Forward and reverse design of adhesive in batteries via dynamics and machine learning algorithms for enhanced mechanical safety.

The growing popularity of electric vehicles brings opportunities and challenges to the battery industry. Designers need to develop reliable battery packs to ensure the safety of consumers' property and passengers' lives. Due to the complex structure of the battery pack, the traditional finite element analysis design consumes a lot of computational resources. The utilization of multibody system dynamics (MSD) and machine learning (ML) methods can assist developers in the efficient design of reliable battery packs. In this work, an MSD model of a battery pack was constructed based on the recursive idea, which can characterize the state information of each cell, such as velocity, acceleration, deformation, etc., during extrusion. By utilizing ML techniques, it is possible to achieve both the forward and reverse design of the adhesive for the battery pack. This enables accurate prediction of battery deformation under various adhesive stiffness and damping coefficients, as well as different battery SOCs. Consequently, the design of the battery adhesive can be guided, resulting in minimal distortion of the battery pack during extrusion and reducing the risk of internal short circuits. This method enables efficient battery pack design and provides ideas for future reliable battery pack designs. • An MSD model of the battery pack was proposed. • Six ML models were utilized to predict the extrusion deformation. • The DNN model which performs best was used to optimize the adhesive. • This method provides ideas for reliable battery pack design. [ABSTRACT FROM AUTHOR]