Manufacturing companies increasingly become “smarter” as a result of the Industry 4.0 revolution. Multiple sensors are used for industrial monitoring of machines and workers in order to detect events and consequently improve the manufacturing processes, lower the respective costs, and increase safety. Multisensor systems produce big amounts of heterogeneous data. Data fusion techniques address the issue of multimodality by combining data from different sources and improving the results of monitoring systems. The current paper presents a detailed review of state-of-the-art data fusion solutions, on data storage and indexing from various types of sensors, feature engineering, and multimodal data integration. The review aims to serve as a guide for the early stages of an analytic pipeline of manufacturing prognosis. The reviewed literature showed that in fusion and in preprocessing, the methods chosen to be applied in this sector are beyond the state-of-the-art. Existing weaknesses and gaps that lead to future research goals were also identified.

To increase the competitiveness of wind energy, the maintenance costs of offshore floating and fixed wind turbines need to be reduced. One strategy is the enhancement of the condition monitoring techniques for pitch bearings, because their low operational speed and the high loads applied to them make their monitoring challenging. Vibration analysis has been widely used for monitoring the bearing condition with good results obtained for regular bearings, but with difficulties when the operational speed decreases. Therefore, new techniques are required to enhance the capabilities of vibration analysis for bearings under such operational conditions. This study proposes the use of indicators based on entropy for monitoring a low-speed bearing condition. The indicators used are approximate, dispersion, singular value decomposition, and spectral entropy of the permutation entropy. This approach has been tested with vibration signals acquired in a test rig with bearings under different health conditions. The results show that entropy indicators (EIs) can discriminate with higher-accuracy damaged bearings for low-speed bearings compared with the regular indicators. Furthermore, it is shown that the combination of regular and entropy-based indicators can also contribute to a more reliable diagnosis.

Virtual sensing is the process of using available data from real sensors in combination with a model of the system to obtain estimated data from unmeasured points. In this article, different strain virtual sensing algorithms are tested using real sensor data, under unmeasured different forces applied in different directions. Stochastic algorithms (Kalman filter and augmented Kalman filter) and deterministic algorithms (least-squares strain estimation) are tested with different input sensor configurations. A wind turbine prototype is used to apply the virtual sensing algorithms and evaluate the obtained estimations. An inertial shaker is installed on the top of the prototype, with a rotational base, to generate different external forces in different directions. The results obtained in the performed tests are analyzed to determine the most efficient sensor configurations capable of obtaining accurate estimates. Results show that it is possible to obtain accurate strain estimations at unmeasured points of a structure under an unknown loading condition, using measured strain data from a set of points and a sufficiently accurate FE model as input and applying the augmented Kalman filter or the least-squares strain estimation in combination with modal truncation and expansion techniques.

Despite its influence on wind energy service life, condition-based maintenance is still challenging to perform. For offshore wind farms, which are placed in harsh and remote environments, damage detection is critically important to schedule maintenance tasks and reduce operation and maintenance costs. One critical component to be monitored on a wind turbine is the pitch bearing, which can operate at low speed and high loads. Classical methods and features for general purpose bearings cannot be applied effectively to wind turbine pitch bearings owing to their specific operating conditions (high loads and non-constant very low speed with changing direction). Thus, damage detection of wind turbine pitch bearings is currently a challenge. In this study, entropy indicators are proposed as an alternative to classical bearing analysis. For this purpose, spectral and permutation entropy are combined to analyze a raw vibration signal from a low-speed bearing in several scenarios. The results indicate that entropy values change according to different types of damage on bearings, and the sensitivity of the entropy types differs among them. The study offers some important insights into the use of entropy indicators for feature extraction and it lays the foundation for future bearing prognosis methods.

Diagnosis and prognosis are key issues in the application of condition based maintenance. Thus, there is a need to evaluate the condition of a machine. Physics-based models are of great interest as they give the response of a modelled system in different operating conditions. This strategy allows for the generation of synthetic data that can be used in combination with real data acquired by sensors to improve maintenance. The article presents an electromechanical model for a rotating machine, with special emphasis on the modelling of rolling element bearings. The proposed model is validated by comparing the simulation results and the experimental results in different operating conditions and different damaged states. This comparison shows good agreement, obtaining differences of up to 10% for the modelling of the whole rotating machine and less than 0.6% for the model of the bearing.