Prognosis of a Wind Turbine Gearbox Bearing Using Supervised Machine Learning

Predicting System Degradation with a Guided Neural Network Approach

Evaluating the physical degradation behavior and estimating the lifetime of engineering systems and structures is crucial to ensure their safe and reliable operation. However, measuring lifetime through actual operating conditions can be a difficult and slow process. While valuable and quick in measuring lifetimes, accelerated life testing is often oversimplified and does not provide accurate simulations of the exact operating environment. This paper proposes a data-driven framework for time-efficient modeling of field degradation using sensor measurements from short-term actual operating conditions degradation tests. The framework consists of two neural networks: a physics discovery neural network and a predictive neural network. The former models the underlying physics of degradation, while the latter makes probabilistic predictions for degradation intensity. The physics discovery neural network guides the predictive neural network for better life estimations. The proposed framework addresses two main challenges associated with applying neural networks for lifetime estimation: incorporating the underlying physics of degradation and requirements for extensive training data. This paper demonstrates the effectiveness of the proposed approach through a case study of atmospheric corrosion of steel test samples in a marine environment. The results show the proposed framework's effectiveness, where the mean absolute error of the predictions is lower by up to 76% compared to a standard neural network. By employing the proposed data-driven framework for lifetime prediction, systems safety and reliability can be evaluated efficiently, and maintenance activities can be optimized.

Machine Learning-Based Analysis of a Wind Turbine Manufacturing Operation: A Case Study

This study analyzes the lead time of the bending operation in the wind turbine tower manufacturing process. Since the operation involves a significant amount of employee interaction and the parts processed are heavy and voluminous, there is considerable variability in the recorded lead times. Therefore, a machine learning regression analysis has been applied to the bending process. Two machine learning algorithms have been used: a multivariate Linear Regression and the M5P method. The goal of the analysis is to gain a better understanding of the effect of several factors (technical, organizational, and experience-related) on the bending process times, and to attempt to predict these operation times as a way to increase the planning and controlling capacity of the plant. The inclusion of the experience-related variables serves as a basis for analyzing the impact of age and experience on the time-wise efficiency of workers. The proposed approach has been applied to the case of a Spanish wind turbine tower manufacturer, using data from the operation of its plant gathered between 2018 and 2021. The results show that the trained models have a moderate predictive power. Additionally, as shown by the output of the regression analysis, there are variables that would presumably have a significant impact on lead times that have been found to be non-factors, as well as some variables that generate an unexpected degree of variability.

Early Fault Diagnosis Strategy for WT Main Bearings Based on SCADA Data and One-Class SVM

To reduce the levelized cost of wind energy, through the reduction in operation and maintenance costs, it is imperative that the wind turbine downtime is reduced through maintenance strategies based on condition monitoring. The standard approach toward this challenge is based on vibration monitoring, which requires the installation of specific tailored sensors that incur associated added costs. On the other hand, the life expectancy of wind parks built during the 1990s wind power boom is dwindling, and data-driven maintenance strategies issued from already accessible supervisory control and data acquisition (SCADA) data is an auspicious competitive solution because no additional sensors are required. Note that it is a major issue to provide fault diagnosis approaches built only on SCADA data, as these data were not established with the objective of being used for condition monitoring but rather for control capacities. The present study posits an early fault diagnosis strategy based exclusively on SCADA data and supports it with results on a real wind park with 18 wind turbines. The contributed methodology is an anomaly detection model based on a one-class support vector machine classifier; that is, it is a semi-supervised approach that trains a decision function that categorizes fresh data as similar or dissimilar to the training set. Therefore, only healthy (normal operation) data is required to train the model, which greatly expands the possibility of employing this methodology (because there is no need for faulty data from the past, and only normal operation SCADA data is needed). The results obtained from the real wind park show that this is a promising strategy.

Optimum Selection of Coated Piston Rings and Thrust Bearings in Mixed Lubrication for Different Lubricants Using Machine Learning

