BHBenito Pérez Hernández

Benito Pérez Hernández

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Benito Pérez Hernández has a total of 4 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Vibration-based fault detection and classification in a jacket-type offshore wind turbine using supervised machine learning

    by José Eligio Moisés Gutiérrez Arias, Benito Pérez Hernández, Gabriela Pérez Osorio, Josefina Castañeda Camacho, Gabriela Aquino González

    Structural health monitoring of jacket-type substructures of offshore wind turbines is key to ensuring their availability under limited access and combined wind–wave loading. This work presents a functional horizontal-axis wind turbine prototype on a modular, demountable A36-steel jacket substructure inspired by the OWEC Quattropod model, conceived to induce structural damage in a controlled and repeatable manner. A synchronized wireless acquisition system with seven triaxial ADXL355 accelerometers was implemented over TCP/IP, together with a methodology to classify structural faults from vibration signals. The resulting database comprises the healthy state and three damage conditions at four rotational speeds, over 48 balanced experimental runs. After sliding-window segmentation, z -score normalization and dimensionality reduction by principal component analysis, six classifiers were compared at five window lengths using leave-one-repetition-out cross-validation, which prevents data leakage between partitions. The best configurations were the support vector machine (SVM) with a quadratic kernel ( n = 50 , 0.20 s; accuracy and macro- F 1 score of 0.993 ± 0.006 ) and linear discriminant analysis (LDA) ( n = 196 , 0.78 s; accuracy and macro- F 1 score of 0.987 ± 0.012 ), both with an area under the ROC curve and average precision of at least 0.997. These figures should be interpreted considering that the Fault 1 condition was acquired under a differentiated excitation regime; by restricting the evaluation of the original classifier predictions to the three conditions recorded under an identical protocol, the accuracy is 0.995 for the quadratic SVM and 0.982 for LDA. LDA combines stability, low computational cost and interpretability: training in less than a second and processing over 24 000 observations per second, indicating its computational viability for online structural monitoring. The results constitute a first laboratory validation of the proposed methodology. Its application to a real installation requires addressing the environmental and scale conditions of a marine site.