Battery Digital Twin: The Future of Battery Intelligence

As electric vehicle (EV) adoption continues to rise, battery analytics become crucial. The battery serves as the new engine, directly influencing the vehicle’s performance, reliability, and safety. Advanced battery analytics enable predictive maintenance, optimize battery lifespan, and predict unexpected failures by continuously monitoring key electrical, thermal, and performance parameters. Real-time data analysis enhances safety by detecting anomalies like overheating or voltage fluctuations, mitigating the risks of malfunctions.

Battery Digital Twin (BDT) is a digital twin of a battery used in an electric vehicle battery or battery energy storage system (BESS) enabling monitoring performance, predicting outcomes, and optimizing operations to make a physical instance of the battery more performant, reliable, and safe. A BDT is a virtual representation of the physical battery structure, state, and behavior. The BDT enables a deeper understanding of internal battery dynamics through a virtual representation that is continually synchronized with the real-time operational condition of the battery and predictive models to forecast future states and conditions.

This blog post delves into a comprehensive approach of BDT data collection strategy to help create a living BDT and a solution approach to help address various battery use cases. BDT can help derive actionable insights based on the data collected from connected vehicles and other IoT devices by users of the BDT.

Data Collection Strategy

EVs’ connected vehicle interfaces typically provide battery data through bus communication, like controller area network (CAN), between the telematics control unit and the Battery Management System (BMS) Controller. A software application running on the telematics control unit can provide filtering, edge analytics, or event-based data logging.

The BMS often uses algorithms like the Kalman Filter to process current battery data. However, these algorithms work on current data and can’t take advantage of historic data and apply AI/ML capabilities to detect faults or any risky situations. By sending battery data to the cloud, you can use the cloud’s computational resources to help apply more advanced data analytics or AI/ML algorithms. This approach allows you to analyze the historical battery data stored in the cloud, which can help provide deeper insights and more accurate predictions about potential battery failures.

The critical data points and the frequency of collecting them, each discussed below, are crucial elements underlying value and cost optimization for BDT systems: –

Data Points: EV battery performance is influenced by a variety of factors and collecting the right data is essential for powering accurate predictive analytics. Key parameters that should be considered when implementing BDT include:

• Electrical Data: Voltage (cell, module, pack level), current flow, state of charge (SoC), state of health (SoH), internal resistance, and power input/output.
• Thermal Data: Cell and module temperatures, cooling system efficiency, and temperature gradients.
• Performance Metrics: Charging/discharging cycles, capacity degradation, energy efficiency, and charge acceptance rates.
• Environmental Data: Ambient temperature, humidity, vibration levels, and operating conditions.

Frequency: Collecting EV battery performance data at the right frequencies is crucial for balancing data accuracy and processing efficiency. Different data points require varied sampling rates to effectively capture their behavior in a cost-effective way:

• High-Frequency (<10 seconds): Voltage, current, and temperature at the cell, module, and pack levels to capture rapid fluctuations.
• Medium-Frequency (1-5 minutes): SoC, internal resistance, cooling system performance, and temperature distribution for tracking gradual changes.
• Low-Frequency (hourly/daily): SoH, capacity measurements, energy efficiency, and environmental conditions for long-term trend analysis.

In addition to continuous battery data collection, an event-based data collection approach can also be implemented. This allows you to capture specific events or conditions that may be indicative of potential battery issues or degradation.

Furthermore, you can optimize the data collection process by adopting an adaptive sampling approach. This technique adjusts the data collection frequency based on the operating conditions of the vehicle and battery. For example, an adaptive sampling approach can increase the sampling rate during high-stress situations, such as uphill driving or exposure to extreme temperatures. Conversely, it can reduce sampling frequency during normal operation to minimize data transfer and storage costs.

In the event of a detected failure or anomaly, the connected vehicle edge agent may need to temporarily increase the data collection rate for a specific battery or battery model to gather more detailed information. Therefore, the connected vehicle edge agent should be intelligent enough to dynamically adjust data collection parameters based on the current conditions and requirements.

By implementing both event-based and adaptive sampling techniques, you can strike a balance between capturing comprehensive battery data and optimizing the system’s resource utilization. This approach allows you to collect the most relevant data while minimizing the overall costs associated with data transfer and storage.

Solution Approach

Battery data can reveal multiple different faults or anomalies. By bringing all the battery data together, you can obtain the necessary information to help identify and predict these faults and anomalies. This consolidated data can also help establish correlations between different parameters.

To create a BDT system, you need to combine four distinct types of data, including real-time, current state, specification, and predictive value data.

Battery Real-time Data (e.g., voltage, current, temperature, etc.): Real-time data from batteries, such as voltage, current, temperature, and other metrics, can be collected using an intelligent edge agent. One of the advantages of deploying an intelligent edge agent solution is that it can support various data collection mechanisms, such as condition-based, time-based, edge analytics, etc.

