Non-contact intelligent fault diagnosis based on thermography, unsupervised feature modelling and deep-feature learning for assessing faults in electromechanical systems
Industrial processes employ different configurations of electromechanical systems (ES) to perform specific tasks, indeed, the operativity and continuity of processes are subjected to machine availability. In this regard, the proposal of novel condition monitoring strategies remains mandatory in orde...
| Autores: | , , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/443764 |
| Acceso en línea: | https://hdl.handle.net/2117/443764 https://dx.doi.org/10.1016/j.infrared.2025.106039 |
| Access Level: | acceso embargado |
| Palabra clave: | Electromechanical system Thermography Intelligent fault detection Maps Deep auto encoder Àrees temàtiques de la UPC::Enginyeria elèctrica::Electromecànica |
| Sumario: | Industrial processes employ different configurations of electromechanical systems (ES) to perform specific tasks, indeed, the operativity and continuity of processes are subjected to machine availability. In this regard, the proposal of novel condition monitoring strategies remains mandatory in order to guarantee the continuous operation of machines. Thus, this work lies in the proposal of an intelligent fault diagnosis method capable of detecting the occurrence of faults in Induction Motors (IM) and Gearboxes (GB), the method considers the acquisition of thermal images from an ES where both, IM and GB, are linked. The novelty and contribution of this work consider a thermography-based feature space that consists of processing thermal images through the estimation of thermal matrices and their corresponding calculation of statistical features and geometric features as fault-related patterns. Moreover, the novelty includes the modelling and learning of fault-related features through a two-stage unsupervised modelling-learning approach supported by Self-Organizing Maps (SOM) and a deep Stacked Auto Encoder (SAE) structure. Finally, the automatic detection and identification of faults is carried out by a single SoftMax layer, the achieved results depict the robustness of the proposed method since the training and validation under a 5-fold cross-validation scheme lead to average classification rates higher than 99%. The result proves the capability of the method for detecting multiple faults in IMs and different severities of wear in GBs. |
|---|