Enhanced Deep Learning Approaches for Diagnosing Drilling Machine Failures Using Gramian Angular Field and Markov Transition Field Encoding

• 出版年: Journal of Applied Science and Engineering
• 主要な著者: Aroui Tarek, Dorbez Fradj
• フォーマット: 論文
• 言語: 英語
• 出版事項: Tamkang University Press 2025-02-01
• 主題: valmet ab drilling machine; gramian angular field; markov transition field; resnet50; inceptionv3
• オンライン･アクセス: http://jase.tku.edu.tw/articles/jase-202509-28-09-0004
• ISSN: 2708-9967; 2708-9975

In the era of Industry 4.0, applying deep learning models for analyzing sensor data in machinery is a fundamental step toward developing predictive maintenance strategies. Deep learning models can automatically learn complex patterns and features from sensor data, making them highly effective in identifying early signs of faults or anomalies in machinery. This capability is often beyond the reach of traditional analysis methods (e.g., statistical features), which may miss subtle or non-linear patterns. By identifying these issues early, deep learning models enable proactive scheduling of maintenance activities, reducing unplanned downtime and preventing catastrophic failures. This study adopts a specific approach by focusing on diagnosing failures in the VALMET AB drilling machine. We use sound signals captured by AudioBox iTwo Studio microphones as our primary data source. The dataset, which includes 134 sounds categorized into Anomaly and Normal classes, is augmented using advanced techniques. We then employ the Markov Transition Field and the Gramian Angular Field encoding methods to represent the sound signals as images. These encoded images are subsequently used to train two deep-learning models with distinct architectures: ResNet50 and InceptionV3. The study’s results are promising, affirming the efficacy of our approach in detecting and diagnosing failures in drilling machines.