The cloud backend must also possess intelligence to support campaign-based data collection capabilities. These campaigns provide instructions to the edge agent software on how to select, collect, and transfer data to the cloud. Intelligent cloud backend and campaign-based data collection allows for targeted and efficient data gathering as the campaigns can be tailored to specific needs and requirements.

This condition-based collection scheme allows for the capture of data when certain predefined conditions are met, such as exceeding a specific voltage threshold or temperature range. This can be useful for monitoring critical battery parameters and triggering alerts or actions based on real-time conditions. In contrast, the time-based collection scheme enables the regular and periodic collection of data, regardless of specific conditions. This can be valuable for long-term trend analysis and monitoring of battery performance over time.

Battery Current state Data (e.g., SoH, SoC, Charge Cycle etc.): AWS IoT Device Shadow can be used to store various state data associated with a device. Shadows can store both medium and low frequency data points related to a device’s state. This allows the device’s current state to be made available to apps and other services, regardless of whether the physical device is currently connected to AWS IoT or not.

AWS IoT Device Shadow acts as a virtual representation of a physical device and maintains the device’s reported state even when the device is offline. This helps enables greater flexibility and responsiveness, as applications can continue to monitor and interact with the device without being blocked by connectivity issues.

Battery Specification Data (e.g., cycle-life, BMS version, etc.): The battery specification data, such as number of cells, modules, charge cycle, capacity, etc., would come from other enterprise systems. Amazon SageMaker Data Wrangler can be used to retrieve specification and initial performance characteristic data from internal systems. It has the capability to quickly select, import, and transform data with SQL and over 300 built-in transformations without writing code.

Predictive Value (e.g., SoH prediction, RUL, etc.): By continuously monitoring and analyzing battery data, potential battery failures can be predicted before they cause critical issues such as thermal runaway, accelerated degradation, lithium plating, etc. Amazon SageMaker Unified Studio can be used to develop AI-driven analytics that identify warning signs such as:

• Unusual temperature patterns indicating thermal runaway risks.
• Rapid capacity loss signaling potential battery degradation.
• Voltage imbalances leading to reduced efficiency.
• Abnormal internal resistance changes that could point to early-stage failures.
• Irregular charging/discharging behavior affecting long-term performance.

AWS has published a sample Guidance for Battery Digital Twin on AWS. Please note, Amazon Forecast is referenced in the Guidance but is no longer available to new customers; however, Amazon SageMaker Canvas is a viable substitute to Amazon Forecast. Amazon SageMaker Canvas allows developers to build ML models using a visual, no-code interface. Amazon SageMaker Canvas supports multiple ML techniques, including linear regression, logistic regression, deep learning, time-series forecasting, and gradient boosting and creates a custom model that makes the most accurate predictions based on your dataset.

The battery degradation process is highly complex and influenced by numerous factors. Temperature, charge state, and discharge rate can significantly impact a battery’s performance over time. Various electrochemical side reactions and operational conditions of the anode, electrolyte, and cathode can further degrade a battery’s capabilities, ultimately affecting its lifespan. The performance of EV batteries will decline both with calendar time (calendar aging) and usage (cycle aging)—a phenomenon known as battery aging. To effectively monitor this aging process, a BDT approach is needed, which requires continuously updating the underlying ML model.

To update the ML model, you have two possible scenarios: (a) periodic retraining of the ML model; or (b) triggering updates when the model drifts beyond a predefined threshold. We can use Amazon SageMaker Model Monitor to continuously monitor the quality of models in real time and set up an automated alert triggering system when there are deviations in the model quality, to retrain the model. We can use Amazon SageMaker Canvas to periodically retrain ML models with updated datasets to constantly learn and improve the model performance. This allows automation of model training and deployment, eliminating the need for manual intervention.

As the industry explores data collection strategies and the utilization of data to achieve a meaningful outcome through a structured approach of BDT, it has become evident that executing advanced analytics or developing ML models necessitates a deep understanding of data. Generally, data science teams lack domain knowledge which impedes the development of such applications. Refer to the blog Empowering ML Teams with Amazon Q Automotive Expertise, which highlights how Amazon Q can bridge the domain gap for data science teams.

Conclusion

With real-time monitoring and predictive insights, EV manufacturers, fleet operators, and service providers can enhance battery lifespan, reduce unexpected failures, and improve overall EV efficiency. The future of EV battery management requires data-driven intelligence to help ensure safer, more reliable, and cost-effective electric mobility.

You can explore the demo code and user interface of BDT in our guidance github repository. Please don’t hesitate to reach out to AWS’s Solution Architecture team for guidance on building the ML model or to use the “Ask AWS” chat feature on the AWS home page. We’ll keep you updated with future blogs related to electric vehicles.