The transition to electric powertrains brings about new challenges in predicting when something might go wrong and in keeping the powertrain healthy. While traditional diagnosis methods have their merits, they often provide only a limited opportunity to address the dynamic evolution of potential dangerous conditions in electric powertrains, where the interplay of numerous intricate electronic components and complex systems requires a more sophisticated approach to monitoring and diagnostics.

Purpose of the system

ESACO (Electric powertrain System Anomaly detector and Conditions Observer) has been developed in response to the urgent need for improving the safety, reliability, and performance of electric powertrains. As electric vehicles (EVs) become increasingly prevalent in the automotive landscape, ensuring their safety and reliability is paramount.

ESACO's scope extends beyond just enhancing individual electric powertrains; its goal is to optimize their usage and integration into the transportation landscape. Utilizing advanced diagnostic and predictive technologies, ESACO aims to enhance the reliability and performance of electric vehicles, thus facilitating their wider adoption. This contributes to mitigating environmental impact by reducing emissions and advancing sustainable transportation solutions.

Ai based alghoritms

The availability of comprehensive electric vehicle powertrain data, along with the capacity to effectively manage it, enables the utilization of AI-based trained algorithms. This integration represents a powerful and innovative approach for real-time monitoring of powertrain health, allowing for the early detection of anomalies and potential issues.

By leveraging advanced artificial intelligence techniques, such as machine learning and predictive analytics, these algorithms can analyze vast amounts of data to identify patterns and trends indicative of developing problems. This proactive approach not only enhances the reliability and performance of electric vehicles but also contributes to minimizing downtime and maintenance costs, thereby optimizing the overall efficiency and usability of electric powertrains in various mobility applications.

Test Bench Case Studies

li-Ion cell Thermal runaway

• Anomalous variation in maximum cell temperature trend from 500s.
• Anomaly score rises even if cell temperature value is only 5°C above the nominal one.

Battery Cell Temperature
Anomalous Operation

• 5s freezing of the maximum battery cell temperature during DC current request.
• Anomaly score during freezing shifts to high values.

Stator Temperature
Anomalous Operation

• Electric motor stator temperature, with sudden oscillation from 440s to 445s.
• Strong non-correlation between stator temperature and the operating points.
• Anomaly has been suddenly detected.

Battery health and energy saving are crucial factors to consider for any electric vehicle powertrain systems. A battery that fails unexpectedly can be a major inconvenience, leading to huge performance degradation, costly repairs and lost time. To address this issue, AI-powered algorithms have been developed to monitor battery State of Health and operating conditions in Real-Time. These algorithms provide alerts for battery issues and even predict potential failures before they occur.

Our battery performance and health predictor algorithm has been developed to release power and usable energy increase. Implementing advanced State of Health estimation and reliable safety logics, this AI-based system makes the Electric Powertrain Management System more powerful.

AI algorithms need a huge amount of data to well train the models. Indeed, data acquisition and management for AI based applications are usually critical issues for the model development. This is the reason why, as explained in our recent news, R&D engineers’ efforts have been firstly dedicated to the implementation of reliable and accurate models. In our approach, simulation results can be a source of data to train the AI model, improving and speeding up its training process.

A sneak peek of our Battery Performance Algorithm application is here reported. In the experiment performed, a battery pack internal resistance variation has been triggered. The algorithm analyses the input signals during a specified time buffer and from that moment on, it is able to estimate an overall warning index related to the operating and health conditions of the battery pack.   

Even if the battery electrical and thermal features continue to be inside the acceptable range, our AI algorithm detects the anomaly operation and it is able to trigger both alarms and safety logics if needed.

#Thermal Modelling

Battery pack performances are widely affected by the temperature of the cells. It is basically all related to their chemical and physical nature. Moreover, cell temperature relies on several factors such as cell internal resistance and cell heat exchange.

An accurate battery pack thermal model implies a deep knowledge on thermo and fluid dynamics of the system components, not only inside the battery but also in all the sub-systems working with the heat transfer medium.

The figure below shows the results of an experiments carried out during our validation activities of a racing electric powertrain with a liquid cooled battery pack. Cell and coolant temperatures are modeled with a maximum error of ±1.5°C.

Model inputs:

• battery cell thermal model calibration;
• vehicle speed;
• electric motor torque;
• thermal management system (air cooling, heat pumps or customer based cooling system).

The model is able to simulate the temperature of each cell inside the battery pack, keeping the computational effort low, and allowing the use of the model even in Real-Time applications. The model is flexible and easily tunable to cover the whole range of thermal management solutions.

#Voltage Modelling

Compact models with accurate results is the main goal we aim to accomplish.

In this case, a brief comparison between simulated battery pack voltage and State of Charge with experimental data is shown. Our battery pack model is an easly tunable model that can be used to simulate the electrical performances of HV battery packs composed with different Li-ion cell types.

Thanks to several calibration activities, our model allows us to simulate the battery pack voltage with a relative error lower than 1% with respect to the experimental data and the battery State of Charge with an absolute error even lower than 1%.

The variety of applications in which the model has been deployed enables us to check simulations reliability and precision, verifying the goodness of the results in 800V automotive powertrains architectures and many other.

What distinguishes us from existing solutions:

• Precision and flexibility;
• Complete emulation of electrical and thermal flows;
• Ability to drive any power supply on the market.