Abstract
Transmissions are extensively employed in mechanical gearboxes when power conversion is required. Being able to provide specific maintenance is a crucial factor for both economics and reliability. However, although periodic transmission maintenance increases the systems’ longevity, it cannot prevent or predict sporadic major failures. In this context, structural health monitoring (SHM) represents a possible solution. Identifying variations of a specific measurable signal and correlating them with the type of damage or its location and severity may help assess the component condition and establish the need for maintenance operation. However, the collection of sufficient experimental examples for damage identification may be not convenient for big gearboxes, for which destructive experiments are too expensive, thus paving the way to model-based approaches, based on a numerical estimation of damage-related features. In this work, an SHM approach was developed based on signals from numerical simulations. To validate the approach with experimental measurements, a back-to-back test rig was used as a reference. Several types and severities of damages were simulated with an innovative hybrid analytical–numerical approach that allowed a significant reduction of the computational effort. The vibrational spectra that characterized the different damage conditions were processed through artificial neural networks (ANN) trained with numerical data and used to predict the presence, location, and severity of the damage.