Este trabalho apresenta um procedimento para auxiliar na determinação da vida útil de pavimentos flexíveis através da determinação de tensões e deformações causadas pela solicitação de um eixo padrão na estrutura de pavimentos flexíveis utilizando Redes Neurais Artificiais. Para treinamento e validação das redes foram utilizadas bacias de deflexões hipotéticas geradas com o auxílio do programa ELSYM5, simulando o carregamento com falling weight deflectometer. Foram criados quatro conjuntos de bacias hipotéticas, dois para pavimentos de três camadas e dois para pavimentos de quatro camadas. As redes neurais artificiais foram treinadas e validadas utilizando-se o simulador EasyNN-plus, que utiliza redes multilayer perceptron com algoritmo de aprendizagem backpropagation. Os dados de entrada das redes são as espessuras das camadas do pavimento e a bacia de deflexão. Como saída, têm-se as tensões e deformações na face inferior do revestimento e no topo do subleito e os módulos de resiliência das camadas do pavimento. Foram determinadas retas de regressão, coeficientes de regressão e histogramas de erros entre os valores reais (ELSYM5) e os valores previstos (RNA). Os resultados obtidos pelas redes neurais artificiais apresentaram boa correlação com os valores reais, demonstrando a capacidade das redes neurais para auxiliar na determinação da vida útil de pavimentos flexíveis, ao estimar diretamente as tensões e deformações em pontos específicos da estrutura. / This paper presents a procedure to assist the evaluation of the remaining life of flexible pavements by means of the determination of stresses and strains caused by a standard load in flexible pavements structures using artificial neural networks. Hypothetical deflections basins, generated by the ELSYM5 program, simulating the load applied by a falling weight deflectometer, were used to train and to validate the networks. Four sets of hypothetical basins were created, two for pavements with three layers and two for pavements with four layers. The artificial neural networks were trained and validated using the EasyNN-plus simulator, which uses multilayer perceptron networks with back-propagation learning algorithm. The networks input data are the pavements layers thickness and the deflection basin. The networks outputs are the stresses and strains in the bottom of the asphalt layer and at the top of the subgrade and resilience modulus of the pavement layers. The results obtained by the artificial neural networks showed good correlation with the real values, demonstrating that neural networks have capacity to assist in the evaluation of the remaining life of flexible pavements, estimating directly the stresses and strains of specific points of the pavement structure.

The problem of vibration induced structural faults has been a real one in engineering over the years. If left unchecked it has led to the unexpected failures of so many structures. Needless to say, this has caused both economic and human life losses. Therefore for over forty years, structural damage identification has been one of the important research areas for engineers. There has been a thrust to develop global structural damage identification techniques to complement and/or supplement the long-practised local experimental techniques. In that respect, studies have shown that vibration-based techniques prove to be more potent. Most of the existing vibration-based techniques monitor changes in modal properties like natural frequencies, damping factors and mode shapes of the structural system to infer the presence of structural damage. Literature also reports other techniques which monitor changes in other vibration quantities like the frequency response functions, transmissibility functions and time-domain responses. However, none of these techniques provide a complete identification of structural damage. This study presents a damage detection technique based on operational response monitoring, which can identify all the four levels of structural damage and be implemented as a continuous structural health monitoring technique. The technique is based on monitoring changes in internal data variability measured by a test statistic <font face="symbol">c</font>2Ovalue. Structural normality is assumed when the <font face="symbol">c</font>2Om value calculated from a fresh set of measured data is within the limits prescribed by a threshold <font face="symbol">c</font>2OTH value . On the other hand, abnormality is assumed when this threshold value has been exceeded. The quantity of damage is determined by matching the <font face="symbol">c</font>2Om value with the <font face="symbol">c</font>2Op values predicted using a benchmark finite element model. The use of <font face="symbol">c</font>2O values is noted to provide better sensitivity to structural damage than the natural frequency shift technique. The analysis carried out on a numerical study showed that the sensitivity of the proposed technique ranged from three to thousand times as much as the sensitivity of the natural frequencies. The results from a laboratory structure showed that accurate estimates of damage quantity and remaining service life could be achieved for crack lengths of less than 0.55 the structural thickness. This was due to the fact that linear elastic fracture mechanics theory was applicable up to this value. Therefore, the study achieved its main objective of identifying all four levels of structural damage using operational response changes. / Dissertation (MSc (Mechanics))--University of Pretoria, 2007. / Mechanical and Aeronautical Engineering / unrestricted

