72 Focus | Thermal runaway prevention May/June 2024 | E-Mobility Engineering risk-prediction algorithms on customer vehicles. As a result, we’ve started to generate warnings for them to manage their fleet in a much safer manner. Additionally, the engineering team has developed a transfer learning methodology to enable deployment of its products to new batteries and fleets much faster by requiring much less, and sometimes no, data from customer’s cells and batteries for characterisation and fine-tuning. “Finally, as a more general and higher-level safety measure that also contains the ability to detect issues that could lead to thermal runaway, we’ve developed a completely unsupervised learning-based approach to anomaly detection to ease deployment into fleets where we have not had the chance to characterise the batteries extensively.” AI in battery testing AI is becoming increasingly important in battery design and testing, particularly in relation to safety in general and to thermal runaway specifically. For example, machine learning is helping engineers create better test plans. Batteries are difficult to simulate using traditional tools, so engineers rely heavily on physical testing to ensure different cell chemistries meet performance requirements and respond safely under varying conditions. Such tools can be used to create models that help focus testing on the most important parameters, ranges and operating conditions to ensure the coverage of critical safety areas. This can help with cell selection and validation testing, and the detection of signals that might lead to a runaway, says an expert from a company that creates such tools. The company’s AI platform enables engineers to create models from lab and field data to look for potentially dangerous anomalies. With deeplearning anomaly detection tools, they can monitor thousands of channels for a wide variety of subtle abnormalities in voltage and thermal measurements. A recent enhancement to the anomaly detection capability allows the parsing of hundreds of test data channels within minutes, along with the inspection of raw test data to ensure it does not contain anything erroneous from faulty sensors, poor measurement logic or sensor placement, or system malfunctions. Using this approach, engineers can quickly identify when batteries undergoing long-running ageing tests are showing early signs of failure and quickly shut them down for repair or retesting. Similarly, measurements from fielded batteries can reveal voltage drifts that may indicate an impending thermal event in time for the battery to be shut down to avoid damage or injury, the expert says. While machine learning-based AI models can only reconstruct battery behaviour from within the design space of values on which the models were trained, as more battery chemistries are tested and tracked with field data, the reach and capability of trained models will grow. “As more thermal runaway events are captured in telemetry data, engineers can create more accurate models to predict and prevent similar events,” the expert concludes. Acknowledgements The author wishes to thank the following for their help with this article: John Williams, vice-president of technical services at Aspen Aerogels; Conor Sheehan, automotive application development engineer at DuPont; Can Kurtulus, CTO of Eatron Technologies; Laila Bonilla, business development manager, and Laure Bertrand, tech service manager, at Elkem; Dr Kim Dana Kreiskoether, specialist battery cell/ system at Henkel; Dr Richard Ahlfeld, founder and CEO of Monolith; Dr Stefan Reichle, EMG advanced manufacturing & market unit manager, alternative mobility at Parker Hannifin; Dr Andy Richenderfer, senior research engineer and Dr Chris Rasik, technology development manager, at Lubrizol; and Shuang Ma, global battery safety platform leader at PPG. Module-based pack architectures provide inherent barriers to particle ejection, while polymer cases can provide extra fire resistance (Image courtesy of LANXESS)
RkJQdWJsaXNoZXIy MjI2Mzk4