Automated modal parameter-based anomaly detection under varying wind excitation

Efficient wind energy harvesting becomes more and more important as a consequence of the increasing interest in renewable energy in the European Union [1]. This leads to growing sizes of wind turbines (WTs), and with it, larger WT blades (WTBs). The structural designs of these WTBs are created to optimize the potential energy output, where low mass is a key requirement. However, high flexibilities and lower buckling capacities are further results of these developments [2], thus certain damage scenarios become significant. Intelligently designed structural health monitoring (SHM) systems can help to reduce the associated operations and maintenance costs. Even though, several techniques are already developed for structural damage detection (SDD) in WTBs, the majority of these methods is not suitable for inservice measurements. This paper presents a SDD and structural damage localization (SDL) method based on the partial autocorrelation function (PACF) of vibration responses. The approach is applied to a numerical model of a large WTB, where the acceleration responses are obtained from transient dynamic simulations with a simplified aerodynamic loading approach. The novel damage sensitive feature (DSF) is developed as the Mahalanobis distance between a baseline and current vector of PACF coefficients. First, numerical modal analysis of the finite element (FE) WTB model is performed in order to estimate the effect of a disbonding damage scenario on the vibration characteristics. Second, the behaviour of the PACF for time series of the healthy system is discussed. Third, the SDD results on the basis of statistical hypothesis testing are assessed for two selected sensor locations and increasing damage extents. Finally, the performance of the proposed DSF with respect to SDL is illustrated for multiple locations on the WTB’s surface. This study demonstrated the efficiency of a DSF based on the PACF for SDD and SDL, which is promising for future developments of vibration-based SHM techniques in WTBs.

The development of large wind turbines that enable to harvest energy more efficiently is a consequence of the increasing demand for renewables in the world. To optimize the potential energy output, light and flexible wind turbine blades (WTBs) are designed. However, the higher flexibilities and lower buckling capacities adversely affect the long-term safety and reliability of WTBs, and thus the increased operation and maintenance costs reduce the expected revenue. Effective structural health monitoring techniques can help to counteract this by limiting inspection efforts and avoiding unplanned maintenance actions. Vibration-based methods deserve high attention due to the moderate instrumentation efforts and the applicability for in-service measurements. The present paper proposes the use of cross-correlations (CCs) of acceleration responses between sensors at different locations for structural damage detection in WTBs. CCs were in the past successfully applied for damage detection in numerical and experimental beam structures while utilizing only single lags between the signals. The present approach uses vectors of CC coefficients for multiple lags between measurements of two selected sensors taken from multiple possible combinations of sensors. To reduce the dimensionality of the damage sensitive feature (DSF) vectors, principal component analysis is performed. The optimal number of principal components (PCs) is chosen with respect to a statistical threshold. Finally, the detection phase uses the selected PCs of the healthy structure to calculate scores from a current DSF vector, where statistical hypothesis testing is performed for making a decision about the current structural state. The method is applied to laboratory experiments conducted on a small WTB with non-destructive damage scenarios.

Modal analysis has developed into a major technology for the study of structural dynamics in the past several decades. Through it, complex structural dynamics phenomena can be represented in terms of structural invariants, i.e., the modal parameters: natural frequencies, damping ratios and mode shapes. Operational Modal Analysis (OMA) deals with the estimation of modal parameters on vibration data measured for operational conditions, when the excitation on the structure is not measured. In this work, OMA is performed on a wind turbine blade undergoing wind tunnel testing. These tests included different wind speeds, pitch angles and also different health conditions of the blade, where masses of different magnitude were fixed to the blade on different locations to emulate damage conditions. In order to monitor the modal parameters across multiple days of varied tests in the wind tunnel, the Polymax modal parameter estimator was implemented, coupled with an Automated Modal Analysis methodology. This methodology included an automatic modal parameter selection technique, using a Machine Learning (ML) clustering algorithm, coupled with a modal tracking procedure which applied statistical thresholds on the modal parameters’ values. The tracking procedure searches for modes similar to the ones calculated for healthy conditions. The results show how the modal parameters of the wind turbine blade vary with the different measured conditions in the wind tunnel. Moreover, a damage detection methodology is implemented to differentiate between the healthy and damaged conditions on the blade, by leveraging an anomaly detection algorithm using the Multivariate Gaussian Distribution (MGD). This algorithm takes as input the modal parameters calculated by the previous Automated Modal Analysis methodology and detects statistical deviations among them which could indicate the presence of damage. All steps of this work contribute to developing an automatic framework able to detect damages on a wind turbine blade, and therefore perform Structural Health Monitoring (SHM) for different operational conditions.