Tato práce se zabývá vytvořením simulačního modelu dvojčinného pneumatického pístu s mechanickou sestavou, včetně modelů snímačů, s následujícím odhadem parametrů a aproximací chování demonstračního zařízení. Dalším cílem je prezentace různých přístupů prediktivní údržby na datové sadě měřené na demonstračním zařízení. Na měřený datový soubor se aplikovaly signal-based techniky bez použití simulačního modelu a model-based metody, které vyžadují použití simulačního modelu. Výsledkem této práce je ověření možnosti monitorování stavu zařízení pomocí nainstalovaných senzorů a vyhodnocení efektivity senzorů z hlediska přesnosti a finančních nákladů.

Essay is devoted to assessment of the existing structure. Primarily is focused on the buildings history, performed structures upgrades, design of structural plans and design process to improve remaining working life of the building.

There are three ways to deal with component failure: reactive maintenance, preventive maintenance, and predictive maintenance. Reactive maintenance is to repair only once something breaks. Preventive maintenance is to repair before it breaks, independent of actual wear. Predictive maintenance is performed on the basis of real time operational data, repairing when components cross a certain degradation threshold.  With classification models one can determine the health state of a component. Regression models, on the other hand, allow the user to calculate a more precise estimate of remaining useful life. Previous research on regression models have exclusively used sensory data while classification models have used both sensory data as well as log-data. Research on predictive maintenance using regression models have found most success using SVM regression, decision trees, random forest regression, artificial neural networks and LSTM models.  Companies have more and more data to their disposal about the performance of their machines, but usually in the form of log-data. The goal of this research is to find if it is possible to use log-data for regression models. If this is the case, more sophisticated regression models can be used to apply predictive maintenance more accurately on a broader scale than is currently the case. The project was performed through a case study at a company in the semiconductor industry in the Netherlands, with years of log-data of their product that are gradually degrading over time. After quantifying the log-data and trying all kinds of different regression models in combination with different time scales, the results were unilaterally abysmal and were unable to make any decent prediction.  The reason for this according to several experts in the field of data science is that there was no in depth understanding of the data. They say it is required to have an integral understanding of the log-data and to closely collaborate with field engineers who know the data in and out. If a field engineer can say something about the degradation of a machine using only the log-data, a machine learning model can do it too. If a machine learning model is unable to purposefully overfit on the training data and the results are bad, there is no signal in the dataset and the task is impossible. It does not matter if the data was originally sensory or log-based, the only thing that matters is understanding what the data means and the presence of the degradation signal within.

