Scholarlink

Purpose <p>Machine learning algorithms demonstrate a strong potential for condition monitoring&#xa0;and fault detection of industrial equipment. However, their reliance on experimental data&#xa0;for training limits their applicability in industry.</p> Method <p>To address this limitation, this paper proposes a novel approach based on hybrid data, combining available experimental data with simulated data generated by a digital twin model. The construction of such hybrid dataset presents a challenging task to achieve a high accuracy. Two ways of data enrichment are considered: (i) increasing the volume of available states (quantitative enrichment) and (ii) adding unavailable states (qualitative enrichment). The approach’s performance is evaluated using different well-established classifiers.</p> Results <p>Results show an improvement in accuracy compared with the approach using solely simulated data. The accuracy is higher than 90<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\%\)</EquationSource> </InlineEquation> with 20<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\%\)</EquationSource> </InlineEquation> of experimental data for quantitative enrichment and with 50<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\%\)</EquationSource> </InlineEquation> for qualitative enrichment.</p> Conclusion <p>In view of the results, this study provides valuable insights into how the digital twin can be used as a means of enriching data for gear diagnosis.</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

Towards Accurate Gear Fault Diagnosis: Hybrid Data Enrichment by Digital Twins for Machine Learning-Based Monitoring

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Gauthier Ngandu Kalala,
  • Abir Boujelben,
  • Xavier Chiementin,
  • Lanto Rasolofondraibe

摘要

Purpose

Machine learning algorithms demonstrate a strong potential for condition monitoring and fault detection of industrial equipment. However, their reliance on experimental data for training limits their applicability in industry.

Method

To address this limitation, this paper proposes a novel approach based on hybrid data, combining available experimental data with simulated data generated by a digital twin model. The construction of such hybrid dataset presents a challenging task to achieve a high accuracy. Two ways of data enrichment are considered: (i) increasing the volume of available states (quantitative enrichment) and (ii) adding unavailable states (qualitative enrichment). The approach’s performance is evaluated using different well-established classifiers.

Results

Results show an improvement in accuracy compared with the approach using solely simulated data. The accuracy is higher than 90 \(\%\) with 20 \(\%\) of experimental data for quantitative enrichment and with 50 \(\%\) for qualitative enrichment.

Conclusion

In view of the results, this study provides valuable insights into how the digital twin can be used as a means of enriching data for gear diagnosis.