PGPengfei Gou

Pengfei Gou

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Pengfei Gou has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Pre-trained transformer with contrastive learning and regression constraints for remaining useful life prediction

    by Yifei Tang, Pengfei Gou, Xuyun Nie, Mengjuan Liu

    Condition-based maintenance (CBM) aims to plan maintenance activities through continuous monitoring and prognostics of equipment operating conditions, thereby reducing maintenance costs and downtime. remaining useful life (RUL) prediction is one of the core enabling technologies of CBM, and data-driven approaches have become a major research focus in this field. To overcome the limitations of existing methods in learning effective health representations, this paper proposes a pre-training and fine-tuning Transformer framework based on contrastive learning with regression constraints, designed to learn continuous and highly predictable health representations. During the pre-training phase, this method introduces RUL regression constraints on top of supervised contrastive learning to prevent the model from relying solely on relative relationships between samples, thereby ignoring the absolute numerical differences in health status along the lifespan dimension. Additionally, we employ an attention-based pooling mechanism to aggregate features from multiple time steps, allowing the model to adaptively emphasize time-step features that are most relevant to the current degradation state, thus obtaining more comprehensive representations of the global health status. Experimental results on the turbofan engine datasets C-MAPSS and N-CMAPSS demonstrate that our proposed method outperforms existing baseline models on multiple evaluation metrics, proving its effectiveness in the RUL prediction task. All codes have been made available on GitHub: https://github.com/ZeyU1999/RUL-Prediction/tree/master .

  • A Two-Phase Equipment Remaining Useful Life Prediction Model With Sequence-Feature Attention Mechanism

    by Mengjuan Liu, Daiwei Zheng, Pengfei Gou, Junhao Zhang, Zhi Lai, Dajiang Chen

    Predicting the Remaining Useful Life (RUL) is crucial in Prognostics and Health Management (PHM). By utilizing real-time data from equipment sensors to predict RUL, the PHM system can plan proactive maintenance tasks, which helps reduce maintenance costs and downtime. While deep learning has been successful in predicting RUL, current mainstream RUL prediction models have limitations in extracting operating state information from time-series data. Additionally, the limited availability of training data poses a significant challenge to RUL research. This paper introduces a new model for predicting RUL for multi-sensor equipment. The model employs a two-phase training process. First, multiple datasets are combined to pre-train the model to learn representations of equipment operating states. Following pre-training, a single dataset is used to fine-tune the pre-trained model to tailor it to specific working conditions. The new model also incorporates a sequence-feature attention mechanism to integrate the multi-dimensional time-series data. It assigns varying weights to different feature parameters at different time steps. Finally, we have evaluated the proposed model on the C-MAPSS and N-CMAPSS datasets, and the results demonstrate that it surpasses existing state-of-the-art models in most cases. All codes and results have been made available on GitHub to assist researchers in replicating our experiments quickly. (https://github.com/mjliu-advertising/RUL).