Remaining Useful Life

What is Remaining Useful Life (RUL)?

Remaining Useful Life (RUL) is a key function declared by the battery management system. As per the title it gives you the remaining predicted lifetime of the battery based on its usage and degradation to the failure threshold [1]. It represents the period from the observation to the end of life (EOL) [3]. EOL refers to the time and the number of charge-discharge cycles when the battery characteristic parameters reach the replacement threshold.

Figure 1. SOH vs RUL concept

How is RUL different to SoH?

SOH is the estimate of the battery’s health in terms of capacity fade or internal resistance increase at the present time, whereas the RUL is a prediction of the SOH trajectory in the future (see Figure 1). As can be seen from Figure 2 battery degradation is non-linear. This means that the SOH of a battery does not decrease at a constant rate over time, therefore, extrapolating the SOH from past data may not be accurate or reliable, as the future degradation may deviate from the past trend. This is why the research into effective RUL methods that can effectively predict RUL, has increased over the last few years.

Why is Remaining Useful Life (RUL) important?

With the ability to predict the RUL accurately, the users, OEM’s, fleet operators or insurance companies can have a better understanding of the time at which the battery will reach its EOL and, based on that, plan any maintenance activities proactively. The accurate RUL also plays a crucial role in the second-hand car market, helping end-users gain trust and confidence in the battery pack’s longevity. It also enables fleet operators to plan the redeployment of used packs in second-life applications.

Figure 2. Battery ageing in real world condition.


Battery operation is governed by complex electrochemical reactions, resulting in degradation over time. The battery degradation is a highly non-linear process and encompasses two types of ageing:

  • Calendar ageing (mostly affected by factors such as resting duration, temperature, and state of charge)
  • Cycling ageing (mostly affected by temperature, charge, and discharge rate, and depth of discharge, and state of charge)

Therefore, for the RUL estimation method to be accurate, both types of ageing should be taken into account. RUL estimation methods can be typically categorised into:

  • model-based methods (such as physical models, electrochemical models, etc.)
    • Pros: can be very accurate and represent the individual ageing processes very well
    • Cons: difficult to characterise, requires specialist battery knowledge, and requires high computation cost to execute
  • data-driven methods (such as machine learning)
    • Pros: simple model is sufficient, requires less computation cost, doesn’t require specialist battery knowledge
    • Cons: requires large set of training data, no representation of individual ageing processes
  • or a combination of both, known as a hybrid method.

What inputs are typically used to predict RUL?

Voltage, temperature and current measurements of the cells over time are required. The accuracy and sampling frequency will highly depend on the method chosen.

For training of the data-driven methods, both lab ageing data and the field ageing data can be used.

Can this be done within the BMS and what advantages are there to using cloud data and algorithms?

The RUL function can be implemented within the BMS, but it will depend on the algorithm choice and the available computational resources on the HW. Some data-driven RUL algorithms may require a special microcontroller equipped with neural accelerators or a parallel processing unit (PPU). A trade-off between the computational cost and performance is likely. However, most of the electric vehicles are equipped with connectivity units allowing for the key battery parameters to be broadcasted into the cloud, where even the most complex RUL algorithms can be executed. Since ageing is a relatively slow process, the frequency of measurement to be transmitted doesn’t have to be very high. Another advantage is that in the cloud the RUL models can benefit from collecting the data from the fleet and periodically (as more data becomes available) repeating the training process, resulting in improved accuracy and robustness over time.

What typical accuracy would you get?

An achievable accuracy of >90% should be expected from effective industrially available RUL algorithms. Higher accuracies can be achieved with high fidelity physics-based methods, but this would come at the expense of execution cost, and in extreme cases may make it practically not yet achievable with available microcontroller. RUL accuracy in the first 100 cycles may not be high, as the cell could be experiencing shift in discharge voltage with no change in capacity.

Other aspects of the RUL feature, like:

Should also be assessed, as these can yield useful insights into battery lifetime management strategies.

There are companies like Eatron Technologies, who are already deploying industrially available RUL software helping OEM’s extending the lifetime of their batteries.

What can cause instability in the prediction of Remaining Useful Life (RUL)?

The effective RUL prediction needs to consider both the dynamic driving behaviours (cyclic ageing) and idling conditions (calendar ageing).

EV users driving styles and unique usage conditions make RUL prediction differ from lab experiments where cells are charged and discharged under controlled states.

Another important noise factor is the difference in the measurement accuracy of the laboratory equipment vs measurement system inside the actual battery, which needs to be effectively compensated for during the training phase using lab data [2].


  1. Wang Shunli, Jin Siyu, Deng Dan, Fernandez Carlos, A Critical Review of Online Battery Remaining Useful Lifetime Prediction Methods, Frontiers in Mechanical Engineering, 7, 2021
  2. Carlos Cesar Martins Ferreira, Gil Manuel Gonçalves, Remaining Useful Life prediction and challenges: A literature review on the use of Machine Learning Methods, Journal of Manufacturing Systems, April 2022
  3. Gabriele Pozzato, Matteo Corno, Closed-loop Battery Aging Management for Electric Vehicles, IFAC-PapersOnLine, Volume 53, Issue 2, 2020

Eatron Technologies

Our intelligent software platform approach significantly reduces cost, risks and time to market. Eatron offers embedded applications for High and Low Voltage Battery Management that are not only automotive grade, safe and robust but also integrated with AI & cloud layers with analytics, offering OTA updates and continuous software improvements to enable its customers to achieve superior performance and reliability over the lifetime of the vehicles.

Leave a Comment