Due to the advances in digitalization, condition monitoring systems have found numerous applications in the industry due to benefits such as improved reliability and lowered costs through condition-based or predictive maintenance. Condition monitoring systems typically involve elements, such as data acquisition via suitable sensors, data preprocessing, feature extraction and selection, diagnostics, prognostics and (maintenance) decisions based on diagnosis or prognosis. For the application-specific development of a suitable condition monitoring system, each of these elements requires individual settings. Due to the uncertainty of the future, an open question arises in the condition monitoring field, which is reflected in unknown future operating and environmental conditions. This uncertainty needs consideration in all elements of a condition monitoring system. This article focuses on feature extraction and selection, building on the hypothesis that the remaining useful life of a technical system can be predicted with high accuracy utilizing suitable features. In this article, health indicators derived from time-domain features that permit the monitoring of the health of critical system components are presented for predicting the remaining useful life of technical systems. Three distinct application examples based on rubber-metal elements and rolling-element bearings are evaluated to validate the suitability of the presented methods. Experimental data from accelerated lifetime tests conducted under non-stationary operating and environmental conditions are considered to take possible future uncertainties into account. It can be concluded from the acquired results that health indicators derived from the presented time series toolbox are robust to varying operating and environmental conditions.

Several methods, including order analysis, wavelet analysis and empirical mode decomposition have been proposed and successfully employed for the health state estimation of technical systems operating under varying conditions. However, where information such as the speed of rotating machinery, component specifications or other domain-specific information is unavailable, such methods are often infeasible. Thus, this paper investigates the application of classical time-domain features, features from the medical field and novel features from the highly comparative time-series analysis (HCTSA) package, for the health state estimation of rotating machinery operating under varying conditions. Furthermore, several feature selection methods are investigated to identify features as viable health indicators for the diagnostics and prognostics of technical systems. As a case study, the presented methods are evaluated on real-world and experimentally acquired vibration data of bearings operating under varying speed. The results show that the selected features can successfully be employed as health indicators for technical systems operating under varying conditions.

Predictive Maintenance as a desirable maintenance strategy in industrial applications relies on suitable condition monitoring solutions to reduce costs and risks of the monitored technical systems. In general, those solutions utilize model-based or data-driven methods to diagnose the current state or predict future states of monitored technical systems. However, both methods have their advantages and drawbacks. Combining both methods can improve uncertainty consideration and accuracy. Different combination approaches of those hybrid methods exist to exploit synergy effects. The choice of an appropriate approach depends on different requirements and the goal behind the selection of a hybrid approach. In this work, the hybrid approach for estimating remaining useful lifetime takes potential uncertainties into account. Therefore, a data-driven estimation of new measurements is integrated within a model-based method. To consider uncertainties within the system, a differentiation between different system behavior is realized throughout diverse states of degradation. The developed hybrid prediction approach bases on a particle filtering method combined with a machine learning method, to estimate the remaining useful lifetime of technical systems. Particle filtering as a Monte Carlo simulation technique is suitable to map and propagate uncertainties. Moreover, it is a state-of-the-art model-based method for predicting remaining useful lifetime of technical systems. To integrate uncertainties a multi-model particle filtering approach is employed. In general, resampling as a part of the particle filtering approach has the potential to lead to an accurate prediction. However, in the case where no future measurements are available, it may increase the uncertainty of the prediction. By estimating new measurements, those uncertainties are reduced within the data-driven part of the approach. Hence, both parts of the hybrid approach strive to account for and reduce uncertainties. Rubber-metal-elements are employed as a use-case to evaluate the developed approach. Rubber-metal-elements, which are used to isolate vibrations in various systems, such as railways, trucks and wind turbines, show various uncertainties in their behavior and their degradation. Those uncertainties are caused by diverse inner and outer factors, such as manufacturing influences and operating conditions. By expert knowledge the influences are described, analyzed and if possible reduced. However, the remaining uncertainties are considered within the hybrid prediction method. Relative temperature is the selected measurand to describe the element’s degradation. In lifetime tests, it is measured as the difference between the element’s temperature and the ambient temperature. Thereby, the influence of the ambient temperature on the element’s temperature is taken into account. Those elements show three typical states of degradation that are identified within the temperature measurements. Depending on the particular state of degradation a new measurement is estimated within the hybrid approach to reduce potential uncertainties. Finally, the performance of the developed hybrid method is compared to a model-based method for estimating the remaining useful lifetime of the same elements. Suitable performance indices are implemented to underline the differences between the results.

