<p>Early and accurate fault diagnosis in rotating machinery is essential to prevent unplanned breakdowns, costly downtime, and safety risks. Conventional fault detection methods are usually based on contact sensors or sophisticated signal processing, which may be intricate, expensive, or unreliable in harsh industrial environments. To overcome these drawbacks, this work introduces an Extended-Adaptive Neuro-Fuzzy Inference System (E-ANFIS) approach for fault diagnosis in rotating machine components by utilizing infrared thermography data from the machine. The method is validated by developing a real-time experimental setup consisting of a motor-driven rotating system with key components, including a bearing, bearing housing, rotating shaft, and belt-pulley mechanism. This test uses a FLIR ONE PRO LT iOS Pro-Grade infrared thermography camera to take thermal images of these components when they work in different operating conditions. The system is configured to run for 16 continuous hours, first under healthy conditions, where recorded temperature includes <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="36" /> </InlineMediaObject> <EquationSource Format="TEX">$47^\circ \text{C }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mn>47</mn> <mo>∘</mo> </msup> <mtext>C</mtext> <mspace width="0.333333em" /> </mrow> </math></EquationSource> </InlineEquation> for a healthy bearing, <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq2.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="48" /> </InlineMediaObject> <EquationSource Format="TEX">$49.5^\circ \text{C }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>49</mn> <mo>.</mo> <msup> <mn>5</mn> <mo>∘</mo> </msup> <mtext>C</mtext> <mspace width="0.333333em" /> </mrow> </math></EquationSource> </InlineEquation> for a scratched bearing, <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq3.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="36" /> </InlineMediaObject> <EquationSource Format="TEX">$51^\circ \text{C}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mn>51</mn> <mo>∘</mo> </msup> <mtext>C</mtext> </mrow> </math></EquationSource> </InlineEquation> for a healthy V-belt, and <InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq4.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="35" /> </InlineMediaObject> <EquationSource Format="TEX">$45^\circ{\rm C}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mn>45</mn> <mo>∘</mo> </msup> <mi mathvariant="normal">C</mi> </mrow> </math></EquationSource> </InlineEquation> for an aligned shaft. Therafter, the system is then tested under fault conditions, where a broken ball bearing reaches <InlineEquation ID="IEq5"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq5.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="48" /> </InlineMediaObject> <EquationSource Format="TEX">$65.5^\circ \text{C }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>65</mn> <mo>.</mo> <msup> <mn>5</mn> <mo>∘</mo> </msup> <mtext>C</mtext> <mspace width="0.333333em" /> </mrow> </math></EquationSource> </InlineEquation>, a cracked outer ring <InlineEquation ID="IEq6"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq6.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="36" /> </InlineMediaObject> <EquationSource Format="TEX">$63^\circ \text{C }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mn>63</mn> <mo>∘</mo> </msup> <mtext>C</mtext> <mspace width="0.333333em" /> </mrow> </math></EquationSource> </InlineEquation> , a loose V-belt <InlineEquation ID="IEq7"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq7.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="48" /> </InlineMediaObject> <EquationSource Format="TEX">$68.5^\circ \text{C }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>68</mn> <mo>.</mo> <msup> <mn>5</mn> <mo>∘</mo> </msup> <mtext>C</mtext> <mspace width="0.333333em" /> </mrow> </math></EquationSource> </InlineEquation> , and a misaligned shaft <InlineEquation ID="IEq8"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="12647_2025_838_Article_IEq8.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="36" /> </InlineMediaObject> <EquationSource Format="TEX">$67^\circ \text{C }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mn>67</mn> <mo>∘</mo> </msup> <mtext>C</mtext> <mspace width="0.333333em" /> </mrow> </math></EquationSource> </InlineEquation>. The method is validated by using five performance parameters, namely accuracy, sensitivity, precision, Jaccard Similarity Index (JSI), and Dice Similarity Index (DSI). The proposed E-ANFIS model achieves an accuracy of 96.19, 94.29 sensitivity, 99.91% JSI, and 97.02% DSI, proving its effectiveness as a non-contact, real-time defect detection method. Additionally, the proposed method is compared with four existing defect detection methods, namely, Conventional ANFIS, Linear Transformation, and Improved Hyper Smoothing based Local Binary Pattern (LTIHLBP), Fuzzy-integrated Local Binary Pattern (Fuzzy-ILBP), and Modified Local Binary Pattern (MANN) method. Although, the accuracy parameter from E-ANFIS is 96.19%, which is lower in comparison to LTIHLBP, Fuzzy-ILBP, and MANN approaches. However, E-ANFIS outperforms all above methods in terms of sensitivity, precision, JSI, and DSI.</p>

The Fault Diagnosis of Different Rotating Machine Elements by Using Infrared Thermography Images and Extended Adaptive Neuro-Fuzzy Inference System: An Experimental Evaluation

期刊论文
发表日期：

2025年7月17日

查看全文

Ekta Yadav,
Viveak Kumar Chawla,
Surjit Angra,
Sanjay Yadav

摘要

Early and accurate fault diagnosis in rotating machinery is essential to prevent unplanned breakdowns, costly downtime, and safety risks. Conventional fault detection methods are usually based on contact sensors or sophisticated signal processing, which may be intricate, expensive, or unreliable in harsh industrial environments. To overcome these drawbacks, this work introduces an Extended-Adaptive Neuro-Fuzzy Inference System (E-ANFIS) approach for fault diagnosis in rotating machine components by utilizing infrared thermography data from the machine. The method is validated by developing a real-time experimental setup consisting of a motor-driven rotating system with key components, including a bearing, bearing housing, rotating shaft, and belt-pulley mechanism. This test uses a FLIR ONE PRO LT iOS Pro-Grade infrared thermography camera to take thermal images of these components when they work in different operating conditions. The system is configured to run for 16 continuous hours, first under healthy conditions, where recorded temperature includes $47^\circ \text{C }$ $47^{\circ} C$ for a healthy bearing, $49.5^\circ \text{C }$ $49 . 5^{\circ} C$ for a scratched bearing, $51^\circ \text{C}$ $51^{\circ} C$ for a healthy V-belt, and $45^\circ{\rm C}$ $45^{\circ} C$ for an aligned shaft. Therafter, the system is then tested under fault conditions, where a broken ball bearing reaches $65.5^\circ \text{C }$ $65 . 5^{\circ} C$ , a cracked outer ring $63^\circ \text{C }$ $63^{\circ} C$ , a loose V-belt $68.5^\circ \text{C }$ $68 . 5^{\circ} C$ , and a misaligned shaft $67^\circ \text{C }$ $67^{\circ} C$ . The method is validated by using five performance parameters, namely accuracy, sensitivity, precision, Jaccard Similarity Index (JSI), and Dice Similarity Index (DSI). The proposed E-ANFIS model achieves an accuracy of 96.19, 94.29 sensitivity, 99.91% JSI, and 97.02% DSI, proving its effectiveness as a non-contact, real-time defect detection method. Additionally, the proposed method is compared with four existing defect detection methods, namely, Conventional ANFIS, Linear Transformation, and Improved Hyper Smoothing based Local Binary Pattern (LTIHLBP), Fuzzy-integrated Local Binary Pattern (Fuzzy-ILBP), and Modified Local Binary Pattern (MANN) method. Although, the accuracy parameter from E-ANFIS is 96.19%, which is lower in comparison to LTIHLBP, Fuzzy-ILBP, and MANN approaches. However, E-ANFIS outperforms all above methods in terms of sensitivity, precision, JSI, and DSI.