Scholarlink

Background <p>Shafts are essential components in mechanical systems because they transmit power and motion. Damage to shafts can reduce system efficiency, degrade performance, and eventually lead to failure. Vibration-based condition monitoring has been widely used for damage detection, but the effectiveness of machine learning models depends on appropriate feature extraction and data preprocessing.</p> Methods <p>This study evaluates the performance of Artificial Neural Networks (ANN) and Support Vector Machines (SVM) for shaft damage classification using vibration characteristics. Experimental tests were conducted on AISI 1045 steel shafts with a length of 250 mm and a diameter of 16 mm under three conditions: intact, 10% damage, and 25% damage. The first and second natural frequencies extracted from vibration signals were used as input features. The dataset consisted of 150 samples for each condition and was divided into 80% training and 20% testing data. Two preprocessing methods, Damage Index (DI) and Standard Scaler normalization, were applied.</p> Results <p>Standard Scaler normalization improved the performance of both models compared with DI-based normalization. SVM achieved the highest classification accuracy of 98.9%, while ANN reached 97.8%. The extracted natural frequencies successfully distinguished the three shaft conditions.</p> Conclusions <p>The findings demonstrate that machine learning combined with appropriate preprocessing can provide reliable and accurate shaft damage detection. SVM with Standard Scaler normalization showed the best overall performance for vibration-based shaft condition monitoring.</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

Vibration-Based Machine Learning Approach for Shaft Damage Detection

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Berli Paripurna Kamiel,
  • Ahmad Habib Rizqi,
  • Sunardi Sunardi

摘要

Background

Shafts are essential components in mechanical systems because they transmit power and motion. Damage to shafts can reduce system efficiency, degrade performance, and eventually lead to failure. Vibration-based condition monitoring has been widely used for damage detection, but the effectiveness of machine learning models depends on appropriate feature extraction and data preprocessing.

Methods

This study evaluates the performance of Artificial Neural Networks (ANN) and Support Vector Machines (SVM) for shaft damage classification using vibration characteristics. Experimental tests were conducted on AISI 1045 steel shafts with a length of 250 mm and a diameter of 16 mm under three conditions: intact, 10% damage, and 25% damage. The first and second natural frequencies extracted from vibration signals were used as input features. The dataset consisted of 150 samples for each condition and was divided into 80% training and 20% testing data. Two preprocessing methods, Damage Index (DI) and Standard Scaler normalization, were applied.

Results

Standard Scaler normalization improved the performance of both models compared with DI-based normalization. SVM achieved the highest classification accuracy of 98.9%, while ANN reached 97.8%. The extracted natural frequencies successfully distinguished the three shaft conditions.

Conclusions

The findings demonstrate that machine learning combined with appropriate preprocessing can provide reliable and accurate shaft damage detection. SVM with Standard Scaler normalization showed the best overall performance for vibration-based shaft condition monitoring.