Real-time prediction of TBM cutter wear rate using pseudo-labeling, multi-task learning, and semi-supervised learning