The application of vibration-based Structural Health Monitoring (SHM) for damage detection is characterised by three fundamental aspects: the features extracted as representative of the structural condition that can be directly linked to some form of damage, the metrics selected as novelty or damage index, and the statistical model or classifier built to identify underlying patterns indicative of changes in the structure’s state. Focusing on the first step to improve the performance of vibration-based SHM strategies, the extracted features should be robust to noise, sensitive to the presence of a specific type of damage. Further, damage should induce patterns that are distinguishable from environmental and operational variabilities and other forms of damage such as ageing phenomena. In this paper, the problem is framed as an outlier detection problem and the the use of different modal parameters as Damage Sensitive Features (DSFs) is investigated, evaluating them based on the detection performance of an unsupervised One-Class Support Vector Machine (OCSVM) classifier. In particular, an artificial dataset is generated from the calibrated numerical model of a three-storey steel frame structure in different damage scenarios. The methodology is validated against available experimental data. For the case investigated the natural frequencies were sensitive to damage and robust to noise.

Fukunaga-Koontz feature transformation for statistical structural damage detection and hierarchical neuro-fuzzy damage localisation

An artificial neural network methodology for damage detection: Demonstration on an operating wind turbine blade

In this paper a vibration-based approach to identify a crack in a wind turbine blade is described and demonstrated numerically and experimentally. Operational modal analysis has been performed before and after a buckling test, and vibration data, gathered from some accelerometers placed along the blade, was used to monitor the integrity of the structure, since the modal parameters are directly influenced by the physical properties of the structures. Additionally a numerical prediction has been done both with a full-scale model and with a one-dimensional model. The results show that this approach is able to estimate successfully the presence of damage and a good numerical and experimental correlation has been found. Finally, some considerations regarding the rotation of the blade in the undamaged and damaged cases have been done.

Application of massive data to structural health monitoring (SHM) may lead to serious problems such as difficulty, computational inefficiency, and low damage detectability. This paper proposes innovative hybrid methods in order to detect damage under massive data, ambient vibration, and environmental and/or operational variability. Each of the proposed methods consists of a three-stage algorithm including response modelling via a time series representation, Gaussian mixture model (GMM) for dimensionality reduction, and an outlier detector. Due to the importance of response modelling under ambient vibration, this article employs the combination of AR with ARX called the ARARX model. In the second stage, a GMM is individually fitted to residual samples extracted from ARARX regarding the undamaged and damaged conditions to obtain low-dimensional features from massive data. Eventually, outlier analyses are carried out by the Mahalanobis distance and an auto-associative neural network to make a decision about the occurrence of damage. Two different kinds of threshold limits are also considered to examine the performances of the proposed methods. A large-scale cable-stayed bridge is applied to validate the reliability and effectiveness of these methods with several comparative studies. Results demonstrate that both proposed methods are effective tools for damage detection.

Vibration-based novelty detection using autoencoders: application to KW51 bridge

The high structural deficient rate poses serious risks to the operation of many bridges and buildings. To prevent critical damage and structural collapse, a quick structural health diagnosis tool is needed during normal operation or immediately after extreme events. In structural health monitoring (SHM), many existing methods will have limited usefulness in the quick damage identification process because (1) the damage event needs to be identified quickly, and (2) postdamage information is usually unavailable. To address these drawbacks, we propose a new damage detection and localization approach based on stochastic time series analysis. Specifically, damage sensitive features, which are extracted from vibration signals, follow different distributions before and after a damage event. Hence, we use optimal change-point detection theory to find the time of damage occurrence. Because existing change-point detectors require the postdamage feature distribution, which is unavailable in SHM, we propose a maximum likelihood method for learning the distribution parameters from the time-series data. The proposed damage detection using estimated parameters achieves optimal performance. Also, we utilize the detection results to find damage location without any further computation. Validation results show highly accurate damage identification in American Society of Civil Engineers benchmark structures and two shake table experiments.

Structural Health Monitoring (SHM) is a condition-based maintenance technology using sensing systems. In SHM, the use of domain knowledge is essential: it motivates the use of machine learning approaches; it can be used to extract damage sensitive features and interpret the results by machine learning. This work focuses on two SHM problems: damage identification and substructure clustering. Our solutions to address them are based on machine learning techniques and robust feature extraction using domain knowledge. In the first problem, damage sensitive features were extracted using a frequency domain decomposition, followed by a robust one-class support vector machine for damage detection. In the second problem, a novel clustering technique and spectral moment feature were utilised for substructure grouping and anomaly detection. These methods were evaluated using data from lab-based structures and data collected from the Sydney Harbour Bridge. We obtained high damage detection accuracies and were able to assess damage severity. Furthermore, the clustering technique was able to group substructures of similar behaviour and detect spatial anomalies.