Assessment of machine health and prediction of future failures are critical for maintenance decisions. Many of the existing methods use unsupervised techniques to construct health indicators by measuring the disparity between the current state and either the healthy or the faulty states of the system. This approach can work well, but if the resulting health indicators are insufficient there is no easy way to steer the algorithm towards better ones. In this thesis a new method for health indicator construction is investigated that aims to solve this issue. It is based on measuring distance after transforming the sensor data into a new space using a feed-forward neural network. The feed-forward neural network is trained using a multi-objective optimization algorithm, NSGA-II, to optimize criteria that are desired in a health indicator. Thereafter the constructed health indicator is passed into a gated recurrent unit for remaining useful life prediction. The approach is compared to benchmarks on the NASA Turbofan Engine Degradation Simulation dataset and in regard to the size of the neural networks, the model performs relatively well, but does not outperform the results reported by a few of the more recent methods. The method is also investigated on a simulated dataset based on elevator weights with two independent failures. The method is able to construct a single health indicator with a desirable shape for both failures, although the latter estimates of time until failure are overestimated for the more rare failure type. On both datasets the health indicator construction method is compared with a baseline without transformation function and does in both cases outperform it in terms of the resulting remaining useful life prediction error using the gated recurrent unit. Overall, the method is shown to be flexible in generating health indicators with different characteristics and because of its properties it is adaptive to different remaining useful life prediction methods. / Estimering av maskinhälsa och prognos av framtida fel är kritiska steg för underhållsbeslut. Många av de befintliga metoderna använder icke-väglett (unsupervised) lärande för att konstruera hälsoindikatorer som beskriver maskinens tillstånd över tid. Detta sker genom att mäta olikheter mellan det nuvarande tillståndet och antingen de friska eller fallerande tillstånden i systemet. Det här tillvägagångssättet kan fungera väl, men om de resulterande hälsoindikatorerna är otillräckliga så finns det inget enkelt sätt att styra algoritmen mot bättre. I det här examensarbetet undersöks en ny metod för konstruktion av hälsoindikatorer som försöker lösa det här problemet. Den är baserad på avståndsmätning efter att ha transformerat indatat till ett nytt vektorrum genom ett feed-forward neuralt nätverk. Nätverket är tränat genom en multi-objektiv optimeringsalgoritm, NSGA-II, för att optimera kriterier som är önskvärda hos en hälsoindikator. Därefter används den konstruerade hälsoindikatorn som indata till en gated recurrent unit (ett neuralt nätverk som hanterar sekventiell data) för att förutspå återstående livslängd hos systemet i fråga. Metoden jämförs med andra metoder på ett dataset från NASA som simulerar degradering hos turbofan-motorer. Med avseende på storleken på de använda neurala nätverken så är resultatet relativt bra, men överträffar inte resultaten rapporterade från några av de senaste metoderna. Metoden testas även på ett simulerat dataset baserat på elevatorer som fraktar säd med två oberoende fel. Metoden lyckas skapa en hälsoindikator som har en önskvärd form för båda felen. Dock så överskattar den senare modellen, som använde hälsoindikatorn, återstående livslängd vid estimering av det mer ovanliga felet. På båda dataseten jämförs metoden för hälsoindikatorkonstruktion med en basmetod utan transformering, d.v.s. avståndet mäts direkt från grund-datat. I båda fallen överträffar den föreslagna metoden basmetoden i termer av förutsägelsefel av återstående livslängd genom gated recurrent unit- nätverket. På det stora hela så visar sig metoden vara flexibel i skapandet av hälsoindikatorer med olika attribut och p.g.a. metodens egenskaper är den adaptiv för olika typer av metoder som förutspår återstående livslängd.

Predictive maintenance plays a crucial role in preventing unexpected equipment failures and maintaining assets in good operating conditions in various systems. One such scenario where predictive maintenance has been widely used is in battery management systems for electronic vehicles based on lithium batteries, where the risk of failure can be reduced by predicting the remaining useful life of the lithium battery. This project developed a DL model based on Long Short-Term Memory networks which was able to generalize new and various kinds of battery. The model was implemented on a low-cost, low-power using embedded artifcial intelligence, which enables local model execution, reducing costs, time, and risks associated with transferring data to the cloud. To further optimize the model and reduce its memory usage, quantization was applied before porting it to an embedded system based on the STM32 MCU. The results show that the model migration was successful, with low memory cost and no signifcant degradation in accuracy. Finally, the memory usage of the prediction model was also analyzed. / Predictiv underhåll har en avgörande roll för att förebygga oväntade utrustningsfel och bibehålla tillgångar i god driftsvillkor i olika system. Ett scenario där predictivt underhåll har använts mycket är i batterihanteringssystem för elfordon baserade på litiumbatterier, där risken för fel kan reduceras genom att förutsäga den återstående användbarhetsperioden för litiumbatteriet. I det här projektet utvecklades djupinlärningsprediktiva modeller med hjälp av Keras sekventiella modell för att representera en ferlagersneural nätverk och en Lång Korttidsminne modell för tidserieprediktion. Dessa modeller implementerades på en lågkostnad, låglägesmikrokontroller med inbyggd artifcial intelligence, vilket möjliggör lokal modellkörning, vilket reducerar kostnader, tid och risker med att överföra data till molnet. För att ytterligare optimera modellen och minska dess minnesfotavtryck tillämpades kvantisering innan den portades till en inbyggd system baserat på STM32 mikrokontroller. Resultaten visar att modellmigrationen var framgångsrik, med låg minneskostnad och ingen signifkant försämring av precisionen. Slutligen analyserades även minnesanvändningen av prediktionsmodellen.

