- A new artificial intelligence model can accurately determine the State Of Health of a battery.
- It uses data captured from the battery in its initial stages and predicts whether the battery would have a long or short lifetime.
Lithium-ion batteries are used in a wide range of application due to their high energy density, long lifetimes and low costs. Over the last few years, the commercialization of hybrid and electric vehicles has stimulated an increasing demand for quality batteries. Thus, analyzing battery “health” has become increasingly important.
However, one of the major obstacles in the development of battery technology is monitoring and testing battery health, which takes a lot of time and the process affects battery life.
A parameter called State Of Health (SOH) represents the battery’s ability to store energy, relative to its ideal or initial conditions. For a new battery, SOH is usually 100% but decreases over time. Assessing SOH is important for safe and correct usage of battery. However, there isn’t any technique that can accurately determine this value without damaging the battery life.
Determining SOH Isn’t Easy
A battery’s SOH is associated with two factors that arise as batteries age –
- Capacity Fade: progressive loss of storage capacity
- Electrical Resistance: progressive increase of impedance that causes the battery power to decline.
In lithium-ion batteries, the increase of impedance and loss of storage capacity occur from numerous interacting processes. Since these processes happen at similar timescales, it is very difficult to analyze them independently. Thus, one cannot use a single direct measurement to evaluate SOH.
Traditional techniques [for determining SOH] involve assessing interactions between the electrodes of the battery. But since this makes the battery unstable, these techniques are unacceptable.
At present, there are two approaches to determine SOH in a less destructive manner: adaptive models and experimental methods. The first approach uses battery-performance data to self-adjust and decrease errors. However, this type of methods needs to be trained on experimental data before they can be actually used in a production environment.
The second approach, on the other hand, can be used to determine particular failure mechanisms or physical processes that happen in a battery. This provides a good estimation of the future rate of capacity degradation. However, such methods fail to identify intermittent malfunctions.
AI Can Accurately Predict Lifetime Of Batteries
Now, researchers at MIT, Stanford University, and Toyota Research Institute have come up with artificial intelligence (AI) model that can accurately determine the SOH of a battery.
The team created a comprehensive data set characterizing the performance of 124 lithium-ion batteries. The data was recorded as batteries underwent different fast-charging conditions. A wide range of charging and discharging cycles (250 – 2,300) was included in the data.
They then used the Machine Learning (ML) method to examine the data and generate models that can accurately estimate battery cycle lives. They analyzed only the first 100 cycles of each battery (before there were clear indications of loss of storage capacity).
Estimated vs observed batteries’ lifetime | The dashed line shows where estimations and observations are equal, for reference | Courtesy of researchers
The best model generated by ML was able to correctly estimate cycle lives for 91% of the batteries. Researchers also used this method to study data from just the first 5 cycles of each battery. This time the aim was to figure out whether batteries would have a long or short lifetime (more than or less than 550 charging-discharging cycles). In this case, the model made correct predictions for 95% of the batteries.
Although the new models were more effective than traditional SOH-determining methods, they were less accurate at predicting cycle lifetime for batteries whose storage capacities were already faded to some extent.
The research team believes that their new approach is a promising way of estimating life cycles of lithium-ion batteries, and could help in the development/improvement of emerging battery technology.