In the industry 4.0 era, there is a growing need to transform unstructured data acquired by a multitude of sources into information and subsequently into knowledge to improve the quality of manufactured products, to boost production, for predictive maintenance, etc. Data-driven approaches, such as machine learning techniques, are typically employed to model the underlying relationship from data. However, an increase in model accuracy with state-of-the-art methods, such as deep convolutional neural networks, results in less interpretability and transparency. Due to the ease of implementation, interpretation and transparency to both domain experts and non-experts, a rule-based method is proposed in this paper, for prognostics and health management (PHM) and specifically for diagnostics. The proposed method utilizes the most relevant sensor signals acquired via feature extraction and selection techniques and expert knowledge. As a case study, the presented method is evaluated on data from a real-world quality control set-up provided by the European prognostics and health management society (PHME) at the conference’s 2021 data challenge. With the proposed method, our team took the third place, capable of successfully diagnosing different fault modes, irrespective of varying conditions.

In all fields, the significance of a reliable and accurate predictive model is almost unquantifiable. With deep domain knowledge, models derived from first principles typically outperforms other models in terms of reliability and accuracy. When it may become a cumbersome or an unachievable task to build or validate such models of complex (non-linear) systems, machine learning techniques are employed to build predictive models. However, the accuracy of such techniques is not only dependent on the hyper-parameters of the chosen algorithm, but also on the amount and quality of data. This paper investigates the application of classical time series forecasting approaches for the reliable prognostics of technical systems, where black box machine learning techniques might not successfully be employed given insufficient amount of data and where first principles models are infeasible due to lack of domain specific data. Forecasting by analogy, forecasting by analytical function fitting, an exponential smoothing forecasting method and the long short-term memory (LSTM) are evaluated and compared against the ground truth data. As a case study, the methods are applied to predict future crack lengths of riveted aluminium plates under cyclic loading. The performance of the predictive models is evaluated based on error metrics leading to a proposal of when to apply which forecasting approach.

Remaining useful lifetime (RUL) predictions as part of a condition monitoring system are focused in more and more research and industrial applications. To establish an efficient and precise estimate of the RUL of a technical product, different uncertainties have to be handled. To minimize the uncertainties of the RUL estimation, a reliable and accurate prognostic approach as well as a good failure threshold are important. Regarding the failure threshold, most often an expert sets a fixed failure threshold. However, neither the a priori known failure threshold nor a fixedthreshold value are feasible in every application. Especially in the case of varying characteristics of the monitored system, an adaptive failure threshold is of great importance concerning the accuracy of the RUL estimation. Rubber-metal-elements, which are used in a wide range of applications for vibration and sound isolation, are mon-itored by thermocouples to allow for lifetime predictions. Therefore, the element’s state is described by its temper-ature during its service life. Aiming to establish accurate RUL predictions of a rubber-metal-element, uncertainties due to nonlinear material characteristics and changing operational conditions have to be considered. Consequently, different temperature-based failure threshold definitions are implemented and compared within a particle filtering approach.

Gummi-Metall-Teile (GM-Teile) werden zur Schwingungsreduktion u. a. in Windenergieanlagen eingesetzt. Mögliche Anwendungen der Teile liegen in Wellen-, Generator- und Getriebelagerungen, Lagern für die Gondel und ihre Komponenten sowie in Drehmomentstützen. Mit dem Ziel eine prädiktive Instandhaltung zu realisieren, soll eine Zustandsüberwachung für die GM-Teile entwickelt werden. Diese Entwicklung basiert auf der Umsetzung diverser Schritte. Neben der funktionalen Betrachtung wird zwingend auch die konstruktive Integration der Sensoren in das überwachte Teil berücksichtigt. Der Schwerpunkt dieser Arbeit liegt auf der verwendeten Messgröße Temperatur, die mittels ausgewählter Sensorik detektiert wird. Dabei werden Lebensdauerversuche unter instationären Betriebsbedingungen durchgeführt, um diese Messdaten zu generieren. In der Datenauswertung werden sie hinsichtlich der Degradierung des GM-Teils analysiert und für die Ermittlung der nutzbaren Restlebensdauer verwendet. Rubber-metal-elements are used for isolation of vibrations e. g. in wind turbines. Possible applications of the elements are shaft bearings, generator bearings, gearbox bearings, bearings for the nacelle and its components and torque supports. In order to realize predictive maintenance, an accurate condition monitoring system for rubber-metal-elements should be developed. During that development different aspects have to be implemented. Additionally to the functional analysis, the constructive integration of the sensors into the monitored part is mandatory. The focus of this work is on the measured variable temperature, which is detected by means of appropriate sensors. Thereby lifetime tests are run under non-stationary operating conditions to generate temperature measurements. During data analysis, the measured data is analyzed regarding the degradation of the rubber-metal-elements and remaining useful lifetimes are estimated.

