Non-invasive Lithium Battery Diagnosis by Magnetic Field Sensing and Machine Learning Assisted Recognition

With the rapid advancement of renewable energy technologies, lithium-ion batteries are increasingly adopted in electric vehicles and energy storage systems. However, battery faults may lead to severe safety incidents, even fire and explosion hazards. In this paper, a non-invasive diagnosis method for lithium-ion batteries, integrating magnetic-field sensing with machine-learning-assisted recognition, is proposed. The magnetic characteristics of cylindrical and pouch-type lithium-ion batteries are analyzed to elucidate the underlying principles of magnetic field generation. A tunnel magnetoresistive sensor is employed to measure the magnetic field distributions of three representative battery states—healthy, aged, and swollen. Since the magnetic field images of batteries under different conditions exhibit significant variations, machine learning algorithms are adopted to extract discriminative features and achieve accurate state identification. The proposed non-invasive, low-cost, and effective method is a promising candidate for diagnosing faults in lithium-ion batteries during quality inspection and recycling processes for electric vehicles and energy storage systems. It can contribute to the reliability of the next generation of power systems.