Although great efforts have been made in developing data science technology for benefitting full-lifespan management of Li-ion batteries, many knowledge gaps still exist. This chapter summarizes these challenges, future trends, and promising solutions to boost the development of data science solutions in the management of battery manufacturing, operation, and reutilization, respectively. This could further inform the selections of data science methodology and academic research agendas alike, thus boosting progress in data science-based battery full-lifespan management on different technology readiness levels.

7.1 Data Science-Based Battery Manufacturing

Efficient data science-based management of battery manufacturing, hampered by the complicated intermediate stages within the battery manufacturing chain, is becoming a key but challenging research direction, as it could play a direct and pivotal role in affecting manufactured battery performance. Considering the huge requirement on battery with high performance for larger-scale energy storage applications, we outline four challenges and ways ahead directions, as illustrated in Fig. 7.1, with the overarching target of achieving more technological innovation and breakthroughs in data science-based management of battery manufacturing.

Fig. 7.1
figure 1

Challenges and ways ahead for data science-based management for better battery manufacturing

Full size image

7.1.1 Continuous Manufacturing Line

The current established battery manufacturing line contains stages with various material flows. For example, coating and calendering are continuous, electrode cutting, electrolyte filling, forming, and testing are discrete, while mixing is batch-wise. Traditionally, as a batch-wise process, slurry mixing would cause variations right from the beginning of electrode manufacturing. A suggested way is to improve mixing technology and make it become a fully continuous process. The obvious benefits of continuous manufacturing line are it requires less manpower to operate but also can include high-precision dosing systems and in-line quality monitoring and analysis. For large-scale battery manufacturing line, achieving continuous manufacturing could also address the bottlenecks in terms of time and cost [1].

7.1.2 Digital Manufacturing Line

Due to the superiorities in terms of predictive maintenance and real-time quality control, Industry 4.0 with digitization is becoming ideal for battery manufacturing [2]. In this case, once the characteristics of the intermediate manufacturing product are measured, the machine corresponding to the next step will adjust its settings for correct further processing. Several measures including equipment automation, digital twin of products and processes have been taken to implement digital manufacturing lines [3], which face their own issues in communication protocols, network security, and financial investment [4]. Therefore, the digital battery manufacturing line becomes a promising direction as the technology cost reduces, the utilization of data science and cloud tool represents less of a barrier. Moreover, with the digitization of the battery manufacturing line, an in-depth understanding of manufacturing equipment, parameters, data collection, and processing is worth being explored.

7.1.3 Advanced Sensing Methodology

Battery manufacturing is a complex process that encompasses multidisciplinary disciplines in material, chemical, and mechanical operations. As the most complex stages, the strongly coupled parameters from coating and drying processes would highly influence electrode and battery performance in terms of material, electrochemical, and mechanical properties. Current researches mostly focus on the off situ electrode characterizations such as surface morphology and element distributions, while in situ methodologies are still poorly adopted. For example, for the drying process, although a few in situ characterization methods have been adopted to explore the drying rate, binder and particle distribution, the dynamic information acquired for drying is still limited [5,6,7]. To further obtain more qualitative and quantitative data for data science activities, advanced sensing methodologies such as X-ray CT [8], Fourier transform infrared microscopy [9, 10], contrast-variation small-angle neutron scattering [11] are suggested to be adopted.

In the light of this, advanced sensing methodology is crucial for the next generation battery manufacturing management in the medium and long runs. Future ways ahead could focus on in situ sensing methods that could accurately access important information about manufacturing parameters, and deeper insight into the electrochemical measurement of battery manufacturing. Moreover, 3D image sensing methodology with computational models has illustrated substantial benefits in battery properties [12]. This methodology also attempts to give more in situ information on battery manufacturing, further benefitting the generation of new data and profound observation for more efficient data science management of battery manufacturing.