As the emerging digitalization of technical systems offers immense opportunities to be exploited by means of bigdata analysis, ubiquitous computing and largely networked systems, the digital twin comes into focus to combineall these aspects to an attendant model of an individual system during design phase as well as during operation.Since state-of-art technical systems are growing increasingly complex due to inherent intelligence and increasingfunctionality, i. e. autonomous behavior so far, it becomes considerably challenging to ensure reliability for thosesystems. Many methods were developed to support a reliability focused design or reliability-by-design approachesto tackle this challenge during design process. In field, data-based methods, i. e. condition monitoring enabled bythe rise of machine learning approaches, are exploited to ensure a reliable operation based on the current conditionof the monitored system. In order to take advantage of existing models of system reliability during design phaseand condition monitoring systems during operation, a method is proposed to combine both approaches in order toset up a digital twin with focus on system reliability. The base model of the digital twin is taken from the systemreliability model from the design phase and is used during operation and therein updated to the current reliabilitybased on the state estimation of the condition monitoring system. The approach is illustrated with a case study of arolling bearing test rig.

Rubber-metal-elements are used in a wide range of applications for vibration and sound isola- tion. Nowadays it is state of the art to calculate the lifetimes of these elements under mechanical stress prior to their service life. To establish more reliable and safer rubber-metal-elements, continuous monitoring by dif- ferent sensors can be used. Especially prognostics enable a rise in reliability, availability and safety. To estab- lish these advantages, estimating the remaining useful lifetime of rubber-metal-elements should be realized during its service life based on current information on its condition. Therefore a suitable measure to monitor the condition of the element is necessary. This work focuses on temperature signals. This approach allows in- cluding the ambient temperature and thereby involving changing operating conditions. For estimating the RUL of rubber-metal-elements a model-based prognostics approach based on particle filtering is proposed. Its performance is analyzed regarding relevant parameters to enable the best performance for the applied data.

Zuverlässigkeit, Sicherheit und Verfügbarkeit gewinnen bei der Anwendung von technischen Systemen eine immer größere Bedeutung. Aus diesem Grund hat sich Condition Monitoring, die Zustandsüberwachung eines technischen Produkts, in verschiedenen Industriebranchen etabliert. Die sensorbasierte Überwachung eines Produkts während seiner Betriebsdauer in Kombination mit Condition Monitoring Methoden ermöglichen die Bestimmung des aktuellen Zustands des Produkts und somit eine Diagnose, ob das Produkt seine ihm zugeschriebene Funktion zum aktuellen Zeitpunkt erfüllt. Neben Diagnosen bietet Condition Monitoring auch die Möglichkeit Prognosen aufzustellen, dabei wird die restliche Nutzungsdauer des Produkts aufbauend auf geeigneten Sensordaten geschätzt. So kann eine intelligente Wartungsplanung umgesetzt werden, die im Gegensatz zu klassischen Ansätzen keine festen Wartungsintervalle benötigt und die Nachteile einer rein reaktiven Wartung kompensiert. Stattdessen ist es möglich ein Element bis vor das Ende seiner Lebensdauer zu nutzen und erst dann zu warten, um eine optimale Nutzung zu gewährleisten. Durch eine Bestimmung der verbleibenden Restlebensdauer während des Betriebs ist eine optimale Wartungsplanung möglich, wodurch die Verfügbarkeit und die Auslastung der überwachten Produkte signifikant gesteigert werden kann. In dieser Arbeit soll ein produktspezifisches Condition Monitoring System für Gummi-Metall-Elemente entwickelt werden. Diese Elemente werden zur Federung, Geräusch- und/oder Schwingungsisolation in vielen verschiedenen Anwendungen eingesetzt, wie bspw. in Nutz- und Schienenfahrzeugen oder Windenergieanlagen. In Industrie und Forschung werden bereits Zustandsüberwachungen von Systemen mit integrierten Gummi-Metall-Elementen eingesetzt, allerdings noch keine Condition Monitoring Systeme zur alleinigen Zustandsüberwachung dieser Elemente. Aktuell ist es üblich die Lebensdauer dieser Elemente aufbauend auf beschleunigten Lebensdauerversuchen und Erfahrungswerten abzuschätzen. Mit dem Ziel die Lebensdauer des fokussierten Produkts präziser vorherzusagen und damit eine intelligente Wartungsplanung zu ermöglichen, wird die Entwicklung eines Condition Monitoring Systems für Gummi-Metall-Elemente angestrebt und in dieser Arbeit erläutert.

