This paper presents an end-to-end condition monitoring co-design model, from vibration measurement to anomaly detection, where conventional signal processing principles are combined with neuromorphic sensing and computing concepts to enable investigations of the potential improvements offered by brain-like information processing technologies.

The use of machine learning in condition monitoring became increasingly popular for intelligent fault diagnosis in the last decade, taking advantage of the rapid developments in deep learning.

However, the high computational cost of training and using deep neural networks prevents the use of such solutions for analysing the bulk of data generated by the resource constrained edge devices, i.e., the condition monitoring sensor systems, as only a minor fraction of data can be transmitted to the cloud or edge servers for analysis.

There is an untapped potential to process this data and thereby improve intelligent fault diagnosis models using event-triggered sensing, spiking neural networks, and neuromorphic processors that substantially can improve the energy efficiency and capacity of embedded machine learning condition monitoring solutions.

The proposed co-design model is evaluated on two use-cases involving rolling element bearing failures, one based on a labelled laboratory environment dataset, and one based on a wind turbine drivetrain bearing failure representing a real-world scenario with stochastic changes of machine state and unknown labels of the bearing condition.

By adjusting co-design parameters, the resulting hybrid conventional/neuromorphic model show a comparable accuracy in detection performance for the laboratory dataset compared to the state-of-the-art reported in the literature.

Similarly, for the wind turbine drivetrain dataset a bearing fault detection time comparable to that in previous work is obtained.

This shows the successful implementation of a hybrid conventional/neuromorphic co-design model for condition monitoring applications, offering novel opportunities to investigate performance trade-offs and efficiency improvements enabled by neuromorphic technologies.

Nowadays, lots of data are generated in industries using vibration sensors to evaluate the equipment’s working condition and identify faults. A significant challenge is that only a small fraction of data can be transmitted for intelligent fault diagnosis and storage. The edge processing capacity is often insufficient for advanced analysis due to time and resource constraints. The neuromorphic signal encoding scheme efficiently reduces the data rate by encoding relevant signal changes into spike trains while discarding redundant information and noise, enabling energy-efficient neuromorphic processing. Due to the presence of dominant operational features and noise in the original measurements, signal pre-processing is required to extract the relevant features before spike coding and processing. The work investigates the effects of different filter banks (pre-processing methods) on the spike encodings for vibration measurements from bearings. This also includes bearing fault features diagnosis based on statistical analysis of generated spikes. The comparative analysis is made for benchmarking different signal pre-processing methods (e.g., envelope, empirical mode decomposition (EMD), and gammatone filter) on bearing vibration datasets. An event-triggered scheme, i.e., Level-crossing analog-to-digital converters (LC-ADCs) is applied to encode the vibration measurement to spikes. Inter-spike intervals (ISIs) statistics are analysed for fault diagnosis of bearings. The results obtained for CWRU bearing databases indicate a possible fault detection and diagnosis with significant data rate reduction and an opportunity for improved computational efficiency. With the developed approach, the envelope filter is found to be the most efficient of all. This work enables a new approach to improve the energy efficiency of condition monitoring systems and further sets a new course of research development in this area using neuromorphic technologies. 

In the last years, the wind industry has increased in a large scale. A wind turbine out of service leeds to high costs due to both maintenance and repair costs and the incapability of producing electricity. A substantial part of the wind turbine failures are in the drivetrain, mainly in generator and gearbox. Several recent works focuses in the study of benefits of the integration of condition monitoring with current maintenance techniques, that would drive to the reduction of costs. For condition monitoring, vibration analysis has been widely accepted as the technique that gives most information about faults in a rotating machine, thus vibration sensors are often used in wind turbine applications. In this work, data from several vibration sensors installed in 18 wind turbines in cold climate were analysed using the Wavelet Packets Transform energy. Signals were acquired for more than four years (from 2011 to 2015), registering failures in gearboxes and generators of the wind turbines. Data were obtained under varying conditions of load and speed as well as varying weather conditions. Signals were analysed with the aim of finding parameters that indicate the presence of a fault. This would be useful to predict a failure with enough time to plan a stop of the wind turbine in the proper moment for similar faults in the future.

