20141031：動力電池與能源管理系統之可靠度工程國際論壇

hlperng

動力電池與能源管理系統之可靠度工程國際論壇
International Forum on Reliability Engineering Applied to Power Battery

隨著動力電池與能源管理科技發展日新月異，動力電池與能源管理的問題向來為產、學、官、研等各界重視的議題，為提昇國內各大專院校在此研究議題之學術水準與教學品質，以及精進產業界實務技術與競爭力，特舉辦動力電池與能源管理系統之可靠度工程國際論壇。

一、日期：103年10月31日(星期五)
二、時間：08:30 - 17:00
三、地點：明志科技大學圖資大樓八樓會議廳
聯絡人：蔡雅萍小姐
　　　　聯絡電話：(02)2908-9899分機4503、傳真電話：(02)2908-5247
　　　　E-mail：ya@mail.mcut.edu.tw

時間	主題	演講者	主持人
08:50 - 09:00	開幕致詞	明志科技大學 劉祖華校長
09:00 - 10:00	Lessons Learned from the 787 Dreamliner on Lithion-Ion Batterry Reliability	美國馬里蘭大學 Prof. Michael G. Pecht	明志科技大學 黃世欽教授
10:00 - 10:40	Principle of Lithium-Ion Power Battery	動能科技公司 詹益村博士	明志科技大學 楊純誠教授
10:40 - 11:10	休息
11:10 - 12:10	Studies on Life, Life Distribution and Quantitative Reliability of Li-Ion Batteries	台灣大學 吳文方教授	明志科技大學 梁晶煒教授
12:10 - 13:30	午餐
13:30 - 14:10	PSM Modeling for Li-Ion Batteries and Remaining Useful Life Estimation	明志科技大學 黃世欽教授 Dr. Kuo-Hsin Tseng	明志科技大學 章哲寰教授
14:10 - 14:50	A Study on the Characteristics of Li-Ion Batteries and the Engergy Management Systems	明志科技大學 楊岳儒教授 吳啟耀教授	明志科技大學 楊純誠教授
14:50 - 15:10	休息
15:10 - 16:30	The Analyses of Lithion-Ion Battery on Failure and Safety Testing	明志科技大學 楊純誠教授	元智大學 陳永樹教授
16:30 - 16:50	討論與閉幕致詞		明志科技大學 黃世欽教授

hlperng

明志科技大學(MCUT)

梁院長：開幕詞
SOC

Keynote Speech I by Prof. Michael Pecht from CALCE of UMChair prof., founder of CALCE
SOH
RUP = remaining useful performance
Catastrophic failure = thermal runaway.
Reliabiity = safety concern
Safety prediction was flawed.  Boeing's reliability.

常數失效率假設的困擾，電池是組件不是系統。
必須試驗多久才能瞭解電池的可靠度。

2013-03-11 Boeing permanent fix.
Tighter voltage control.  Electrical insulator wrapped around each cell.  Thermal insulation installed on all side of the pack.  Wire sleeves more heat resistance.  Bottom holes to allow moisture.  
To this day, no root cause of failure has been identified.  

MD approach Fustion prognostic approach.  RUL estimation.
Possible cause --> dendrite growth, low temperature effect?
Possible cause --> metallic particle defect. manufacturing process.
2014 Boeing battery failure at Narita International Airport in Tokyo.  Smoke from venting of a battery cell.  

Program, excellent presentation.  More gain from the talk.  

1984, $6M, batteries.  Means to reduce emission.  

Rechargeable battery.  

1980, Li-ion start to use.  Battery technology.

Five years ago, Sony battery fires atAirport conference. Last year again.  

Electric car fires.  Tesla, Chevy. Boeing 787 dreamliner.  

Boston Battery,

January 16, 2013.  Kagawa Battery incident.  

World wide grounding in Jan 2013.  $1M per day per plane.  Boeing’s engineering design processs.  

Complex supply chain.

1970 sealed nickel cadmium battery, 1990sealed lead acid battery.

2003 special request for li-ionbattery.  

