Why did LFP battery model fit got worse after optimizing the parameters?

05 September 2021 9 5K Report

I'm trying to characterize a 280Ah LFP battery model with pulse discharge test based on the method described at:

https://www.mathworks.com/company/newsletters/articles/battery-model-parameter-estimation-using-a-layered-technique-an-example-using-a-lithium-iron-phosphate-cell.html

The pulse discharge test was performed at 1% SOC decrements between 100-90%, 10%-0% SOC and 5% SOC decrement between 90%-10% SOC. The cell was rested for 2 hours after each pulse.

After optimization, the model fit the pulse discharge test result well, but when validated with a US06 drive cycle, the fit after optimization was worse than before at about 92% SOC.

The simulated terminal voltage was almost parallel to the experimantal at around 92% SOC, capturing the shape of the voltage response but not the absolute value. We suspected it had something to do with the OCV and added a gain to the equivalent circuit model, but the same issue is still happening.

I would really appreciate it if someone could point me to something I can look into to solve this issue. Thank you in advance.

Ioannis Samaras

Dear Wenlin Zhang ,

the parameters' errors in LFP battery model(s) are, very much, high(er) near the (hard-won) high SOC (90%, 100%) regime(s), including your case study, with your challenging value:

SOC=92%,

due to high(er) electrochemical (thermal, mechanical stress) inhomogeneity[1-3].

So, let me recommend to, re-make/re-start your troublesome case study using a more favorable ('peaceful') regime of SOC value(s), near 50%, accompanied with, very much, low(er) parameters' errors.

1. Electrochemical Thermal-Mechanical Modelling of Stress Inhomogeneity in Lithium-Ion Pouch Cells https://www.researchgate.net/publication/336246259_Electrochemical_Thermal-Mechanical_Modelling_of_Stress_Inhomogeneity_in_Lithium-Ion_Pouch_Cells

2. Stochastic many-particle model for LFP-electrodes https://air.unimi.it/retrieve/handle/2434/642548/1219640/CMAT_SDE_revised.pdf

3. Discrete-time modeling of Li-ion batteries with electrochemical overpotentials including diffusion https://www.researchgate.net/publication/351402096_Discrete-time_modeling_of_Li-ion_batteries_with_electrochemical_overpotentials_including_diffusion

Wenlin Zhang

Hi Ioannis Samaras, thanks for your response. Could you elaborate on what you mean by using a more favorable regime? SinceI'm doing characterization of the cell, wouldn't it need to cover the entire SOC range?

Ioannis Samaras

Hi Wenlin Zhang:

Yes, you need to cover the entire SOC range.

However, the parameters' errors in LFP-battery model(s) are, very much, low(er) around the half (50% +-30%) SOC ('favorable & peaceful') regime.

So, start your next case study using, at least-3-SOC value(s)[1], near 50%.

Then, 'goto 92' (%SOC); it will be more easy for you.

1. A proposal 'educational' list of (initial) SOC value(s): 35%, 50%, 65%.

Djef Brak

Hello,

I'd like to share my initial thoughts when reading your question. I've done some charaterization tests on LFP modules before with my previous job..

I was wondering if the operating conditions under which the parameters were initially estimated were the same as that of the drive cycle test? Temperature of the cells has an important influence on the battery parameters (same for the C-rate but to a lesser extent). Another issue could be SOC drift. Due to the flat V-OCV curve of LFP, the estimation of the SOC is more difficult compared to other battery chemistry. So perhaps the SOC estimate you are receiving is not correct. This can potentially be resolved by using the proper CCCV charging current and voltage cutoff provided by the supplier to reset the SOC estimation.

On the model side, do you use the entire relaxation period (the measurement points) in your battery parameter estimation model? If so, perhaps lowering the sample frequency during these moments could help the estimation as to give more weight to the areas where the voltage is changing strongly and/or current is being applied.

Just my thoughts, take them with many grains of salt please!

Cheers,

Wenlin Zhang

Thanks everyone for the input. Just wanted to say that I seemed to have solved this issue by setting proper upper & lower bounds for my parameters. They seemed to have been trapped in local minimum during the LS optimization. I observed the parameter trends and set limits to avoid too much deviation from the previous parameters and that seemed to have solved the problem for now.

12% ledeburitic cold work tool steel, etched for revealing retained austenite, any help for clarifying?

How to size battery banks?

How to model the battery pack in PSIM or PSCAD?

How to calculate battery consumption for IoT device?

How to delaminate copper in water-based anodes?

Is there an ideal incubation time of RGO-modified SPCE before subsequent use, i.e., electrochemical characterization or further modification?

What is the difference between AC conductivity and DC conductivity? Is AC conductivity = ionic and DC conductivity = electronic?

Where can the fault diagnosis data set of lithium battery be found?

How can I measure the surface interaction or surface interfacial chemistry between polishing abrasives and upper surface of substrate ?

Not able to charge-discharge Li-ion coin cell at higher C rate ?