The purpose of this study is to build a parametric algorithm combining analytical results and Machine Learning in order to improve the tribological performance of coated piston rings and thrust bearings in mixed lubrication using different synthetic lubricants. The friction models for piston ring conjunction and pivoted pad thrust bearing consider the basic lubrication theory, the detailed contact geometry and the complete lubricant action for a wide range of speeds. The data produced from the analytical solutions are used as input for the training of regression models. The effect of TiN, TiAlN, CrN and DLC coatings on friction coefficient are investigated through multi-variable quadratic regression and support vector machine models. The optimum selection is considered when the minimum friction coefficient is predicted. Smooth TiN2 and TiAlN coatings seem to affect better the ring friction coefficient than rougher steel, TiN1 and CrN coatings using an uncoated or coated Nickel Nanocomposite (NNC) cylinder. Using an NNC cylinder for better durability, the friction coefficients were found to be higher by 31.3−58.8% for all the studied rings due to the rougher surface morphology. On the other hand, the results indicate that pads coated with DLC show lower friction coefficients compared to the common steel and TiAlN, CrN, and TiN applications. The multi-variable second-order polynomial regression models were demonstrated to be 1−6% more accurate than the quadratic support vector machine models in both tribological contacts.

Deep-Learning Based Prognosis Approach for Remaining Useful Life Prediction of Turbofan Engine

The entire life cycle of a turbofan engine is a type of asymmetrical process in which each engine part has different characteristics. Extracting and modeling the engine symmetry characteristics is significant in improving remaining useful life (RUL) predictions for aircraft components, and it is critical for an effective and reliable maintenance strategy. Such predictions can improve the maximum operating availability and reduce maintenance costs. Due to the high nonlinearity and complexity of mechanical systems, conventional methods are unable to satisfy the needs of medium- and long-term prediction problems and frequently overlook the effect of temporal information on prediction performance. To address this issue, this study presents a new attention-based deep convolutional neural network (DCNN) architecture to predict the RUL of turbofan engines. The prognosability metric was used for feature ranking and selection, whereas a time window method was employed for sample preparation to take advantage of multivariate temporal information for better feature extraction by means of an attention-based DCNN model. The validation of the proposed model was conducted using a well-known benchmark dataset and evaluation measures such as root mean square error (RMSE) and asymmetric scoring function (score) were used to validate the proposed approach. The experimental results show the superiority of the proposed approach to predict the RUL of a turbofan engine. The attention-based DCNN model achieved the best scores on the FD001 independent testing dataset, with an RMSE of 11.81 and a score of 223.

Remaining Useful Life Estimation Using Deep Convolutional Generative Adversarial Networks Based on an Autoencoder Scheme

Accurate predictions of remaining useful life (RUL) of important components play a crucial role in system reliability, which is the basis of prognostics and health management (PHM). This paper proposed an integrated deep learning approach for RUL prediction of a turbofan engine by integrating an autoencoder (AE) with a deep convolutional generative adversarial network (DCGAN). In the pretraining stage, the reconstructed data of the AE not only participate in its error reconstruction but also take part in the DCGAN parameter training as the generated data of the DCGAN. Through double-error reconstructions, the capability of feature extraction is enhanced, and high-level abstract information is obtained. In the fine-tuning stage, a long short-term memory (LSTM) network is used to extract the sequential information from the features to predict the RUL. The effectiveness of the proposed scheme is verified on the NASA commercial modular aero-propulsion system simulation (C-MAPSS) dataset. The superiority of the proposed method is demonstrated via excellent prediction performance and comparisons with other existing state-of-the-art prognostics. The results of this study suggest that the proposed data-driven prognostic method offers a new and promising prediction approach and an efficient feature extraction scheme.

Misalignment Fault Prediction of Wind Turbines Based on Improved Artificial Fish Swarm Algorithm

A misalignment fault is a kind of potential fault in double-fed wind turbines. The reasonable and effective fault prediction models are used to predict its development trend before serious faults occur, which can take measures to repair in advance and reduce human and material losses. In this paper, the Least Squares Support Vector Machine optimized by the Improved Artificial Fish Swarm Algorithm is used to predict the misalignment index of the experiment platform. The mixed features of time domain, frequency domain, and time-frequency domain indexes of vibration or stator current signals are the inputs of the Least Squares Support Vector Machine. The kurtosis of the same signals is the output of the model, and the 3σ principle of the normal distribution is adopted to set the warning line of misalignment fault. Compared with other optimization algorithms, the experimental results show that the proposed prediction model can predict the development trend of the misalignment index with the least prediction error.