7.1.4 Improved Machine Learning

Data science-based models could bring many merits for parameter analyses in battery manufacturing, such as greatly reducing the expertise required to use model. However, pure data-driven models still present limitations, such as the inability to give physical insights into the battery manufacturing line. This would significantly hinder battery manufacturers to optimize their manufacturing line. In addition, the parameter analysis capabilities of pure data science models are highly influenced by the quality of the data used, because these data must cover enough information to ensure that the model can be well trained. In this context, it is important to explore how to improve the pure data science model by combining it with other powerful tools. One key lies in the design of data physics-driven models by coupling the physical elements of battery manufacturing to machine learning approaches to further help provide physical insights for battery manufacturing parameter analysis. In addition, image-based models that link relevant X-rays and other imaging information to describe battery micro behaviour have also become powerful tools in the battery field [13, 14]. In this context, hybrid tools through combining the benefits of machine learning and image-based model are suggested to improve the interpretability of data science tools and get more valuable information from images for wider battery manufacturing data analyses.

7.2 Data Science-Based Battery Operation

Battery operation management, featured by its multidisciplinary nature, is becoming a fast growing research area, as there are increasingly stringent regulations on battery performance for large-scale transportation electrification applications. Considering potential scientific importance and engineering application requirements, several data science-based research directions and trends in this field, from the perspectives of battery operation modelling and state estimation, lifetime prognostics, fault diagnosis, and battery charging, are outlined, with the overarching target of stimulating more technical innovations and transformative breakthroughs.

7.2.1 Operation Modelling and State Estimation

Although the current data science-based battery operation modelling and state estimation approaches have made great progress, the following challenges still exist and need to be further improved, as illustrated in Fig. 7.2.

Fig. 7.2
figure 2

Ways ahead for better data science-based battery operation modelling and state estimation

Full size image

Robust and simplified operation modelling: For battery operation modelling, the widely utilized equivalent circuit model (ECM) lacks enough chemical significance, making it become difficult to describe many dynamic characteristics within a battery. Therefore, it is recommended to combine battery electrochemical elements into ECMs. The machine learning-based model is usually trained under a specific condition, further causing generalization issues in real battery operations. This requires the model could be trained under comprehensive operation cases. Therefore, the training data should consider the elements of battery degradation, hysteresis, charging/discharging rates, and temperatures. On the other hand, another big issue for battery operation modelling is the trade-off between computational efforts and computing resources. In the light of this, another future research direction should focus on the reduction and simplification of battery operation modelling.

Joint multi-state estimations: There are a number of data science-based battery single state estimation methods reported in the literature, whereas the research of at least two-state joint estimation is still limited. It should be known that battery internal states are actually coupled and interact with each other. Estimating one state independently and ignoring other states can obtain satisfactory results only under certain constraints. In the light of this, according to the multi-field coupling of electro, thermal, ageing, and mechanical conditions of a battery, devising advanced data science-based methods such as fractional-order calculus [14] and multi-time scale estimator [16] to effectively enhance battery multi-state joint estimation performance, with a reliable computing efficiency, becomes another promising research direction.

7.2.2 Lifetime Prognostics

Battery lifetime prognostics is also a key and hot research topic in battery operation management. Although great data science efforts have been made in this field, several challenges from Fig. 7.3 are still existed as:

Fig. 7.3
figure 3

Ways ahead for better data science-based battery lifetime prognostics

Full size image

Battery degradation identification: The pure data science-based solutions, especially just using machine learning technology, are difficult to explain battery degradation mechanisms. It would become meaningful to integrate the information of battery degradation mechanism with the lifetime prognostic approaches. As such, combining machine learning with physical information about battery ageing is a promising research direction. As some battery ageing data curves such as IC/DV contain battery degradation mechanisms, one suggested way is to first collect IC/DV data for uncovering battery ageing mechanisms, then couple this information into machine learning. In this way, IC/DV data could reflect sensitive indicators of battery ageing, further benefitting battery lifetime prognostics.

Self-improving model via online data: Li-ion battery degradation is sensitive to the operation cases. Effectively predicting battery lifetime under conditions different from the training cases is a challenge. The difference between the laboratory cases for model development and the real operation conditions limits the wider applications of data science-based approaches. This could be improved in two ways: (1) the scale of an experimental ageing dataset can be increased to cover more battery degradation information under wider operation cases, further improving the predictability of derived data science-based solutions. However, this would also lead to the increased cost of the battery ageing tests. (2) Improving the dynamic updating capability of data science-based models developed offline is worthy of further research, because it could pave a way to the self-improving model.

