Large and expensive mechanical equipment such as wind turbines generally has limited fault datasets from real-world operations for digital model development. This often leads to poor accuracy in implementing a model based the life prediction. To address this data shortage issue in developing deep learning models, a remaining useful life prediction approach is proposed in this paper, which combines digital twin technology with transfer learning theory and embedded convolutional long short-term memory extended model. Firstly, the main bearing of a direct-drive wind turbine is mapped to the digital world by digital twin technology, allowing for the fault datasets of main bearings to be generated and thereby ensuring the model to be trained sufficiently with a balanced dataset. The convolutional long short-term memory network then performs convolutional operations on input-to-state and state-to-state transitions, thereby integrating the time dependence and time-frequency characteristics of the data. In the meantime, a transfer learning was used to transfer the trained model to the wind field for real-word fault diagnostics and the life prediction of main bearings. Finally, the approach is applied to predicting the life of the main bearings, which is also compared with other methods of similar types. The results verified that the proposed approach can effectively overcome the low data density of large equipment, greatly improving the accuracy of life prediction.

In industry, accurate remaining useful life (RUL) prediction is critical in improving system reliability and reducing downtime and accident risk. Numerous data-driven RUL prediction approaches have been proposed and achieved impressive performance in RUL prediction. However, most of them are still faced with the dilemma of limited samples, and most of popular transfer learning and domain adaptive methods adopt single-source domain adaptation (DA), ignoring the domain-shift within source domain and failing to fully utilize the multi-condition data. This article proposes a model-data fusion life prediction method based on digital twin (DT) and multi-source regression adversarial domain adaptation (MRADA) to address the aforementioned issues. For data-driven life prediction model, the model-based DT technology among them offers a significant amount of multi-condition training data. The proposed MRADA fully utilizes the benefits of DT simulation data by using intra-group alignment strategy, inter-group alignment strategy, adversarial learning, and regressor alignment strategy to learn domain-invariant features and supervision from multiple sources. The experimental findings demonstrate that the proposed fusion life prediction method can successfully address the issue of small samples and improve the accuracy of rolling bearing life prediction results.INDEX TERMS Digital twin, domain adaptation, remaining useful life prediction, small sample size.

No abstract