Health and performance diagnostics in Li-ion batteries with pulse-injection-aided machine learning

Performance metric diagnostics of lithium-ion batteries are important for electric vehicles. A novel diagnostics method during vehicle charging is proposed using a feedforward neural network and the battery voltage response to a current pulse perturbation, hence the name 'pulse-injection-aided machine learning' (PIAML). Performance metrics are quantified using state of health and state of power, representing capacity and power fade. Data is collected for lithium-ion battery cells at various states and pulsing scenarios, resulting in 5,184 unique voltage responses for evaluating the technique. PIAML is shown to estimate states of health and power with high fidelity, and can also be used to initialize the state of charge. In the best-case, average trial error is 0.0057 for state of health estimation, 0.0069 for power, and 0.0072 for charge. Neither charging history nor battery parameters are required, and diagnostics can be performed in less than 3 min. Results show that PIAML is a high-accuracy general-purpose technique with potential for wider applications.

Automated summary

This paper proposes a novel diagnostics method called 'pulse-injection-aided machine learning' (PIAML) for lithium-ion batteries during vehicle charging. The PIAML method utilizes a feedforward neural network and the battery voltage response to accurately estimate the states of health, power, and charge, without requiring charging history or battery parameters.