Main battery,  stored at cathode(陰極)，放電時，電子由陰極流向陽極(anode)。

3.7 volts per cell.  

GM research lab, battery explosion, 5people, A123 battery, $55M

Hot,

Degradation mechanisms,  

Capacity fade curve.  Max Ah that can be withdrawn from a fullycharged battery.  

Capacity vs. Discharge cycle. Over 600dischareg decrease signicantly.  

Systems issue, 討厭的精靈，如何約束？

Discharge and temperature effects onbattery capacity.  

After driving the 1.2 miles to daycare, theindicated range had plummeted from 31 down to 22 miles,  Nine miles of range dropped in just 1.2 milesof

Nisson leaf

State of charge (SOC)

SOC = (Qmax – f(t))/Qmax

State of health (SOH) = Qmax (c) / Qmax (1)

Remaining Useful Performance prediction(RUP)

MTBF(e) = ta/r

Failure rate is constant?

Constant failure rate assumption

Data analysis,

Mahalanobis based detection methodology

CALCE heath monitoring approach for anomaly(safety) detection.  

MD,C=correlateon matrix, parameter anomaly detection.  

38 min prior to anomaly.  

Fusion prognostics approach

Electrical field failures,

Lessons Learned

-    Needto monitor the actual usage condtions

-    Becareful on the assumptions of failure characteristics

-    Needto monitor key parameters

-    In-sitemonitoring for anomaly detection, diagnostics and prognostics, provides valuefor reliability and safety.  

Neil tests,

Malasia Li-ion battery the possible cause?

Q: take flight, take 787 or not.  

A: Air bus used li-ion on 380.  Cacel because Boeing case.  

Q: True cause of the 787. Shut down batteryearly.  Air bus use li-ion?

A:: Battery density.  At some stage concern with li-ion.  European requirements.  Boeing take risk use tranditional battery,not to use it.  Change regulateon need recertifythe airplane.  Air worthy processs.  

Q: Satellite use li-ion.  Outgassing effects.  

A: whole range depend on many factors.  Battery tested hundred cycles.  Transportation EV, 1000 cycles is the target.800 is not too bad.  Weight and space issues.  Money.

Q: SOC, SOH not direct measurement of thebattery.  They are not reliable.

A: many approach on Laptop and NB are not accuracy.  Get improvement. Cellphone far behind.  Normal.

A: Life tested over 1000 cycles.  

Q: Battery in aircraft

A: See problems can we stop it.  We cannot stop it.  Thermal runaway means it is too late. Howearly can we stop it or prevent it.  Itis a risk management problem.  Fire exitngesher.  Every effort is just for tuture useonly.  

hlperng

Prof.  Wen-Fang Wu
Battery Management
What is Relibility? Strong, Robust design.  related to product's life.  MTBF, failure rate.  
Life is a random variable.  MIL-HDBK, FIDES, address reliability issue.
fit = failure in time.  
λ = λ_b π_V π_T ...
Failure time = time to faillure,
degradation curve, degradation process and damage accumulation.  
Failure defintion? = reachs 80 % of SOH
Degradation reality = Systematic trend in time + systematic variation + random variation
evolution in time → stochastic processLife prediction (算命), RUL ⇒ performance prognostic (預後), RUP
POF = Physics of Failure
causes of degradation, related to physical theories.  
Three MS theses on Electric Vehicle studies.  1. Modeling of Capacity Fading and Investigation of Maximum Capacity for Lithium-Ion Battery
Mechanism models (failure rate model, failure time based) vs. External characteristic models (failure strength model, failure number based)
Use dataset from NASA website.  
ambient temperature: 4 °C, 24 °C, 43 °C
Capacity fading = Q_LUse curve fitting technique to manipulate the dataset.  
Q_L = α(T, DOD_V) N_c^{β(T, DOD)}

Polynomial fit vs. inverse power fit vs. exponential fit

2. Life Prediction Modeling of Lithium-Ion Batteries based on Electrochemical Analysis

3. SOC Estimation of Lithium Battery Systems by Markov Chains

Diffusion Markov chain model, ← by number counting
Degradation model, → by time calcuation

RV expressed by statistics (first order, second order, third order, fourth order, and higher ...)

