Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process
Identifying and categorising gear faults forms an important aspect in predictive maintenance and industrial safety. Traditional methods of fault detection, such as vibration-based analysis, are restricted in terms of sensor placement, high sensitivity to environmental noise, and sheer incapacity to...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
EDP Sciences
2025-01-01
|
| Series: | EPJ Web of Conferences |
| Online Access: | https://www.epj-conferences.org/articles/epjconf/pdf/2025/13/epjconf_icetsf2025_01012.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849422761386049536 |
|---|---|
| author | Bundele Shubham Kane P.V. |
| author_facet | Bundele Shubham Kane P.V. |
| author_sort | Bundele Shubham |
| collection | DOAJ |
| description | Identifying and categorising gear faults forms an important aspect in predictive maintenance and industrial safety. Traditional methods of fault detection, such as vibration-based analysis, are restricted in terms of sensor placement, high sensitivity to environmental noise, and sheer incapacity to identify subtle gear anomalies. To overcome these challenges, the present study employs acoustic data for gear fault diagnosis, transforming temporal sound pressure signals into image representations. Hence, this proposed method gives a systematic analysis of various gear conditions, including cracks, misalignment (under load and no-load conditions), broken teeth, and normal operational conditions, under varying RPM and load conditions. The methodology consists of converting acoustic time-series data to two-dimensional arrays and normalizing them to 8-bit grayscale images, after which the data are categorized based on types of fault. Data diversity is enhanced on these augmented images through image data augmentation using resizing, rotation, flipping, color jittering, and normalization. It is then used to train deep learning models, EfficientNetB0 and EfficientNetB3 that are superior on feature-extraction and computational efficiency. The comparative analysis indicates that EfficientNetB3 outperforms EfficientNetB0 based on all four metrics of accuracy, precision, recall, and overall classification performance. Model validation is conducted using k-fold cross Validation to ensure robustness and generalizability. This research proves that acoustic-based fault analysis combined with advanced deep learning models achieves capture of efficiency, compared to conventional vibration-based diagnostics. The proposed method improves early fault detection, provides an accurate classification coupled with a non-intrusive, scalable solution for industrial gear health monitoring and thus takes a step forward toward advancement in predictive maintenance strategies in mechanical systems. |
| format | Article |
| id | doaj-art-bc9163ce62a84c828b41c265db0aaa8f |
| institution | Kabale University |
| issn | 2100-014X |
| language | English |
| publishDate | 2025-01-01 |
| publisher | EDP Sciences |
| record_format | Article |
| series | EPJ Web of Conferences |
| spelling | doaj-art-bc9163ce62a84c828b41c265db0aaa8f2025-08-20T03:30:56ZengEDP SciencesEPJ Web of Conferences2100-014X2025-01-013280101210.1051/epjconf/202532801012epjconf_icetsf2025_01012Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning ProcessBundele Shubham0Kane P.V.1Visvesvaraya National Institute of Technology (VNIT)Visvesvaraya National Institute of Technology (VNIT)Identifying and categorising gear faults forms an important aspect in predictive maintenance and industrial safety. Traditional methods of fault detection, such as vibration-based analysis, are restricted in terms of sensor placement, high sensitivity to environmental noise, and sheer incapacity to identify subtle gear anomalies. To overcome these challenges, the present study employs acoustic data for gear fault diagnosis, transforming temporal sound pressure signals into image representations. Hence, this proposed method gives a systematic analysis of various gear conditions, including cracks, misalignment (under load and no-load conditions), broken teeth, and normal operational conditions, under varying RPM and load conditions. The methodology consists of converting acoustic time-series data to two-dimensional arrays and normalizing them to 8-bit grayscale images, after which the data are categorized based on types of fault. Data diversity is enhanced on these augmented images through image data augmentation using resizing, rotation, flipping, color jittering, and normalization. It is then used to train deep learning models, EfficientNetB0 and EfficientNetB3 that are superior on feature-extraction and computational efficiency. The comparative analysis indicates that EfficientNetB3 outperforms EfficientNetB0 based on all four metrics of accuracy, precision, recall, and overall classification performance. Model validation is conducted using k-fold cross Validation to ensure robustness and generalizability. This research proves that acoustic-based fault analysis combined with advanced deep learning models achieves capture of efficiency, compared to conventional vibration-based diagnostics. The proposed method improves early fault detection, provides an accurate classification coupled with a non-intrusive, scalable solution for industrial gear health monitoring and thus takes a step forward toward advancement in predictive maintenance strategies in mechanical systems.https://www.epj-conferences.org/articles/epjconf/pdf/2025/13/epjconf_icetsf2025_01012.pdf |
| spellingShingle | Bundele Shubham Kane P.V. Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process EPJ Web of Conferences |
| title | Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process |
| title_full | Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process |
| title_fullStr | Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process |
| title_full_unstemmed | Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process |
| title_short | Design of an Improved Model for Gear Fault Diagnosis Using Acoustic Data and EfficientNet-Based Deep Learning Process |
| title_sort | design of an improved model for gear fault diagnosis using acoustic data and efficientnet based deep learning process |
| url | https://www.epj-conferences.org/articles/epjconf/pdf/2025/13/epjconf_icetsf2025_01012.pdf |
| work_keys_str_mv | AT bundeleshubham designofanimprovedmodelforgearfaultdiagnosisusingacousticdataandefficientnetbaseddeeplearningprocess AT kanepv designofanimprovedmodelforgearfaultdiagnosisusingacousticdataandefficientnetbaseddeeplearningprocess |