Programa del congreso

Unplanned diesel-engine failures lead to relevant operational and economic impacts in heavy-transport settings. Predictive maintenance approaches based on remaining useful life (RUL) estimation can mitigate downtime; however, real labeled degradation datasets are often scarce. This work presents a simulation-based, reproducible framework to generate multivariate synthetic OBD-II–like signals and evaluate RUL prediction using a Long Short-Term Memory (LSTM) network. In addition, a Genetic Algorithm (GA) is applied to search LSTM hyperparameters using validation performance as the fitness criterion. The system is assessed as a regression problem using 𝑹𝟐, RMSE, MAE, and relative RMSE under an asset-based split, where an unseen engine is reserved for testing. Results show that both the baseline LSTM and the GA-tuned LSTM satisfy the predefined acceptance criteria (𝑹𝟐 ≥ 𝟎. 𝟖𝟓 and relative RMSE ≤ 𝟏𝟓%), while the baseline model achieves a lower test RMSE than the GA-tuned configuration, highlighting that improved validation fitness does not necessarily translate into better generalization on unseen assets. The proposed framework provides a controlled proof of concept for future studies incorporating real OBDII data and broader validation.