Enhancing Electric Vehicle Battery Pack Assembly with YOLOv8-Based Real-time Detection and Bayesian Optimization

In the assembly of the battery pack for an Electric Vehicle (EV) ensuring the correct placement of key components such as battery modules, module connectors, thermal insulation, etc., is crucial for operational efficiency and safety. Out of these components mica plates, which are used for thermal insulation, and busbars, which are used for connecting one module to another, stand out in terms of their utility and safety features. However, during assembly, mica plates or busbars are occasionally missing from their predefined position, and this can lead to significant operational inefficiencies and safety risks. If any of these two parts is missing, then it may cause damage not only to the module but also to other parts in operation. So, the detection of missing parts early in the assembly process is important to address this issue of detecting missing parts efficiently. Thus, here the authors explore the utility of vision systems with Deep Learning (DL) algorithms as they have proven to have a better edge over traditional vision systems. YOLOv8 (You Only Look Once) was utilised due to its exceptional speed and accuracy in object detection, which is essential in high-paced operations to enhance the model’s performance and robustness. Also, a series of data augmentation steps such as geometric transformations, lighting adjustments, etc. to create a data set big enough to ensure the model’s effectiveness under varying conditions and assembly configurations, and Bayesian Optimization was used for selecting the hyperparameters. To analyze the model’s performance four parameters namely mAP-50, Precision, Recall, and mAP50-95, were compared