Leveraging Label Information in a Knowledge-Driven Approach for Rolling-Element Bearings Remaining Useful Life Prediction

Since bearing deterioration patterns are difficult to collect from real, long lifetime scenarios, data-driven research has been directed towards recovering them by imposing accelerated life tests. Consequently, insufficiently recovered features due to rapid damage propagation seem more likely to lead to poorly generalized learning machines. Knowledge-driven learning comes as a solution by providing prior assumptions from transfer learning. Likewise, the absence of true labels was able to create inconsistency related problems between samples, and teacher-given label behaviors led to more ill-posed predictors. Therefore, in an attempt to overcome the incomplete, unlabeled data drawbacks, a new autoencoder has been designed as an additional source that could correlate inputs and labels by exploiting label information in a completely unsupervised learning scheme. Additionally, its stacked denoising version seems to more robustly be able to recover them for new unseen data. Due to the non-stationary and sequentially driven nature of samples, recovered representations have been fed into a transfer learning, convolutional, long–short-term memory neural network for further meaningful learning representations. The assessment procedures were benchmarked against recent methods under different training datasets. The obtained results led to more efficiency confirming the strength of the new learning path.

Estimating Residual Life Distributions of Complex Operational Systems Using a Remaining Maintenance Free Operating Period (RMFOP)-Based Methodology

Recent developments in the area of condition monitoring research have been targeted towards predicting machinery health condition for the purpose of preventative maintenance. Typically, published research uses data collected from rotating components (bearings, cutting tools, etc.) working in an idealized lab environment as the case study for prognosis algorithm validations. However, the operational implementation in industry is still very sporadic, mainly owing to the lack of proper data allowing sufficiently mature development of comprehensive methodologies. The prognosis methodology presented herein bridges the gap between academic research and industrial implementations by employing a novel time period for prognosis and implementing random coefficients regression models. The definition of the remaining maintenance-free operating period (RMFOP) is proposed first, which helps to transform the usefulness of the degradation data that is readily available from data short of failure. Degradation patterns are subsequently extracted from the original degradation data, before fitting into either of two regression models (linear or exponential). The system residual life distributions are then computed and updated by estimating the parameter statistics within the model. This RMFOP-based methodology is validated using real-world degradation data collected from multiple operational railway switch systems across Great Britain. The results indicate that both the linear model and the exponential model can produce residual life distributions with a sufficient prediction accuracy for this specific application. The exponential model gives better predictions, the accuracy of which also improves as more of system life percentage has elapsed. By using the RMFOP methodology, switch system health condition affected by an incipient overdriving fault is recognized and predicted.

Winding Tensor Approach for the Analytical Computation of the Inductance Matrix in Eccentric Induction Machines

Induction machines (IMs) are critical components of many industrial processes, what justifies the use of condition-based maintenance (CBM) systems for detecting their faults at an early stage, in order to avoid costly breakdowns of production lines. The development of CBM systems for IMs relies on the use of fast models that can accurately simulate the machine in faulty conditions. In particular, IM models must be able to reproduce the characteristic harmonics that the IM faults impress in the spatial waves of the air gap magneto-motive force (MMF), due to the complex interactions between spatial and time harmonics. A common type of fault is the eccentricity of the rotor core, which provokes an unbalanced magnetic pull, and can lead to destructive rotor-stator rub. Models developed using the finite element method (FEM) can achieve the required accuracy, but their high computational costs hinder their use in online CBM systems. Analytical models are much faster, but they need an inductance matrix that takes into account the asymmetries generated by the eccentricity fault. Building the inductance matrix for eccentric IMs using traditional techniques, such as the winding function approach (WFA), is a highly complex task, because these functions depend on the combined effect of the winding layout and of the air gap asymmetry. In this paper, a novel method for the fast and simple computation of the inductance matrix for eccentric IMs is presented, which decouples the influence of the air gap asymmetry and of the winding configuration using two independent tensors. It is based on the construction of a primitive inductance tensor, which formulates the eccentricity fault using single conductors as the simplest reference frame; and a winding tensor that converts it into the inductance matrix of a particular machine, taking into account the configuration of the windings. The proposed approach applies routine procedures from tensor algebra for performing such transformation in a simple way. It is theoretically explained and experimentally validated with a commercial induction motor with a mixed eccentricity fault.

Predicting Frequency, Time-To-Repair and Costs of Wind Turbine Failures

Operation and maintenance (O&M) costs, and associated uncertainty, for wind turbines (WTs) is a significant burden for wind farm operators. Many wind turbine failures are unpredictable while causing loss of energy production, and may also cause loss of asset. This study utilized 753 O&M event data from 21 wind turbines operating in Germany, to improve the prediction of failure frequency and associated costs. We applied Bayesian updating to predict wind turbine failure frequency and time-to-repair (TTR), in conjunction to machine learning techniques for assessing costs associated with failures. We found that time-to-failure (TTF), time-to-repair and the cost of failures depend on operational and environmental conditions. High elevation (>100 m) of the wind turbine installation was found to increase both the probability of failures and probability of delayed repairs. Furthermore, it was determined that direct-drive turbines are more favorable at locations with high capacity factor (more than 40%) whereas geared-drive turbines show lower failure costs than direct-drive ones at temperate-coastal locations with medium capacity factors (between 20% and 40%). Based on these findings, we developed a decision support tool that can guide a site-specific selection of wind turbine types, while providing a thorough estimation of O&M budgets.

