Data Analysis Using Artificial Intelligence Methods

In the field of data analysis using artificial intelligence methods, we use both small and big data approaches to discover new correlations and property relationships in the data generated by our laboratory infrastructure. In addition, we analyze market and industry data with our partners and improve operations strategies and quality assurance procedures. Our in-depth domain knowledge, which spans all relevant areas of battery research – from material development and electrochemical processes to production and operating strategies – is incorporated into the interpretation of this data.

We use advanced signal and data processing methods to detect new effects and correlations using artificial intelligence. We use computer vision models to segment and classify objects and anomalies and analyze time series data using principle component analysis, independent component analysis and other signal processing methods.

Classification of Anomalies and Defects

Schematische Darstellung der Defektanalyse von Batteriezellen
© Fraunhofer ISE
Schematische Darstellung der Defektanalyse von Batteriezellen mittels Scanning Acoustic Microscopy: Die KI-basierte Detektion von Anomalien kann sowohl in Bilddaten als auch in Zeitreihendaten erfolgen.

By applying artificial intelligence methods to data analysis, we can identify, locate, and classify anomalies and defects in battery cells or their components. To do this, we use advanced technologies such as computed tomography and scanning acoustic microscopy to capture large data sets.  

We analyze gray-scale images using semantic segmentation, which we preprocess using image filtering. This process divides an image into regions that are assigned to specific classes, such as defects or normal areas.  

In parallel, we process time series data directly. Here, we use, for example, principal component analysis (PCA) as a method for feature extraction. In addition to PCA, we use supervised and unsupervised learning approaches to identify anomalies directly in the time series data.  

Machine Learning for Battery Material Evaluation and High-Throughput Experiments

Neuronale Netzwerke zur Materialentwicklung
© Warod - stock.adobe.com
Neuronale Netzwerke kommen bei der Entwicklung neuer Batteriematerialien zum Einsatz.

The development and evaluation of optimized or novel battery materials is often a challenging and resource-intensive undertaking. To successfully transfer and scale expert and domain knowledge, we develop machine learning (ML) models that accurately predict battery material properties. Machine learning-assisted experiment planning can make experimental investigation and validation more efficient and targeted.  

Datasets with limited experimental information (small data) can be complemented and extended by computational methods such as density functional theory (DFT) and chemoinformatics.

Condition Monitoring

Grafischer Vergleich von neuronalen Architekturen
© Fraunhofer ISE
Grafischer Vergleich von neuronalen Architekturen zur Vorhersage des Ladezustands der Batterie Ladezustandes (SOC) über die Zeit.

To determine indirect parameters such as state of charge (SOC), internal resistance, state of health (SOH) or the safety status (SOS) of a battery storage unit or cell, complex methods are needed that determine these based on the fundamental parameters (voltage, current and temperature of the battery cell). Especially with the high number of cells in modern battery systems, processing a large number of values in real time is challenging. In addition to model-based state determination, ML models are therefore increasingly being used. Fraunhofer ISE is therefore working with researchers in the field of artificial intelligence to develop and validate AI models to determine the state of various battery types and applications precisely and with efficient computing.  

Cell Characterization

Vorhersage des DRT-Verlaufs
© Fraunhofer ISE
Vorhersage des DRT-Verlaufs aus einem synthetischen Messpuls und Vergleich mit dem zugrundeliegenden Verhalten.

The overvoltages under electrical load and the internal resistance of a cell are very characteristic and correlate with various electrochemical effects and influences (e.g. temperature or aging). Although the internal resistance can be measured quickly compared to the storage capacity, the evaluation and differentiation of the individual internal resistance components and their interpretation, however, poses a major challenge. Fraunhofer ISE is therefore developing an ML tool that is able to determine the internal resistance components from a current pulse using the distribution of relaxation times (DRT). 

R&D Infrastructure

At Fraunhofer ISE, we have access to the following infrastructure for our research and development activities:

 

Center for Electrical Energy Storage

From novel materials and production technologies for battery cells, to battery system design and safety testing, to integration – the “Center for Electrical Energy Storage” offers a unique research infrastructure along the entire value chain of batteries.

Selected Research Projects

 

BatterieDigital_real

Pilot Project for the Fraunhofer Research Data Space

 

SAMBA

Scanning Acoustic Microscopy-based Battery Analysis

 

Quaze

Optical Test Method for Determining the Quality of Battery Cells