus under the real operation of the Chiang Mai University transit system is used for the evaluation. The result from the field test proves that the proposed discrete Coulomb counting method works correctly. The comparative error with the benchmark BMS is 0.1%. The result of the remaining battery life estimation is that the battery should be replaced after 2,888 cycles. Presently, the battery end-of-life is set to 80% of the initial capacity. With this range, the remaining capacity of a Lithium-ion battery is linearly faded. Lastly, the life cycle cost analysis of a Lithium-ion battery with a battery management system, and the original Lead-acid battery is calculated. It is found that the use of a Lithium-ion battery can reduce the total cost by 20%. With this amount of cost reduction, it would attract the transit system operator to replace the Lead-acid battery with the Lithium-ion battery.