An Improved LightGBM Algorithm for Online Fault Detection of Wind Turbine Gearboxes

It is widely accepted that conventional boost algorithms are of low efficiency and accuracy in dealing with big data collected from wind turbine operations. To address this issue, this paper is devoted to the application of an adaptive LightGBM method for wind turbine fault detections. To this end, the realization of feature selection for fault detection is firstly achieved by utilizing the maximum information coefficient to analyze the correlation among features in supervisory control and data acquisition (SCADA) of wind turbines. After that, a performance evaluation criterion is proposed for the improved LightGBM model to support fault detections. In this scheme, by embedding the confusion matrix as a performance indicator, an improved LightGBM fault detection approach is then developed. Based on the adaptive LightGBM fault detection model, a fault detection strategy for wind turbine gearboxes is investigated. To demonstrate the applications of the proposed algorithms and methods, a case study with a three-year SCADA dataset obtained from a wind farm sited in Southern China is conducted. Results indicate that the proposed approaches established a fault detection framework of wind turbine systems with either lower false alarm rate or lower missing detection rate.

Wind Turbine Noise Prediction Using Random Forest Regression

Wind energy is one of the most widely used renewable energy sources in the world and has grown rapidly in recent years. However, the wind towers generate a noise that is perceived as an annoyance by the population living near the wind farms. It is therefore important to new tools that can help wind farm builders and the administrations. In this study, the measurements of the noise emitted by a wind farm and the data recorded by the supervisory control and data acquisition (SCADA) system were used to construct a prediction model. First, acoustic measurements and control system data have been analyzed to characterize the phenomenon. An appropriate number of observations were then extracted, and these data were pre-processed. Subsequently two models of prediction of sound pressure levels were built at the receiver: a model based on multiple linear regression, and a model based on Random Forest algorithm. As predictors wind speeds measured near the wind turbines and the active power of the turbines were selected. Both data were measured by the SCADA system of wind turbines. The model based on the Random Forest algorithm showed high values of the Pearson correlation coefficient (0.981), indicating a high number of correct predictions. This model can be extremely useful, both for the receiver and for the wind farm manager. Through the results of the model it will be possible to establish for which wind speed values the noise produced by wind turbines become dominant. Furthermore, the predictive model can give an overview of the noise produced by the receiver from the system in different operating conditions. Finally, the prediction model does not require the shutdown of the plant, a very expensive procedure due to the consequent loss of production.

Model-Based Fault Detection and Identification for Prognostics of Electromechanical Actuators Using Genetic Algorithms

Traditional hydraulic servomechanisms for aircraft control surfaces are being gradually replaced by newer technologies, such as Electro-Mechanical Actuators (EMAs). Since field data about reliability of EMAs are not available due to their recent adoption, their failure modes are not fully understood yet; therefore, an effective prognostic tool could help detect incipient failures of the flight control system, in order to properly schedule maintenance interventions and replacement of the actuators. A twofold benefit would be achieved: Safety would be improved by avoiding the aircraft to fly with damaged components, and replacement of still functional components would be prevented, reducing maintenance costs. However, EMA prognostic presents a challenge due to the complexity and to the multi-disciplinary nature of the monitored systems. We propose a model-based fault detection and isolation (FDI) method, employing a Genetic Algorithm (GA) to identify failure precursors before the performance of the system starts being compromised. Four different failure modes are considered: dry friction, backlash, partial coil short circuit, and controller gain drift. The method presented in this work is able to deal with the challenge leveraging the system design knowledge in a more effective way than data-driven strategies, and requires less experimental data. To test the proposed tool, a simulated test rig was developed. Two numerical models of the EMA were implemented with different level of detail: A high fidelity model provided the data of the faulty actuator to be analyzed, while a simpler one, computationally lighter but accurate enough to simulate the considered fault modes, was executed iteratively by the GA. The results showed good robustness and precision, allowing the early identification of a system malfunctioning with few false positives or missed failures.