One of Bosch Rexroth’s newest developments is the ABPAC power unit, which is both modular and configurable. The modular design of the ABPAC is enhanced by a selfcontained Condition Monitoring System (CMS), which can also be used to retrofit existing designs. This dissertation shows how Industry 4.0-Technology provides special advantages for the diverse user profiles. Today, Hydraulic Power Units have either scheduled intervals for preventive maintenance or are repaired in case of component failures. Preventive maintenance concepts, until now, did not fully utilize the entire life expectancy of the components, causing higher maintenance costs and prolonged downtimes. Risk of unscheduled downtime forces the customer to stock an array of spare parts leading to higher inventory costs or in the event a spare is not readily available, the customer may encounter long delivery times and extended downtime. Bearing this in mind, we’ve conceived the idea of a self-contained intelligent Condition Monitoring System including a predictive maintenance concept, which is explained in the following.

To further improve the reliability of STATCOM solutions a predictive maintenance system can give the user valuable readings of capacitor health an online and non-invasive, and especially readings of individual capacitors. This allows the user to replace individual deteriorating capacitor links in an MMC configuration before triggering any safety system. There are a few ways to get readings of capacitor health, equivalent series resistance (ESR) and total capacitance are two parameters that can give the user a good health indication. However, due to the large operating voltage of the STATCOM the ESR are too small to give any reliable reading. The total capacitance is the most reliable health indicator that can be used. This predictive maintenance system utilizes the voltage drop caused by capacitor discharge and the current that goes through. If one looks at the integral of that current along with voltage difference it is possible to make an estimate of the total capacitance. However, in order to add further accuracy to the monitoring system an adaptive filtering method will be applied, the Recursive Least Squares filter or RLS. With this estimation technique it is possible to limit the estimation error to approximately 1% even with a small SNR (signal to noise ratio). To give additional functionality of the system it can also be ran in parallel to the control system for the STATCOM on the FPGA control board, and thus not affecting the performance of the control systems. However, it is not uncommon for the voltage and current to be acquired on different boards and in different places. Thus, due to communication delays it is possible measurement might get skewed or phase shifted towards each other. Therefore, a system which detects this skewing are also of interest, as it could improve the accuracy of the capacitance estimations further.

Predictive Maintenance (PM) has been increasingly adopted in the Automotive industry, in the recent decades along with conventional approaches such as the Preventive Maintenance and Diagnostic/Corrective Maintenance, since it provides many advantages to estimate the failure before the actual occurrence proactively, and also being adaptive to the present status of the vehicle, in turn allowing flexible maintenance schedules for efficient repair or replacing of faulty components. PM necessitates the storage and analysis of large amounts of sensor data. This requirement can be a challenge in deploying this method on-board the vehicles due to the limited storage and computational power on the hardware of the vehicle. Hence, this thesis seeks to obtain low dimensional descriptive features from high dimensional data using Representation Learning. This low dimensional representation will be used for predicting vehicle faults, specifically Turbocharger related failures. Since the Logged Vehicle Data (LVD) was base on all the data utilized in this thesis, it allowed for the evaluation of large populations of trucks without requiring additional measuring devices and facilities. The gradual degradation methodology is considered for describing vehicle condition, which allows for modeling the malfunction/ failure as a continuous process rather than a discrete flip from healthy to an unhealthy state. This approach eliminates the challenge of data imbalance of healthy and unhealthy samples. Two important hypotheses are presented. Firstly, Parallel StackedClassical Autoencoders would produce better representations com-pared to individual Autoencoders. Secondly, employing Learned Em-beddings on Categorical Variables would improve the performance of the Dimensionality reduction. Based on these hypotheses, a model architecture is proposed and is developed on the LVD. The model is shown to achieve good performance, and in close standards to the previous state-of-the-art research. This thesis, finally, illustrates the potential to apply parallel stacked architectures with Learned Embeddings for the Categorical features, and a combination of feature selection and extraction for numerical features, to predict the Remaining Useful Life (RUL) of a vehicle, in the context of the Turbocharger. A performance improvement of 21.68% with respect to the Mean Absolute Error (MAE) loss with an 80.42% reduction in the size of data was observed.

A big hurdle for many companies to start using machine learning is that trending techniques need a huge amount of structured data. One potential way to reduce the need for data is taking advantage of previous knowledge from a related task. This is so called transfer learning. A basic description of it would be when you take a model trained on existing data and reuse that for another problem. The purpose of this master thesis is to investigate if transfer learning can reduce the need for data when faced with a new machine learning task which is, in particular, to use transfer learning on ultrasound spectrograms of weld joints for predictive maintenance. The base for transfer learning is VGGish, a convolutional neural network model trained on audio samples collected from YouTube videos. The pre-trained weights are kept, and the prediction layer is replaced with a new prediction layer consisting of two neurons. The whole model is re-trained on the ultrasound spectrograms. The dataset is restricted to a minimum of ten and a maximum of 100 training samples. The results are evaluated and compared to a regular convolutional neural network trained on the same data. The results show that transfer learning improves the test accuracy compared to the regular convolutional neural network when the dataset is small. This thesis project concludes that transfer learning can reduce the need for data when faced with a new machine learning task. The results indicate that transfer learning could be useful in the industry.