The growing demand for sustainable technology has led to an increased application of power electronics. As these devices are often exposed to harsh conditions, their reliability is a primary concern for both manufacturers and users. Addressing these reliability challenges involves a set of activities known as Prognostics and Health Management (PHM). In PHM, predicting the Remaining Useful Life (RUL) is crucial. This prediction relies on identifying failure precursors, which signify the presence of degradation. These precursors are then used to construct a degradation model that enables the prediction of the remaining time that the device can work before failure. The project focuses on examining a MOSFET aging dataset from the NASA PCoE dataset depository and a diode aging dataset from Fraunhofer ENAS. The prediction of the remaining useful life of devices using failure precursors has been done by applying recurrent neural network (RNN) methods. However, the prediction results from a single feature is significantly deviated from the actual values. To improve the prediction, the age of the device was proposed as an additional feature. RNNs with a similar number of weights and RNNs with the same hyperparameters are implemented and their performance is evaluated by the accuracy of prediction. The results show that all the RNN models implemented manage to capture the characteristics of the aging data. Despite its simpler structure, the vanilla RNN manages to produce a comparable result with the GRU and LSTM by simpler mechanism and less number of weights. The results also reveal that the characteristics of the data have a significant impact on the final results. / Den växande efterfrågan på hållbar teknik har lett till en ökad tillämpning av kraftelektronik. Eftersom dessa enheter ofta utsätts för tuffa förhållanden är deras tillförlitlighet ett primärt bekymmer för både tillverkare och användare. Att ta itu med dessa tillförlitlighetsutmaningar innebär en uppsättning aktiviteter som kallas Prognostics and Health Management (PHM). I PHM är det avgörande att förutsäga det återstående användbara livet (RUL). Denna förutsägelse bygger på identifiering av felprekursorer, som anger förekomsten av nedbrytning. Dessa prekursorer används sedan för att konstruera en nedbrytningsmodell som möjliggör förutsägelse av den återstående tiden som enheten kan fungera innan fel. Projektet fokuserar på att undersöka en MOSFET-åldringsdataset från NASA PCoE-datauppsättningen och en diodåldringsdataset från Fraunhofer ENAS. Förutsägelsen av den återstående livslängden för enheter som använder felprekursorer har gjorts genom att använda metoder för återkommande neurala nätverk (RNN). Förutsägelseresultatet från en enskild funktion avviker dock avsevärt från de faktiska värdena. För att förbättra förutsägelsen föreslogs enhetens ålder som en extra funktion. RNN med ett liknande antal vikter och RNN med samma hyperparametrar implementeras och deras prestanda utvärderas av förutsägelsens noggrannhet. Resultaten visar att alla implementerade RNN-modeller lyckas fånga egenskaperna hos åldrande data. Trots sin enklare struktur lyckas vanilj RNN producera ett jämförbart resultat med GRU och LSTM genom enklare mekanism och färre antal vikter. Resultaten visar också att uppgifternas egenskaper har en betydande inverkan på de slutliga resultaten.