Lifetime prognostics at pack level: To date, most of battery lifetime prognostics research is explored under battery cell level. However, in real applications, numerous cells require to be connected with series or parallel forms to construct a battery pack for providing enough energy and power. Understanding battery pack degradation requires knowledge beyond the cell level, considering additional effect elements including cell inconsistency, electrical imbalance and temperature variations among cells. All these issues complicate accurate lifetime prognostics modelling for a battery pack. The advances in the state-of-the-art deep learning tools are foreseen to introduce some ways to these issues. Some deep neural networks such as convolutional neural network and generative adversarial network have the ability for highly complicated nonlinear fitting and become good candidates for handling these issues. The use of such deep learning tools or similar self-learning solutions also becomes a promising way for battery lifetime prognostics at the pack level.

7.2.3 Fault Diagnostics

Battery failure is a very large potential hazard to vehicles, so the battery fault diagnosis is becoming a research hotspot. At the same time, battery failures are concealed and have a long incubation period, which brings great challenges to its diagnosis. The current fault diagnosis is still based on feature detection, and the early warning capability of hidden faults needs to be improved. With the development of sensor and artificial intelligence technologies, future development of battery fault diagnosis has the following trends, as illustrated in Fig. 7.4.

Fig. 7.4
figure 4

Ways ahead for better data science-based battery fault diagnosis

Full size image
  1. (1)

    Diversification of sensing signals. The development of smart sensors makes it possible for the smart perception of batteries. Some currently unmeasurable parameters will become measurable through built-in smart sensors, such as the battery internal temperature, pressure or strain inside the battery, internal gas concentration and composition, positive and negative absolute voltages, and impedance spectra. These signals bring important input to the fault diagnosis of the battery, changing the current defect that there is only insufficient information such as battery voltage, current, and temperature. It is worth mentioning that the online measurement technology of EIS provides very valuable information for the battery fault diagnosis, and the development of EIS chips in future is an important direction.

  2. (2)

    High-fidelity battery models. The current battery model has a contradiction between complexity and accuracy. However, the high-precision state estimation of the battery requires high-precision physical and electrochemical models. Cloud computing makes the application of this highly complex model and the adaptive updating of model parameters based on data become a reality. For the fault diagnosis model, the key technology is how to establish the mapping relationship between the sensor signal and the internal state of the battery from the battery mechanism, and how to establish the mapping relationship between the fault type and the battery model.

  3. (3)

    Intelligent diagnosis and decision technologies. The battery fault characteristics are deeply mined by machine learning. Model, data, and machine learning are becoming the three core elements for battery fault diagnosis. The development of big data has produced a large amount of data. Machine learning technology is an important means and powerful weapon to explore massive data. The model-driven and data-driven fault diagnosis methods urgently need the support of machine learning. The parameters of the physical model need to be learned and improved by machine learning. In addition, how to extract fault features from massive data is a very challenging task. Machine learning can effectively and deeply mine the hidden fault features in the data, which can greatly improve the fault recognition rate and reduce the false alarm rate.

  4. (4)

    End-edge-cloud collaboration. The application of digital twin technology gives a new concept of battery network management and service. The digital twin technology and cloud collaboration of the future battery management system will establish a battery fault diagnosis algorithm based on outlier mining, break through the limitations of computing power and storage space of traditional battery management, and realize the refined safety management of the full battery life cycle.

7.2.4 Battery Charging

Figure 7.5 illustrates the ways ahead for better data science-based battery charging management, which includes three main parts: robustness improvement, thermal management, and pack level charging.

Fig. 7.5
figure 5

Ways ahead for better data science-based battery charging management

Full size image

Robustness improvement: For battery charging, as numerous data science-based explorations on charging management are experimental or empirical in nature, their performance has been explored only under a limited range of battery chemistry, operation factors and cases. These results are difficult to be extended to other battery types or conditions, as supported by the frequent conflicting observations from different reports. Besides, lots of model-based optimal charging strategies are based on equivalent circuit model or single particle model, and only be validated under a specific operating condition. Such strategies would become inaccurate especially under high power or high current cases. In this context, more in-depth exploration of battery charging behaviours and robust data science-based battery models are required for efficient battery charging strategy design that could generalize well under different operational conditions.

