Fault diagnosis based on machine learning has been widely used in the Industrial Internet of Things (IIoT). However, due to variation of operating conditions, training and testing data are drawn from different distributions. Existing methods usually need prior knowledge of the relationship between label space of the source and target domains and suffer the negative transfer problem if the knowledge is unavailable. To handle these issues, a novel transferability weighted universal domain adaptation network (TWUAN) is proposed for the universal domain adaptation problem. TWUAN consists of a weighted adversarial module and a weighting learning module. Among them, a transferability measure is embedded into the adversarial domain adaptation network, which helps TWUAN classifies samples in the common label space correctly and identifies unknown classes. Experiments on the CWRU dataset demonstrate that the proposed TWUAN can bridge the distribution discrepancy and achieve satisfactory diagnosis accuracy in different settings.