[ Paper Review ] Parameters for degradation diagnosis

Effective and practical parameters of electrochemical Li-ion battery models for degradation diagnosis 

• Journal : Energy Storage’ 21 (IF 8.9)
• 32 citations

 

[Summary]

1. params 중에는 fixed params, dynamic params가 있음
   1. 15개의 dynamic params는 aging에 따라 변함
2. 15개의 params에 대해 GA 알고리즘으로 파라미터 최적화
3. 15개의 params 중에 battery degradation diagnosis하는 파라미터 선택
   1. 3가지 수렴 조건 , CI (Confidence Interval) 사용
4. 결과 : degradation diagnosis 파라미터
   1. cathode particle surface area
   2. porosity
   3. stoichiometry limits

+) 일반화를 위해 3개의 profile에서 실행

 

🔗 figure에 대한 상세설명 

---

1. Problem Statement

• SOH(State of Health) is not enough for diagnosis of the battery’s internal state
  • SOH is defined using capacity, internal resistance (the ratio of the present capacity to initial one / the present equivalent series resistance)
  • Capacity, internal resistance are insufficient to describe the current state of battery
• This paper proposes a practical method for identifying and selecting effective P2D model parameters that change significantly as a battery ages

⇒ Main theme : Identify more effective indicator(params) for accurate degradation diagnosis

 

 

2. Contribution

• P2D model parameters of a Li-ion battery were identified and their confidence intervals were analyzed
• Various operation profiles were used to avoid overfitting problems during parameter identification (generalization)
• Define aging parameters for battery degradation diagnosis

 

 

3. Methodology

• Aim1 : Parameter Optimization using Genetic Algorithm
• Aim2 : Select ‘degradation diagnosis params’ among aging params

1️⃣ Select 15 dynamic parameters (vary with battery aging) in P2D model

 

 

2️⃣ Parameter Optimization using Genetic Algorithm

1. Initial population generated randomly according to the designated search range
2. Select parent chromosome by the roulette wheel algorithm
3. Apply genetic algorithm - uniform mutation, tuning mutation
4. Calculate fitness function for parameter combination (parent chromosomes)

Parameter setting for genetic algorithm

Fitness Function of Genetic Algorithm

 

 

3️⃣ Select ‘degradation diagnosis params’ among aging params (using convergence criteria & CI )

 

[ criteria ]

• The parameters should converge to similar values for any initial condition and sample trajectory in the P2D model
• The parameters at BOL and EOL should converge to distinguishable values
• Parameter variation from BOL to EOL should be physically appropriate

 

4️⃣ Result : params for LIB degradation diagnosis

 

 

4. Experiment Result Analysis

 

4.1 Experimental Setup

• Experiment with 5 cells :
  • cell #1 - BOL(beginning-of-life battery)
  • rest - EOL (end-of-life battery)

 

 

• Profile setting : CC-CV / HPPC / Driving cycle
  • purpose
    • Avoid overfitting issues during parameter identification
    • Identify parameters that can reflect the characteristics of the battery under various operating conditions

• CC-CV : observe battery response at various charge/discharge rates
• HPPC : observe pulse response at various SOC and current levels
• Driving cycle : simulate real electric vehicle usage patterns

 

 

 

4.2 Result Analysis

 

 

4.2.1 Parameter identification results

 

1-1. obtained optimization paramter value using GA

Mean values of parameter identified 10 times of GA

95% CI for each params

 

 

 

 

1-2. voltage error

To verify the reliability of the optimized parameter values

• Identified params show small voltage errors compared to existing result → Identified params using GA is effective
• CC-CV voltage error bigger than others
  1. OCV fitting error
  2. CC-CV profile has only one continuous dataset

 

 

4.2.2 Aging parameter selection

 

Select effective and practical aging parameters for battery degradation diagnosis

 

[ Criteria ]

• The parameters should converge to similar values for any initial condition and sample trajectory in the P2D model
• The parameters at BOL and EOL should converge to distinguishable values
• Parameter variation from BOL to EOL should be physically appropriate

[ Confidence Interval (using Table.8, Fig.10) ]

95% CI for each params

 

 

 

 

• cathode particle suface area (a_p)
  • BOL과 EOL에서 구별 가능한 값으로 converge
  • EOL에서 값이 감소하는 경향을 보임 → 반복된 충방전으로 인한 particle isolation, active dissolution ans cracking of the active marterial, SEI layer formatio 등으로 설명 가능
• stoichiometry limits (θmaxp, θmaxn, θminp, θminn)
  • BOL과 EOL에서 구별 가능한 값으로 converge
  • Theoretically, battery aging에 따른 reduction of active material → 각 전극에서 less lithiated state
• porosity (εp, εs, εn)
  • 상대적으로 큰 CI을 가지지만, EOL에서 값이 감소하는 경향을 보임
  • corresponding params converge to similar values for any initial condition
  • Theoretically, porosity may play a role as diagnosing the degradation status of battery

 

 

5. Conclusion

Define aging parameters for battery degradation diagnosis :

 

[ Limitations and weakness ]

• This study deals with BOL and EOL cells only
  • The middle of life (MOL) cells were not available
  • Difficult to elaborately predict the intermediate stage of degradation
• Large confidence intervals for anode stoichiometry
  • OCP of the anode has a potential plateau(평탄 구간) over a wide range
    → response of potential to lithium concentration change is very weak
  • difficult to make the stoichiometry limits converge using the proposed scheme that considers the output voltage error to be a fitness functionLarge confidence intervals for anode stoichiometry

• Too small number of samples
  • Identification performed 10 times for each cell using GA
  • CI is calculated using the t-distribution ( less than 30 )
  • The more parameter samples identified, the narrower the confidence interval
• Validation limits of identified parameters
  • validate thorugh comparison with measured real voltage outputs
  • For more accurate validation, true parameters should be directly compared (True parameters are measured with non-destructive and non-invasive test methods (EQCM-D) or X-ray tomography)

 

---

 

Background Knowledge

• P2D model

 

The governing equations of the P2D model (5 equations) :

 

• Genetic Algorithm
  • Roulette wheel 방식으로 부모 염색체 선정 → fitness가 높을수록 부모로 선택될 확률이 높아지는 방식
  • fitness가 높은 염색체가 더 높은 확률로 선택되도록 선택 영역 조정 가능

 

---

❓질의 응답 

 

Q. 3개의 profile으로 실험한 이유

→ To identify selected aging parameters distinctly correlated with the battery degradation states

 

Q. BOL , EOL 구분해서 실험한 이유

→ 배터리의 Beginning-of-Life (BOL) 및 End-of-Life (EOL) 시에 뚜렷한 상이한 값으로 수렴하는 특정 파라미터가 aging parameter로 선택

→ 배터리 열화를 진단하기 때문에, 다양한 aging의 배터리 상태에서 실험하는 것이 적절함

 

•