In vielen verschiedenen Industriezweigen hat sich Condition Monitoring aufgrund seiner finanziellen und sicherheitstechnischen Vorteile bereits etabliert. Um die Verlässlichkeit und die Auslastung zu steigern, sowie um die Lebenszykluskosten zu reduzieren, steigt auch im Schienenfahrzeugbereich die Anzahl an eingesetzten Condition Monitoring Systemen. Studien zu Versagensmodi von Schienenfahrzeugen haben gezeigt, dass Versagensursachen meistens in den Radprofilen oder im Fahrwerk liegen [1]. Wird das Fahrwerk heute mittels Condition Monitoring überwacht, werden hierfür häufig Sensoren an den Wagons angebracht, um bspw. deren Schwingungen zu kontrollieren [2, 3]. In dieser Arbeit liegt der Fokus auf Gummi-Metall-Elementen (GM-Elementen) der Jörn GmbH.; als elastische Lager im Drehgestell sind diese Teil des Fahrwerks eines Schienenfahrzeugs. Mit dem Ziel die Wartungsplanung dieser Elemente zu optimieren, ist untersucht worden, ob diese Elemente einzeln mittels Condition Monitoring überwacht werden können. Die hierfür durchgeführten beschleunigten Lebensdauertests werden im nächsten Abschnitt erläutert. Anschließend werden die modellbasierten Methoden dargestellt, die aufbauend auf den im Versuch aufgezeichneten Daten eine Prognose der nutzbaren Restlebensdauer (RUL, remaining useful lifetime) der GM-Elemente aufstellen. Im letzten Abschnitt folgen eine kurze Zusammenfassung und ein Ausblick.

In der Windenergieindustrie haben die Größen Zuverlässigkeit, Sicherheit und Verfügbarkeit eine enorme Bedeutung erlangt aufgrund des Trends Windenergieanlagen zur optimalen Windausnutzung an schwer zugänglichen Positionen aufzustellen, wie bspw. Offshore. Dies führt zu erschwerten Wartungsbedingungen und damit zu höheren Kosten. Der Einsatz von Condition Monitoring hat sich in dieser Industrie etabliert, denn diese Technik ermöglicht eine Zustandsdiagnose des überwachten Systems und eine Prognose seiner nutzbaren Restlebensdauer (remaining useful life: RUL), jeweils basierend auf geeigneten Sensordaten. In dieser Arbeit wird ein Konzept für ein produktspezifisches Condition-Monitoring-System für Gummi-Metall-Elemente (GM-Elemente) vorgestellt, welches den Schwerpunkt auf die Prognose der RUL dieser Elemente setzt. In Windenergieanlagen werden zahlreiche GM-Elemente zur Geräusch- und Schwingungsisolation verwendet. Der Einsatz des hier vorgestellten produktspezifischen Condition-Monitoring-Systems kann somit einen erheblichen Beitrag zum verlässlichen Betrieb von Windenergieanlagen liefern, da die Überwachung einzelner Komponenten in die Zustandsüberwachung der gesamten Anlage integriert und dadurch der Betrieb der Anlage optimiert werden kann. In dieser Arbeit werden einige Herausforderungen diskutiert, die sich bei der Entwicklung eines Condition-Monitoring-Systems für GM-Elemente ergeben. So wird evaluiert, welche Größen sich zur Beschreibung der Alterung eines spezifischen Elements eignen und wie diese gemessen werden können. Temperaturen werden bereits in einigen technischen Systemen, wie auch in Windenergieanlagen, aufgezeichnet und ausgewertet, aber ihr Potential für die Bestimmung der RUL der überwachten Komponente ist noch nicht ausgeschöpft. Hier wird eine Lösungsmöglichkeit vorgestellt, die auf Temperatursensoren aufbaut. Als Grundlage für die Entwicklung des Condition-Monitoring-Systems wurden beschleunigte Lebensdauerversuche der GM-Elemente auf einem Versuchsstand zur Schwingungsanalyse durchgeführt. In diesen Lebensdauerversuchen wird die mechanische Alterung eines GM-Elements über einen kraftgeregelten Hydraulikzylinder erzielt. Dabei wird das Ende der Lebensdauerversuche in einem ersten Schritt über die Wegamplitude des Zylinders bestimmt. Während dieser Versuche wurden diverse Sensoren eingesetzt. Die aufgezeichneten Temperaturdaten zeigen, dass sich Temperaturmessungen eignen die Lebensdauer von GM-Elementen mittels Condition Monitoring Prognosemethoden zu schätzen.