Transportsektorn är en av världens största producenter av växthusgas därav är dess avkarbonisering essentiell för att uppnå Parisavtalets mål för CO2-emissioner. Ett viktigt steg för att uppnå dessa mål utförs genom elektrifiering. Litium-jon-batterier (eng. litium-ion batteries, ’LIB’) har blivit väldigt populära energilagringssystem för batteridrivna elektriska fordon (eng. battery electric vehicles, ’BEV’) men tenderar att åldras, precis som alla andra batterier. Därav krävs forskning kring batteriföråldring på grund av nedbrytningsprocessernas inverkan på prissättningen, prestationerna och miljöpåverkan av BEV. Olika modeller används för att beskriva batteriernas åldrande. Datadrivna modeller som förutspår batteriers livstid ökar i popularitet vars noggrannhet och prestationer till stor del beror på indatats kvalitet. Formatet för tidsinhämtade data kräver enorma mängder lagringsutrymme, hög processkapacitet och längre processer; något ’reducerad’ eller ’aggregerad’ data delvis åtgärdar. Denna avhandling fokuserar på att utveckla en metodik för användning av dynamiska fordonsdata i ’aggregerad’ form. Tidsloggade data inhämtade från kallklimatstesting av Scanias BEV-prototyp användes varav interaktionseffekterna mellan diverse fordonsparametrar samt deras effekt på batteriåldring utifrån en batteriåldringsmodell analyserades. Olika tillvägagångssätt för strukturering av dynamiska fordonsdata i modellen undersöktes också. Tolv aggregeringsscenarion designades och testades. Dessutom valdes tre scenarion för uppskattningar och jämförelser av återstående användbar livslängd (eng. remaining-useful-life, ’RUL’) tillsammans med resultat från tidsinhämtade data. Slutligen drogs slutsatser om: parameterinteraktioner, struktur av dynamiska fordonsdata och RUL. Flera framtida utvecklingsområden har också föreslagits bland annat: tester av andra aggregeringstekniker, utöka modellen till tjänstefordon samt kategorisera användningsbeteenden av fordon för att förbättra RUL-uppskattningar. / The transport sector is one of the world’s largest greenhouse gas producing sector and it’s decarbonisation is imperative to achieve the CO2 emission targets set by the Paris Agreement. One important step towards achieving these targets is through electrification of the sector. Lithium-ion batteries (LIBs) have become very popular energy storage systems for battery electric vehicles (BEVs). However, LIBs like all other batteries, tend to age. Hence, the study of the battery ageing phenomena is very essential since the degradation in battery characteristics hugely determines the cost, performance and the environmental impact of BEVs. Different modelling approaches are used to represent battery ageing behaviour. Data-driven models for predicting the lifetime of batteries are becoming popular. However, the accuracy and performance of data-driven models largely depends upon the quality of data being used as the input. Time-sampled format of logging data results in huge data files requiring enormous amounts of storage space, high processing power requirements and longer processing times. Instead, using data in a ’reduced’ or ‘aggregated’ form can help in addressing these issues. This thesis work focuses on developing a methodology for using dynamic vehicle data in an ‘aggregated’ form. Time-sampled data from a Scania prototype BEV truck, recorded during cold climate test, was used. The interaction effects between various vehicle parameters and their effect on battery ageing in a battery ageing model were analyzed. Different approaches to structuring dynamic vehicle data for use in the model were also studied. Twelve aggregation scenarios were designed and tested. Furthermore, three scenarios were selected for making remaining-useful-life (RUL) estimations and compared alongside time-sampled data results. Finally, conclusions about parameter interactions, structuring of dynamic vehicle data and RUL estimations were drawn. Several next steps for future work have also been suggested such as testing other aggregation techniques, extending the model to vehicle fleets and categorizing vehicle usage behaviours to make better RUL estimations.

Remaining useful life (RUL) prediction is a problem that researchers in the prognostics and health management (PHM) community have been studying for decades. Both physics-based and data-driven methods have been investigated, and in recent years, deep learning has gained significant attention. When sufficiently large and diverse datasets are available, deep neural networks can achieve state-of-the-art performance in RUL prediction for a variety of systems. However, for end users to trust the results of these models, especially as they are integrated into safety-critical systems, RUL prediction uncertainty must be captured. This work explores an approach for estimating both epistemic and heteroscedastic aleatoric uncertainties that emerge in RUL prediction deep neural networks and demonstrates that quantifying the overall impact of these uncertainties on predictions reveal valuable insight into model performance. Additionally, a study is carried out to observe the effects of RUL truth data augmentation on perceived uncertainties in the model.