Capacity fading, capacity degradation
Output as reliability measure, should take more uncertainties into consideration.  

Input random variables ⇒ Process ⇒ Output random variables
Process is modeled by Function

Single Particle Model
diffusion equation, ion concentration, current density,
current density changed with time → degradation ⇒ degradation modelEV for UK BEV (battery electric vehicle)
ω = 1 charge every one day, ω =2 charge every two day.
ω = 1, MTBF = 778 cycles
ω = 2, MTBF = ?

discharging current is limited to 1 A.
capacity fading after 400 charge-discharge cycles


POF is important but may be complicated.
Reliability can be improved through technology innovation.
Quantitative reliability is based on consideration of randon failure time.
Quantitative reliability is not easy to be measured and obtainted directly.
Database of previous experience may be helpful.
Safety and reliability are always required, especially in transportation industries.
Q: predict remaining life or average life.  with possibility
A: mean trend, average failure.  Quantitative reliability. failure rate, single life.

瞎子摸象：
Markov ⇒ management,
Degradation time or time to failure ⇒ engineering,
diffusion or chemical process ⇒ physical or science

hlperng

Topic: Regression Models using Fully Discharge Voltage and Internal Resistence for RUL Estimation of Lithium-Ion Battery

Presented by Prof. Shyh-Chin Huang and Dr. Kuo-Hsin Tseng

Prof. Huang (Ph.D. Mechanical Engineering Purdue University
Dr. Tseng, do the simulation. (Ph.D. National Chung Hsing University)
Data provided by CALCE
Electronic devices, phone, NB, vehicle.  
Battery Prognostics and Health Management (PHM) becomes more important than ever.
Battery RUL prediction by:
1. Regression model
2. Regression parameters

Age related parameters:
N=charge/discharge cycle number, most frequently used.

Other possible parameters: Open circuit voltage, discharge voltage, battery internal resistance, charge/discharge time, etc.

Regression models
- Polyomials: mostly used single variable N, coefficients determined by LSE.
- Exponentials: single variable N, coefficients determined by LSE, D-S, MC, etc.

Vdis and R as aging parameters

Better RUL prediction

Dr. Tseng presentation
Batteries in our daily lives.  
Battery Failure = Life failing

Precise battery life estimation becomes important to the reliable operation of the energy supply systems.

Battery aging: environments (temperature), opertion (charging and discharging), indicators, method (EKF, EIS, Neuro-Fuzzy, neural networks)

Framework:
Four batteries, Three parameters: cycle number (N), internal resistence (R), voltage after fully discharged (Vdis)

Highlight: Compare, Different, New technologies, Accurately.

Test Content
LiCoO₂ battery, room temp (25 °C)

SOH (state of health) (%)
Battery failure: ISO 12405-2 (80 %), IPC-9701, fifth data exceed the threshold.
Aging parameters: cycle number (N), internal resistance (R), open circuit voltage (Voc)
Vdis is a better.
Correlation, N, R, Voc vs. SOH
Vdis and R are more adequate than N.

Filtering
R, Vdis vs. N, butterworth filter (first order, low-pass, cut-off frequency was 0.1)

Modeling:
Type: Polynomial and exponential, V & R
Method: LSE, Particle swarm optimizer (PSO)
Evaluation: R-square, Root mean squared error (RMSE)

Model I: N, Model II: V & R (poly), Model III: V &R (exp), Model II fit best.  

what's the relationship between N and V & R.
N is output of the battery.  detected by SOH (80 %)
V_dis and R are process parameters.  
SOH (N) = f (R, V_dis)SOH = f (N, R, V_dis)
N : trending parameters
R and V_dis: systematic parameter
Erms = random errors (samples)
inverse power (single item polynomial):
exponential:

Q: (Pecht) short questions. When u did computer work (2012). It is just one data experiment? Just one example.  Reproduce the process with Taiwan data (not CALCE data).  Three batteries. Use average data to check with models.  Trending data fit the model.  Comparison against the fourth one.  [Xing et al (2012)]
Model work best.  Model less well. Exponential (3 parameters vs. 4 parameters)  

Q: only have 3 samples, average trending. Sample size require.  two sets of data to decide the correlation of the model.  
A: PSO is powerful method.  
Q: Manufacturrer, 30 samples is minimum. Check normality.  
Q: (Pecht) Model tends to fit mid section of the curve.  Deviation is more inaccurante.  time trend and systematic bias are more rely on engineering.  Less samples can be used.  

hlperng

Topic: Principle of Lithium-Ion Power Battery
Presented by 詹益松博士，Dr. Yih-Song Jan
動能科技公司

Fault tolerant and fail safe.  
Factors affect battery performance: High rate discharge, stacking type, old winding type, new winding type.
Rapic charge no effect onf the charge efficiency (%).  
DSC measurement,
Above 200, 250 deg C, the battery will catch fire. Heat flow (W/g) vs. Temperature (deg C).  on Cathode side only.  
LiFePO4 doped with M⁺² (substituted for Fe⁺²)
Coating carbon on LiFePO₄ to increase the electric conductitivity.  
Q: chemical, Why LiMn₂O₂ is better. New winding type reasons and disadvantage.
A: LiMn2O2 is better.  Winding type depend on the volume issue.
Q: oven 200 deg C, flash point or others.
A: p.29. DSC measurements of full charged (4.3 V) electrodes with electrolyte.  Perkin Elmer DSC7.  Structure issue.  
A: catch fire, heat go to electron, O₂ inside + heat → unstable electrons, chemistry.  Heat reaction release below 200 °C.  It is chemical problem.

hlperng

Topic: A Study on the Characteristics on Li-Ion Batteries and the Energy Management Systems
Prof. Chi-Yao Wu
Prof. Yueh-Ru Yang

Characteristic Tests of LiFePO₄ for Estimating SOH
Major tasks of BMS


Method of Estimating SOH
(1) Full-charge capacity estimation. need to take a long-term charging
(2) Internal Impedance estimates

Cycle life vs. Temperature
Cycle life vs. discharge rate
Q_loss = B exp(-E_a/RT) A_h^z

Conclusion:Full charge capacity (FCC) estimations need to take a long-term charging and discharging test.
Internal impedance estimation may cause large error due to the effect of environmental impedance.
How to develop a fast and accurate SOH estimation method?  

Q: 30 min rest, why?
A: 3 min to 1 hour voltage difference is small.  
Q: (Wu) charaterisitics test.  curve phenomena before ripple. relation of occuring time with discharge or charge.   
A: set overcharge voltage.  
Q: chart display issues.

Battery Management System
Presented by Yueh-Ru Yang
BMS
Why battery pack need BMS?
Why cell string need balance?

BMS = Battery control unit + BMU#1 + BMU#2 + ... + Battery Charger
BMU = Cell Stack#
The failure of a battery pack is closely related to the battery management system.
Use CAN bus to communicate the battery charge.  
The internal impedance and charge capacity of each individual cell in a cell string are not uniform.  

Test Battery: A123 ANR26650M1-B Cell
Charge for Multi-cell LiFePO₄ battery

The failure of a cell closely related to the imbalance of cells.
The imbalance is caused by the differential of internal impedance and charge capacity of each cell.  
THe imbalance can short the cycle life of cell string.
To protect the cell and prolong the cycle life, BMS needs to do equalizing charge.  

Dissipative voltage equalizer
The charge current for a Li-ion cell is 0.3 C.  
36 V/10 Ah battery pack, the charge current is 3 A.  
Power loss = 4.2 (V) x 3 (A) x 9 = 113.4 W > 50 W rated power.  