Thermal management under low/high temperatures: Thermal management is a key element in battery charging. To date, existing charging strategies are mainly designed under normal temperature conditions without considering the cases of low or high temperature charging. Due to the increased number of electrical vehicles is deployed in both colder and hotter climates, battery charging strategies under such conditions become increasingly critical. In this context, experimental data under extreme charging conditions are required to design suitable data science-based charging strategies. Then these charging strategies could equip with suitable thermal management solutions to preheat or cool batteries. This could significantly improve battery safety during the charging period and performance such as battery lifetime.

Pack level charging: To further benefit the charging performance of each individual cell within a pack, charging strategies considering the cell inconsistency are required to avoid local overcharge or safety issues. Currently, multi-objective charging strategies have been well designed for single battery cells, but their transferability, influence and costs in battery packs have not been fully explored. Charging strategies with the ability to improve the performance of a single cell would also lead to uneven currents or temperatures when executed on the battery pack. As fast charging would amplify heterogeneity, such charging research considering the effects of cell-to-cell difference is urgently required. In this context, experimental data of different cells within battery pack charging are required. Then the data science-based battery pack charging solutions could be explored to integrate cell and pack management, further improving charging performance at the pack level.

7.3 Data Science-Based Battery Reutilization

The echelon utilization and material recycling are the key links to building the closed-loop management of Li-ion batteries in the full life cycle. They have obvious significance in resource recycling and environmental protection. With the explosive development of EVs and the rapid development of the Internet, big data, and artificial intelligence (AI) technologies, battery management is entering the digital and intelligent stage in the entire life cycle. As summarized in Fig. 7.6, there are three important trends.

Fig. 7.6
figure 6

Schematic diagram of the entire life cycle and digital management of Li-ion batteries, reprinted from [15], with permission from Elsevier

Full size image
  1. (1)

    New industrial structure is forming and reconstructing. Massive zero/low marginal cost Li-ion batteries in EVs will greatly enrich the flexible resources of the power grid and promote the rapid development of the mobile energy Internet. In 2030, EV ownership in China will reach 100 million, and the power of on-board power Li-ion batteries will exceed 1 billion kilowatts, which is equivalent to 50 Three Gorges power stations. The on-board power batteries of mass EVs can absorb the excess power in the power grid. What we need to do is to make these EVs charge and discharge at the right time. The power battery will make the energy Internet connect more widely, flexibly, stably, and strengthen the cooperation with other elements. Under the framework of energy Internet, smart grid, battery leasing, battery recycling, echelon utilization, sharing economy, and other new industrial structures and modes are forming and developing.

  2. (2)

    Circular economy is further developing and deepening. The entire life cycle of Li-ion batteries can be divided into production, service, and retirement stages. Li-ion batteries provide power for EVs during service, which is also an important part of the energy Internet. When the battery is retired from the EV and enters the stage of echelon utilization, it can directly participate in the energy Internet. When Li-ion battery cannot meet the requirements of echelon utilization scenarios, it will enter the recycling link. Recycling valuable metal of the spent battery has great environmental protection and economic value. The recycled battery materials can be reused for battery remanufacturing, which is of great significance to alleviate the resource crisis. By constructing the cycle economy model of the full life cycle of Li-ion batteries, the value of Li-ion batteries can be maximized, which is of great significance to the sustainable development of Li-ion batteries.

