TECHNICAL PAPER Hybrid SOM–PCA method for modeling bearing faults detection and diagnosis Mohamed Lamine Fadda 1 • Abdelkrim Moussaoui 1 Received: 6 April 2016 / Accepted: 16 April 2018 Ó The Brazilian Society of Mechanical Sciences and Engineering 2018 Abstract Rolling bearing faults always stay a thorny problem in the renewable energy field; therefore, the necessity of research and development in this area is essential. In this document, we propose a new approach in the faults detection and diagnosis of bearings during their operation. The principal aim of this study was to ensure a fast and efficient modeling of the unknown signals and their diagnostic, hence minimizing the damage in systems and the maintenance time and costs. The modeling step is based on the frequency’s analysis of residues produced by (SOM–PCA) algorithm, and then, the diagnostic step is relied on the classification of the unknown signals in four operation cases possibly using the SOM model. Results indicate the efficiency of method to faults detection and diagnostic in experimental data such as the unsupervised classification of most faults. Keywords Bearing failure  FDI  Diagnostic  SOM  PCA  Multi-PCA  FFT 1 Introduction The development of mechanic systems become remark- able, such as renewable energy systems, since wind energy play the important role compared with all other energy sources, concerning the construction, design, and produc- tion [1]. Recently, the scientific-research activities are looking the developement of mechanical systems with the reduc- tion of all costs related [2, 3]. Where the financial impact of maintenance costs is important, and at the same time the inevitability of early maintenance procedure depends on diagnostic of possible faults, which causes fatigue, more downtime, and total or partial damage [4–6]. Current studies show that most criticals components contributor in wind turbine (WT) systems damages are the rotor, the blades, gear box, bearings [7], and the electric generator). The WT components diversity and their faults shows that the frequency of faults of component are not equal (Fig. 1)[4, 8, 9]. Additionally, the most cases of failure in WT is mechanical, the bearings faults intervenes in the principal part of failure, so that cannot cite WT problems without citing the bearings defects. Where bearings is important in rotative parts of WT. However, the true crucial choice to avoid sudden failure and ensure the integrity is recommended. The enormous attention given to scientific research in the bearing faults diagnosis field explains the diversity of approaches in the research literature [8, 10–12], and those methods are generally based on data exploration and analysis techniques, compression, data space transforma- tion representation, and neural network models. Generally, the objective of this study is to develop a fault detection technique in frequency domain to prepare the characteristic models of classification and diagnosis of faults, used in bearings experimental database vibration signals, based on the famous multivariate classification techniques, self-organizing map (SOM), principal compo- nent analysis (PCA) [13], and fast Fourier transform (FFT), Technical Editor: Fernando Antonio Forcellini. & Mohamed Lamine Fadda fadmamfad@yahoo.fr Abdelkrim Moussaoui a_k_moussaoui@yahoo.fr 1 Electrical Engineering Laboratory of Guelma (LGEG), Electrical Engineering Department, University 08 Mai 1945-Guelma, Guelma, Algeria 123 Journal of the Brazilian Society of Mechanical Sciences and Engineering (2018)40:268 https://doi.org/10.1007/s40430-018-1184-7