Non-dissipative voltage equalizer
1. charge shuttling method and fliying capacitor balances the charge of series-connected cells.
2. Transfer overcharge energy with inductors.
3. Using switched flyback transformer
4. Using shared flyback transformer. When switch in turned ON, the energy from entire stack is transferred to the transformer.  When switch is turned OFF, the energy is mostly returned to the lower voltage cell.  
5. Using multiple flyback transformer.  The lower voltage cell is charged by other cells.  
6. Using multiple-flyback transformer.  THe lower voltage cell is charged by other cells.  

Battery management controller


BMS fault monitoring with the LTC6801

Problems and discussion
1. Regenerative braking can cause problems, because regenerative inrush current causes battery voltage to increse suddenly, possibly damages the cells.  
2. End-of-charging cell-balancing methods are useful only for the batteries that are fully charged between each use cycle.  
3. The batteries of hybrid electric vehicles are not maintained fully charged, resulting in unpredictable end-of-charge conditions to enact the balancing mechanism.
4. Imbalance results in the cycle life of cell stack less than a cell.
5. The design of BMS depends on the batteries and applications.  

Q: Problem 2 justisfication.  
Q: Use or going to use BM controller in the research and what's result.  
A: Use LTC6802, performanc is better but more expensive. It monitor cell voltage, cell temperature.
Q: charge and discharge curve.  
A: Battery must be matched before use the BMS.  
Q: Reguler. 2nd is worst than 1st.  
A: charge can be removed by controller.
Q: based on and see Chart: why a battery pack needs BMS (1/3).  Can improve cell B.  
A: use active cell?

hlperng

Topic: The Analysis of Lithium-Ion Battery of Failure and Safety Testing
Presented by Prof.  Chun-Chen Yang

Some modification on LiFePO4/C Composite Cathode Materials
LiFePO₄ Cathode is current trend
LiFePO₄ Battery (18650) Tesler (EV)

Need to improve electrical chemical properties.
most costs come from cathode.  
Industry is interested on easy, light, cheaper cathode materials.  
particle diameters: 100, 200, 300 nm  (up to 900 nm)

La Doping, Nb Doping,
Cl doping in LVP,
Na⁺ and Cl^- doping,
doping element will going to where, it does not know.  

Physical properties measured by Crystallinity (XRD), Residual carbon (mirco-Raman), Surface morphology (SEM), Doping element composition (EDX/mapping), Carbon coating layer (TEM), Particle size distribution (DLS), Electron conductivity (Auto-lab)
Bunch of experimental charts shown.

Comparison wih literature data.  



Conclusions
1. At 0.1C rate, at 700 °C, LiFe_0.99Nb_0.01PO₄/C shows the discharge capacity of ca. 152.76 mAh g^-1; at 1C rate, ca. 136.04 mAh/g.
2. At 650 °C, at 0.1C rate, LiFe(PO₄)_1-XCl_X/C with 3 % Cl show the discharge capacity of 152.57 mAh/g, at 10C rate, 122.52 mAh/g
3. It was found that doping Cl^- on LFP is more easily producing Fe₂P and Fe₃P impurity phases that are of doping Nb⁵⁺ on LFP.  It was also found that too much content of Li₃PO₄ cause decreasing the discharge copacity; on the contrary, a small amount of Fe₂P and Fe₃P will improve the discharge capacity.  
4. According to the CV, AC impedance results, it was discovered that doping Nb⁵⁺ can improve the electrochemical performance; in particular, the doping Cl^- site (M1 site) of LiFe_1-XNb_XPO₄/C material (i.e., optimazed at 1 % Nb), also doped with Cl_- anion on O^2- site of LiFe(PO₄)_1-XCl_X/C (i.e., optimazed at 3 % Cl).  
綠色能源電池研究中心 founded on 2012. 12. 18.

