Predictive maintenance strategies are widely used in industry to reduce maintenance costs and avoid unexpected downtime that can affect production. Predictive maintenance strategies are usually based on monitoring the product in use, where, based on historical operating data and analysis of the evolution of acquired signals, the failure of critical components is predicted.

However, the development of unique custom products, increasingly demanded in the current market, implies the absence of historical data that allow such maintenance tasks to be carried out until such data is generated. Therefore, carrying out predictive maintenance in this type of installation is a major challenge today.

In this regard, various authors in the literature point to the enormous potential of hybrid virtual twins fed by both simulation data and empirical data. In this way, virtual data obtained through simulation tools play a fundamental role in predicting the product's service behavior. Likewise, as usage data becomes available, these models feed the hybrid virtual twins, improving their level of accuracy and allowing more efficient maintenance tasks to be performed.

To develop hybrid virtual twins of electromechanical products, it is necessary to consider and integrate different layers:

• The control layer establishes the product's operating operations, speeds, accelerations, trajectories, etc. responsible for the efforts generated in the different components of the product.
• The structural layer is responsible for modeling the transfer of efforts to the different components, considering the resistance and stiffness of the product's mechanical structure.
• The behavioral layers are those that relate the life of the components to the efforts transmitted through the structure.

Thus, by integrating the different layers, hybrid virtual twins allow the life of critical components to be predicted under different product operation scenarios.