Increasing levels of modernization in different countries worldwide have lead to an exponential growth in national energy demand. India with its huge population base had a total grid connected electric power close to 285 GW during 2015 with annual generation of 1106 TWh accounting to a per-capita energy consumption of 746 kWh. The Government of India under its ambitious plan “Power for all” envisages 24 × 7 power supply for the entire domestic and Industrial loads by the year 2022. This power sector initiative also promises the generation through green path by targeting reduction in carbon footprint. The plan has sanctioned funds for production of 190 GW power based on go green initiative of United Nations Organization and a step forward to create a better environment for future generations. The proposed work aims at design of wind-solar hybrid system for light load decentralized applications. The computational tools were used to model and analyze wind turbine (WT) blade structure for low power applications to suit decentralized power generation based on renewable energy. The comparative studies of test aerofoil with standard NACA 0018 aerofoil indicated conformance of the test aerofoil as a profile for WT blades. The test aerofoil has the feature of easier manufacturability as against standard NACA 0018 profile adopted in commercial WT blades. The studies conformed suitability of test profile at specified angle of attack with WT blade structure and hub construction using balsa wood. The choice of balsa wood was based on properties of lighter weight and adequate strength that was supported on the designed tower of tripod type. The designed WT system was for a rated power output of 75 W while operating at an average wind speed of 4.5 m/s. The system developed will be a feasible option for powering households in rural India.

Vibration-based damage detection approaches have received considerable attention in the field of structural health monitoring. Modal parameters are often adopted to define damage-sensitive features, since their strong physical meaning can help interpreting the structural condition. On the other hand, they are also sensitive to changes to environmental and operational conditions. This aspect is critical for an automatic damage detection, because changes in modal parameters due to varying external factors (e.g. temperature) can be greater than those caused by damage.In this context, this paper proposes an application to real data coming from long-term structural health monitoring of axially-loaded beam-like structures under realistic environment. These very common structural elements are usually subject to an axial load that changes under environmental and operating conditions. Since the axial load is not directly known in most real applications, assessing damage using modal-based damage features is a complicated task.In this paper, a data driven approach that does not require a knowledge of the axial load is proposed to filter out the environmental effects on modal-based damage features. The strategy has been successfully tested on data acquired in an uncontrolled environment, and resulted in being a promising solution for real structural health monitoring applications.KeywordsLong-term monitoringDamage detectionEnvironmental and operational variationsVibration-based featureStatistical pattern recognitionBeam-like structureTie-rod

Rotating machines are key components in energy generation processes, and faults can lead to shutdowns or catastrophes encompassing economic and social losses. Structural Health Monitoring (SHM) of structures in operation is successfully performed via Operational Modal Analysis (OMA), which has advantages over traditional methods. In OMA, white noise inputs lead to the accurate extraction of modal parameters without taking the system out of operation. However, this excitation condition is not easy to attain for rotating machines used in power generation, and OMA can provide inaccurate information. This research investigates the applicability of machine learning as a pre-stage of OMA to differentiate adequate from inadequate excitations and prevent inaccurate extraction of modal parameters. Data from a rotor system was collected under different conditions and OMA was applied. In a training stage, measurements were characterized by statistical features and K-means was used to determine which features provided information about the excitation condition, that is, which excitation was adequate to extract the rotor’s modal parameters via OMA. In a testing stage, data were successfully classified as adequate or not adequate for OMA, achieving 100% accuracy and revealing the technique’s potential to support SHM of rotating machines. The technique is extendable to other monitoring systems based on OMA.

The torque generated by a wind turbine blade is dependent on several parameters, one of which is the angle of attack. Several models for predicting the angle of attack in yawed conditions have been proposed in the literature, but there is a lack of experimental data to use for direct validation.To address this problem, experiments were conducted at the University of Waterloo Wind Generation Research Facility using a 3.4 m diameter test turbine. A five-hole pressure probe was installed in a modular 3D printed blade and was used to measure the angle of attack, a, as a function of several parameters. Measurements were conducted at radial positions of r/R = 0.55 and 0.72 at tip speed ratios of λ = 5.0, 3.6, and 3.1. The yaw offset of the turbine was varied from -15° to +15°.Experimental results were compared directly to angle of attack values calculated using a model proposed by Morote in 2015. Modeled values were found to be in close agreement with the experimental results. The angle of attack was shown to vary cyclically in the yawed case while remaining mostly constant when aligned with the flow, as expected. The quality of results indicates the potential of the developed instrument for wind turbine measurements.