Q: why low rate only run 30 cycles.  
A: It takes 2 years to do the experiment.  300 cycles for most industry cases.
Q:
A: 100 cycles on most literatures  
Q: 650 °C and 700 °C, CV curve (test on room temperature)
台塑鋰鐵生產材料，能源危機，新能源，人類找到提供生活品質的交通工具。磷酸鋰鐵是否最佳，不一定。

學校400坪、8000萬，材料試驗費用低。

hlperng

綜合討論：

Q to Prof. Wu: Environmental impedance.
A: Environmental conditions.  accuracy of sensors.
Q: Internal impedance sense to temperature or not.
A: not sure.  DC resistance will increase with temperature.  
Q: which one is easier to measure?
A: can be tested by pulse.  
Q: resistance is more important parameter in the model.  

Q: to Pecht, why to measure
A: Because we can.  change of the dimension during the process.  Like Doctor, not direct measure. Battery with problems, cannot put. monitoring may get more inside.  
Q: to Yang.  Indirect measurement.  measure inside the cell without influence the reaction process. It is possible.  
A: (Yang) More early monitor the cell condition. Input/output cannot measure without break the operation.  
A: (Pecht) Cell for example, transparent plate, optical sensors inside.
A: (Pecht) Well control condition, real cell and pack.  drive li-ion to danger situation.  600 Ah to 3000 Ah, it is not good for li-ion.  The case Doctor put medicine may be used.  
A: Measurement technique improvement.  Eng. Prof. in mainland try to measure fluid dynamic with non-contact methodologies.  Optical measurement is an option.  Battery reliability issues depending on the battery capacity, and manufacture technology.  
Q: (Wu) reliability engineering not easy to explain.  Ask Pecht to give some ideas.  
A: Dr. Wu explained very well.  People don't like reliability engineer and like R&D engineer.  Reliability engineering is very complex.  It gives 20 different ways to choose.  From business, people goes to Pi factors.  Reliability engineering is complex, mechanical, chemical, statistical.  Boeing for example, not to test so many.  It is a huge money.  Chevy.  People won't to buy.  
Q: (Wu) Reliability engineering in Taiwan, CSQ.  A lot thing to do, but we don't know.  Buyer always ask how reliable.  But engineer do not know how to answer.
A: Hunday as example, car is bad. Consumer report is bottom.    10 years 100,000 mile warranty.  Korea knows put reliability into car.  Not to guaranty 10 years.  It is great exercise.  
Q: Car environments affect the components.  What factors will be effective.
A: material and cell point of view.  Key is uniform, batch by batch.  Materials go manufacturer.  Uniform and consistent.  EV 3C 100 M, same and consistent problems.  cannot judge, too few amount of EV on street right now.  Just like toys.  Battery must be made in consistent. same cell and same performance, then it is easy to control the reliability.  Always get inconsistent results.  Power shut down the results are different.  Uniformity of cell and pack.  
原料廠量不夠大，BMS不是最佳化，沒有一致性。EV資訊不夠。費一簇即可，充電時必須well controlled.  
A: Temperature effects are very important.  Low temperature battery performance.
A: Military spec.  battery every body can used is too expensive.  
A: Nission car have temperature problem in Arizon, it is very hot here.  Life of two years. Bring in as warranty.  cost 16 - 17,000 dollars.  It costs 1/3 of car's price.
Q: find algorithm.
A: Equations for physics.  One can put on the computer and solve it.  Coeficients but don't known how to measure.  Can find out any DNA for example.  It is time problems.  
A: (Wu) Presented based on experimental data. SOH as a output.  Dependant on variables.  SOH as related to some parameters.  It is semi-physics approach.  We need real physical approach too.  Engineering judgement.  It is not helpful. Explain the data. Prediction algorithm.  
A: (Pecht)  If company can make same batteries, statistical models would be fine.  We don't have consistent processes now.  So many different changes.  So Model-based approach is chosen recently.  A: (Yang) only known average value now.  Cannot break the system.  Data are under assumption.  Average value and deviation, need more sample size.  Poor engineering skill makes poor battery.  All are related.  Tesler (18650).  Not easy to control.  Most safe product now.  Need more cell, more tests.  Find out which parameter is most important.
Q: Put sensors inside the battery, which is the most important.  Temperature, put electronic surface.  local area only.  temperature distribution. don't known where is the most hot one.  
A: Down to - 20 °C everything is stop!