Scholarlink

Background <p>Gearboxes are critical components in industrial power transmission systems, and their unexpected failure can lead to operational disruption, increased maintenance cost, and reduced system reliability. Prognostics and Health Management (PHM) has emerged as an important strategy for enabling condition-based maintenance through remaining useful life (RUL) prediction. However, reliable gearbox RUL estimation is still difficult due to heterogeneous signal behaviour, irregular sampling, dataset variations, and the dependence of many methods on expert-defined health indicators that do not always exhibit monotonic degradation.</p> Purpose <p>This study presents an adaptive Bayesian framework for gearbox RUL prediction that is designed to overcome these challenges and remain effective under realistic monitoring constraints.</p> Methods <p>The proposed framework begins with vibration signal denoising using discrete wavelet transform, followed by extraction of time- and frequency-domain features that are filtered through monotonicity-based selection to identify trend-consistent descriptors. These features are normalized using the Feature Degradation Ratio to construct a cumulative Health Indicator (HI), which is then modelled through Adaptive Bayesian Regression RUL (ABR-RUL) estimation. This method was validated using a single-stage gearbox run-to-failure dataset under six different monitoring conditions. These included dense, moderate, and irregular sampling, along with early, mid-life, and late-life truncations to simulate practical industrial scenarios.</p> Results <p>The model accurately predicted failure at 410.89 h compared to the actual 407 h, yielding a deviation of less than 1%. For the dense sampling dataset, RMSE was approximately 0.11 with MAE of 0.04, while moderate and irregular sampling achieved RMSE around 0.12 and MAE below 0.05. The truncated datasets showed RMSE of 0.25, 0.16, and 0.12 with corresponding MAE of 0.18, 0.07, and 0.05.</p> Conclusions <p>The results confirm the robustness and effectiveness of the proposed framework for gearbox RUL prediction dense, moderate, irregular, and truncated monitoring conditions, supporting its applicability in practical PHM environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-Resilient Condition Monitoring of Gearbox using Adaptive Bayesian Regression

  • 期刊论文

  • 发表日期：

    2026年3月28日

查看全文
  • Anka Datta,
  • Sabareesh Geetha Rajasekharan,
  • Vamsi Inturi,
  • S V V S Narayana Pichika

摘要

Background

Gearboxes are critical components in industrial power transmission systems, and their unexpected failure can lead to operational disruption, increased maintenance cost, and reduced system reliability. Prognostics and Health Management (PHM) has emerged as an important strategy for enabling condition-based maintenance through remaining useful life (RUL) prediction. However, reliable gearbox RUL estimation is still difficult due to heterogeneous signal behaviour, irregular sampling, dataset variations, and the dependence of many methods on expert-defined health indicators that do not always exhibit monotonic degradation.

Purpose

This study presents an adaptive Bayesian framework for gearbox RUL prediction that is designed to overcome these challenges and remain effective under realistic monitoring constraints.

Methods

The proposed framework begins with vibration signal denoising using discrete wavelet transform, followed by extraction of time- and frequency-domain features that are filtered through monotonicity-based selection to identify trend-consistent descriptors. These features are normalized using the Feature Degradation Ratio to construct a cumulative Health Indicator (HI), which is then modelled through Adaptive Bayesian Regression RUL (ABR-RUL) estimation. This method was validated using a single-stage gearbox run-to-failure dataset under six different monitoring conditions. These included dense, moderate, and irregular sampling, along with early, mid-life, and late-life truncations to simulate practical industrial scenarios.

Results

The model accurately predicted failure at 410.89 h compared to the actual 407 h, yielding a deviation of less than 1%. For the dense sampling dataset, RMSE was approximately 0.11 with MAE of 0.04, while moderate and irregular sampling achieved RMSE around 0.12 and MAE below 0.05. The truncated datasets showed RMSE of 0.25, 0.16, and 0.12 with corresponding MAE of 0.18, 0.07, and 0.05.

Conclusions

The results confirm the robustness and effectiveness of the proposed framework for gearbox RUL prediction dense, moderate, irregular, and truncated monitoring conditions, supporting its applicability in practical PHM environments.