摘要详情

ID / 提交时间

194 / 2025-06-13 17:24:40

标题

Deep Learning Data-Driven Thresholding Models Incorporating Box-Cox Transforms

关键字

gas turbine, anomaly analysis, deep learning, thresholding, S-Golay filter, B-CT, DL-DDBC.

主题及专题

4、System Monitoring, Fault Diagnosis and Prediction

状态

终稿

作者

Jingyi Hu / Harbin Engineering University

Yunpeng Cao / Harbin Engineering University

Xiaoyu Han / Harbin Engineering University

Weiying Wang / China Shipbuilding Industry Corporation 703 Research Institute

Chunhui Li / Harbin Engineering University

Weixing Feng / Harbin Engineering University

摘要

Abstracts—Aiming at the problems of high false alarm rates and poor adaptability of the traditional monitoring threshold establishment method based on empirical formulae, this paper proposes a data-driven threshold construction method (DL-DDBC) that integrates Box-Cox transform (B-CT) and deep learning. The method firstly constructs a steady state operating condition data set based on the target rotational speed screening and adopts the Savitzky-Golay filter (S-Golay filter) to achieve signal noise reduction. Then, the target benchmark data after noise reduction is tested for normality, and the feature thresholds are directly extracted from the data that meets the normal distribution, while the non-normal data is firstly corrected by B-CT, and the inverse B-CT is performed after extracting the eigenvalues of the corrected data; after that, a deep learning model is constructed to extract the benchmark eigenvalues of the data that characterize the health state of the equipment; finally, the threshold boundary is constructed by applying the improved threshold boundary generating algorithm. By the data analysis process designed in this paper, experiments are conducted with the gas turbine acoustic pressure signal as the research object, and the results show that the constructed threshold intervals have significant adaptability to the working conditions and anti-noise ability, and provide reliable quantitative evidence for real-time monitoring of gas turbine operation status.