Condition monitoring is central to the efficient operation of wind farms due to the challenging operating conditions, rapid technology development, and a large number of aging wind turbines. In particular, predictive maintenance planning requires the early detection of faults with few false positives. Achieving this type of detection is a challenging problem due to the complex and weak signatures of some faults, particularly the faults that occur on the gearbox bearings of a turbine drivetrain. The results of former studies addressing condition-monitoring tasks using dictionary learning indicate that unsupervised feature learning is useful for diagnosis and anomaly detection purposes. However, these studies are based on small sets of labeled data from test rigs operating under controlled conditions that focus on classification tasks, which are useful for quantitative method comparisons but gives little insight into how useful these approaches are in practice or how can be used by existing condition-monitoring systems. Here, we investigate an unsupervised dictionary learning method for condition monitoring using vibration data recorded over 46 months under typical industrial operations. Thus, we contribute real-world industrial vibration data that are made publicly available and novel test results. In this study, dictionaries are learned from gearbox vibrations in six different turbines, and the dictionaries are subsequently propagated over a few years of monitoring data when faults are known to occur. We perform the experiment using two different sparse coding algorithms to investigate if the algorithm selected affects the features of abnormal conditions. We propose a dictionary distance metric derived from the dictionary learning process as a condition indicator and find the time periods of abnormal dictionary adaptation starting six months before a drivetrain bearing replacement and one year before the resulting gearbox replacement. In addition, we investigate the distance between dictionaries learned from geographically close turbines of the same type under healthy conditions. We find that the features learned are similar and that a dictionary learned from one turbine can be useful for monitoring a similar turbine.

The highest costs due to premature failures in wind turbine drivetrains are related to defects in the gearbox, with bearing failures being overrepresented. Vibration monitoring has been identified as the primary tool to detect and diagnose these types of failures. However, late or no signs of the failures are still being reported. Artificial neural networks (ANNs) has been shown to favourably be used as a classifier of bearing failures to increase the detection and diagnosis performance, which requires labelled data when training for all types of considered failures. However, less work has been done with an ANN used to create descriptive functions of the vibration and turbine operation data relationship and thereby negating inherent variance in the vibration data and increasing the detectability when a defect appears. Therefore, this study utilizes the relationship between the rotational speed recorded during a vibration measurement and the calculated condition indicator values of specific bearing failures in three wind turbine gearbox failures. An ANN establishes a function between the rotational speed and condition indicator values with healthy training data collected before the failure occurred. Thereafter, whole datasets leading up to the changing of the gearboxes is used to predict the condition indicator values without the failure influence. The difference between the predicted and true values show an increased sensitivity of the detection in two cases of gearbox output shaft bearing failures as well as indicating a planet bearing failure which with the previous data had gone undetected.

Dynamic simulations are often used to evaluate the vibrational response of rolling-element bearings experiencing defects. Previously, the optimal accelerometer position has been found to be as close as possible to the bearing. However, further details about the influence of rotational symmetry have not been closely investigated. This paper presents a dynamic simulation model of a radially loaded complete spherical roller bearing with a defect in the inner ring and placed in a housing with 72 equally spaced accelerometer positions around the circumference. Surface accelerations have been extracted and transformed into the frequency domain. Thereafter, the vibrational components indicating the defect have been evaluated around the circumference. The results show an optimal position as close as possible to the primary loaded zone and validation test rig experiments show a reasonable qualitative agreement.

Wind turbine drivetrain bearing failures continue to lead to high costs resulting from turbine downtime and maintenance. As the standardised tool to best avoid downtime is online vibration condition monitoring, a lot of research into improving the signal analysis tools of the vibration measurements is currently being performed. However, failures in the main bearing and planetary gears are still going undetected in large numbers. The available field data is limited when it comes to the properties of the stored measurements. Generally, the measurement time and the covered frequency range of the stored measurements are limited compared to the data used in real-time monitoring. Therefore, it is not possible to either reproduce the monitoring or to evaluate new tools developed through research for signal analysis and diagnosis using the readily available field data. This study utilises 12 bearing failures from wind turbine condition monitoring systems to evaluate and make recommendations concerning the optimal properties in terms of measurement time and frequency range the stored measurements should have. The results show that the regularly stored vibration measurements that are available today are, throughout most of the drivetrain, not optimal for research-driven postfailure investigations. Therefore, the storage of longer measurements covering a wider frequency range needs to begin, while researchers need to demand this kind of data.

Wind turbines are often plagued by premature component failures, with drivetrain bearings being particularly subjected to these failures. To identify failing components, vibration condition monitoring has emerged and grown substantially. The fast Fourier transform (FFT) is the major signal processing method of vibrations. Recently, the wavelet transforms have been used more frequently in bearing vibration research, with one alternative being the discrete wavelet transform (DWT). Here, the low‐frequency component of the signal is repeatedly decomposed into approximative and detailed coefficients using a predefined mother wavelet. An extension to this is the wavelet packet transform (WPT), which decomposes the entire frequency domain and stores the wavelet coefficients in packets. How wavelet transforms and FFT compare regarding fault detection in wind turbine drivetrain bearings has been largely overlooked in literature when applied on field data, with non‐ideal placement of sensors and uncertain parameters influencing the measurements. This study consists of a comprehensive comparison of the FFT, a three‐level DWT, and the WPT when applied on enveloped vibration measurements from two 2.5‐MW wind turbine gearbox bearing failures. The frequency content is compared by calculating a robust condition indicator by summation of the harmonics and shaft speed sidebands of the bearing fault frequencies. Results show a higher performance of the WPT when used as a field vibration measurement analysis tool compared with the FFT as it detects one bearing failure earlier and more clearly, leading to a more stable alarm setting and avoidable, costly false alarms.

