Transferring Vision-Language-Action Models to Industry Applications: Architectures, Performance, and Challenges