  3. (3)

    The new technologies are applied in the full life cycle of Li-ion batteries. First, the ageing and failure mechanism of Li-ion batteries will be clearer, which is very critical for the fine management of Li-ion batteries. Second, the high-quality operation data in the entire life cycle of Li-ion batteries is very valuable. These data are the basis of accurate state estimation and safety management of Li-ion batteries, and they also facilitate the quick evaluation of residual value of the retired Li-ion batteries. In recent years, the intelligent BMS integrating mechanism model, cloud data and artificial intelligence technologies have become a hot research topic, which promotes the accurate and efficient management, optimization and control of Li-ion batteries in the full life cycle. For Li-ion batteries recycling, the development of new recycling methods with characteristics of green, highly efficient, low energy consumption, short process is an important research field. We think that the following three aspects are favourable trends: (a) the study of selective leaching separation of Li will greatly improve the economic benefits of recycling; (b) the development of co-extractant for the extraction of Ni, Co, and Mn metals will shorten the separation process and reduce costs; (c) direct electrode recycling technology is an important development direction.

  4. (4)

    Sustainable and green development have attracted more and more attention. Under the requirements of carbon neutralization, the carbon emission in battery recycling and remanufacturing needs to be paid special attention. Battery recycling has high economic value, but its impact on the environment is a topic that needs to be expounded. As a powerful evaluation tool of environment, resources, and cost, lifecycle assessment (LCA) has received great attention in the full lifecycle management of Li-ion batteries in recent years. The environmental and resource burdens are evaluated by the LCA method in the full life cycle of Li-ion batteries is of great significance to improve process flow and products, risk assessment and policy decision-making.

7.4 Summary

Li-ion battery represents one of the promising energy storage solutions for many applications such as transport electrification and smart grid, owing to its high energy density, reliable service life. Battery management technologies are developing rapidly in the international market and are also a hot research topic. However, there are many technical challenges in battery full-lifespan applications, demanding state-of-the-art data science approaches. Firstly, battery properties such as cost, reliability, energy density, and life are directly determined by its manufacturing process. It is thus vital to develop suitable solutions for understanding and analysing battery intermediate manufacturing processes in the pursuit of smarter battery manufacturing management. Besides, owing to complicated electrochemical dynamics, numerous tasks including battery operation modelling, state estimation, lifetime prognostics, fault diagnosis, and charging must be done to well and efficiently manage battery during its operation stage. Moreover, to make full use of battery residual value, the retired battery will be reutilized in second-life applications such as grid energy storage. To further comply with environmental and health benefits, batteries need to be recycled finally when they get either spoilt or non-functional. With the rapid development of AI and machine learning technologies, data science-based applications have drawn much attention and become a research hotspot in the field of battery full-lifespan management. After well-designing proper data science solutions, significant enhancement can be achieved for more effective battery management from the aforementioned three parts. However, as relative new and prospective research, currently there is no book to systematically introduce and describe the battery full-lifespan management particular from data science application perspective to our best knowledge.

In this book, data science-based battery full-lifespan management strategies are comprehensively reviewed and discussed. In Chaps. 1 and 2, i.e. (1) the introduction to Li-ion battery and related management, (2) the key stages of battery full-lifespan and the basics of data science technology, give illustrative descriptions on the research focus. Then in Chaps. 3–6, the new and emerging data science technologies for full-lifespan management of Li-ion battery from three key aspects, i.e. (1) battery manufacturing management, (2) battery operation management, and (3) battery reutilization management, are discussed with plentiful case studies. Finally, this chapter overviews the key challenges, future roadmap in all these three parts.

In a nutshell, to make full use of battery to support power/energy and ensure its safety, efficiency and performance, reliable data science methods are required in the battery full-lifespan management, but many corresponding technologies are immature. None of the data science solutions can be regarded as a one-size-fits-all solution; instead, there are inherent trade-offs between complexity and performance in different applications. To widen the data science applications for battery management, three key features are given in this book: (1) the concept of full-lifespan management of Li-ion battery is proposed and the state-of-the-art data science technologies to handle related key tasks are described. (2) Case studies of deriving various data science technologies to benefit battery manufacturing, operation and reutilization are systematically introduced, which proves that data science is a promising route to improve the full-lifespan management of battery. (3) Valuable guidance for the challenges, future trends, and promising solutions to benefit data science-based battery full-lifespan management are provided. With the above arrangement, we hope that this book provides useful reference points to support the design of data science-based battery management solutions during its lifespan, while a brand-new hologram to make full use of battery during full-lifespan will be formulated, further boosting the advancement of AI and low-carbon technologies.