Wind power has in the last 20 years grown into one of the main sources of renewable energy in the world, with both the amount and size of the turbines increasing substantially. One of the major challenges for the wind power industry is the premature failures of especially the drivetrain components. These failures cause a lot of turbine downtime, which increases the operation and maintenance costs of the turbines. Failures in the gearbox have been shown to lead to the highest downtime and the multitude of bearings within that subsystem is overrepresented in the total amount of component failures. Vibrationbased condition monitoring is considered the best method to find these types of defects early and avoid prolonged turbine downtime. Previous research has therefore been focused on the different aspects of condition monitoring; i.e. measurement technologies, signal analysis of vibration measurements to improve detection and diagnosis as well as the implementation of machine learning solutions. However, the majority of research work has yet to evaluate the performance of new developments using wind turbine field data, and still no fundamentally new developments have seen a large-scale implementation in the industry. Further, it is known that the positioning of the accelerometer, used to measure the vibrations, affects the ability to detect and diagnose defects. However, it is not known how to optimally position the accelerometers to monitor the individual drivetrain sub-systems. Also, previous research does not show how the influence of the measurement properties of the field data affect the ability to detect and diagnose component failures.

_{Therefore, this thesis provides a comprehensive evaluation of how to improve the detection and diagnosis of bearing failures specifically in wind turbine drivetrains. In this thesis, a simulation model was developed to study how the accelerometer positioning affects the measurement quality. Vibration simulations of a similar sized bearing to ones found in the wind turbine drivetrain show an optimal accelerometer position as close to the primary loaded zone of the bearings as possible. The current placement of the accelerometers in the wind turbine drivetrain are often diametrically opposed to the loaded zone, and the performance of the vibration monitoring with respect to the commonly used signal analysis tools could thereby be increased. Further, wavelet-based signal analysis has been evaluated using historical wind turbine drivetrain field data. A new implementation of the wavelet packet transform to analyse enveloped vibration measurements in the frequency domain was developed, where the measurements were decomposed into packets matching the frequency resolution of the fast Fourier transform, and analysing the packet energy spectra. Finally, an anomaly detection solution utilizing an artificial neural network has been implemented to separate the condition indicator values, used for detection and diagnosis, from their inherent variance due to the dynamic turbine operation seen in the drivetrain rotational speed.}

_{The results in this thesis show the inadequacy of the commonly stored vibration measurements to the condition monitoring databases when used in post failure investigations and application of research developments on available field data. Measurements both taken over a long period of time and covering wide frequency range should be stored, instead of the either/or of today. Otherwise, the real-time monitoring of wind turbine drivetrain bearing failures cannot be replicated and monitoring improvements not fully evaluated. By implementing the wavelet packet transform, the detection and diagnosis performance was increased 250% compared to the fast Fourier transform, in an example of gearbox output shaft bearing failure. By implementing the anomaly detection by the artificial neural network, the performance increased further and was able to provide indications in a planet bearing failure case, which was not possible before. For turbine owners, these results provide both practical actions to take and provide an example of an easily implementable signal analysis tool to improve the detection and diagnosis of drivetrain bearing failures. The anomaly detection, which utilizes available historic data from healthy turbines and does not require any amount of labelled data for all considered types of bearing failures, also shows promise to detect failures in the drivetrain components which has been historically problematic. For the research community, the results mainly provides guidance into using historic field data when evaluating new developments. Also, they highlight potential pitfalls one can face using field data and what data properties to look for to successfully show the potential of your work.}

The maintenance cost of wind turbines needs to be minimized in order to keep their competitiveness and, therefore, effective maintenance strategies are important. The remote location of wind farms has led to an opportunistic maintenance strategy where maintenance actions are postponed until they can be handled simultaneously, once the optimal opportunity has arrived. For this reason, early fault detection and identification are important, but should not lead to a situation where false alarms occur on a regular basis. The goal of the study presented in this paper was to detect and identify wind turbine bearing faults by using fault-specific features extracted from vibration signals. Automatic identification was achieved by training models by using these features as an input for a one-class support vector machine. Detection models with different sensitivity were trained in parallel by changing the model tuning parameters. Efforts were also made to find a procedure for selecting the model tuning parameters by first defining the criticality of the system and using it when estimating how accurate the detection model should be. Method was able to detect the fault earlier than using traditional methods without any false alarms. Optimal combination of features and model tuning parameters was not achieved, which could identify the fault location without using any additional techniques.