Die Identifikation von flexibel einsetzbaren, mächtigen Verfahren zur Massendatenanalyse und die Schaffung von standardisierbaren Vorgehensmodellen zur Integration dieser Verfahren in IT-Systeme sind zentrale Herausforderungen für die moderne Wirtschaftsinformatik. Insbesondere für KMU ist die Entwicklung standardisierter Lösungsansätze von großer Relevanz. Dies gilt über alle Branchen. Finanzdienstleister sind ebenso betroffen wie der Maschinen- und Anlagenbau. Im Rahmen dieser Forschungsarbeit wird eine Wissensbasis geschaffen werden, welche es einer breiten Masse an Unternehmen ermöglicht, geeignete quantitative Methoden zur Datenanalyse zu erkennen und diese für sich nutzbar zu machen.

Predictive maintenance is an automated technique that analyses sensor data from industrial systems to enable downtime planning. Available for this study is unlabelled data from sensors placed in proximity to hydraulic system outlets of a cold isostatic press. There is limited knowledge about degradation processes because of their rarity, but it is still of high importance to minimise them. One approach to overcome this obstacle is by implementing machine learning to recognise deviations from normal behaviour and potentially learn about them. The state-of-the-art machine learning algorithms for situations with little to no knowledge about anomalies in different machines are deep learning variants using unsupervised learning and transfer learning. With the foundation of such research, this study analyses the available data and proposes three deep learning methods. The testing of these algorithms is made by presenting an equal amount of healthy and simulated unhealthy data as input. The output measurement threshold is adjusted to minimise false negatives because of safety reasons. Consequently, the best method (denoising autoencoder) results in 94% accuracy for separating the data and 74% when also identifying the source of error. However, the results should be taken with caution as the simulated faulty data is not fully representative of a real scenario. These algorithms indicate to what extent they are capable of separating deviations from normal data. This thesis provides knowledge about predictive maintenance and lays a foundation for implementing automatic anomaly detection with deep learning in a high-pressure system.

With the developments of industry 4.0, data analytics solutions and their applications have become more prevalent in the manufacturing industry. Currently, the typical software architecture supporting these solutions is modular, using separate software for data collection, storage, analytics, and visualization. The integration and maintenance of such a solution requires the expertise of an information technology team, making implementation more challenging for small manufacturing enterprises. To allow small manufacturing enterprises to more easily obtain the benefits of industry 4.0 data analytics, a full-stack data analytics framework is presented and its performance evaluated as applied in the common industrial analytics scenario of predictive maintenance. The predictive maintenance approach was achieved by using a full-stack data analytics framework, comprised of the PTC Thingworx software suite. When deployed on a lab-scale factory, there was a significant increase in factory uptime in comparison with both preventative and reactive maintenance approaches. The predictive maintenance approach simultaneously eliminated unexpected breakdowns and extended the uptime periods of the lab-scale factory. This research concluded that similar or better results may be obtained in actual factory settings, since the only source of error on predictions would not be present in real world scenarios.

Predictive maintenance is a topic that has been researched and theorized for decades. With the advent of Industry 4.0 and greater technological capabilities in the form of advanced AI, the concept of predicting the need for maintenance in a system or its components is quickly becoming more of a reality for complex processes. The possibility of estimating remaining useful lifetimes would help businesses with maintenance scheduling to avoid unnecessary maintenance actions, but also process failures. Predicting when maintenance is needed would ensure system or component reparation or replacement before they are degraded to the point of negative product quality impact and production losses. While there are many studies on predictive maintenance and how it can be implemented and used in continuous processes, the research on complex batch processes is minimal. Therefore, this thesis aims to construct a framework based on literature for implementing predictive maintenance in batch processes. Parts of the framework are then applied and validated on a complex batch process in the form of sintering at Seco Tools AB. Recommendations are given on how to implement predictive maintenance and what is required in the company’s specific case based on the sintering process’ agreement with the framework. The framework consists of two main parts with several underlying requirements: Data collection and pre-processing and Predictive models. Evaluating the sintering process based on these requirements reveals that many parts of the framework are already in place or possible to implement, while other areas are lacking. There is a need for data cleaning and data related to component health and issues, while the amount of specific parameter data on temperatures, pressures, and similar variables is large. It is possible to predict these parameters accurately through building, training, and validating linear regression models. These predictions can be used as inputs in future models to predict the Remaining Useful Life (RUL) of components or the entire system. Due to the inherent complexity of the sintering process and similar industrial manufacturing processes, which involve numerous interdependent variables affecting product quality and component health, it is imperative to develop machine learning models and neural networks for future predictive maintenance algorithms. Moreover, as highlighted in this thesis, the attainment of predictive maintenance in an industrial environment necessitates prioritizing augmented data collection on component conditions, investing in hardware to bolster computational power, and acquiring the essential expertise to design and implement tailored predictive maintenance algorithms for dedicated manufacturing processes.

This thesis leverages vibration-based unsupervised learning and deep transfer learning to reduce the manual labour involved in building algorithms that perform intelligent fault detection (IFD) on roller element bearings. A review of theory and literature in the field of IFD is presented, and challenges are discussed. An issue is then introduced; current machine learning models built for IFD show strong performance on a small subset of specific data, but do not generalize to a broader range of applications. Signal processing, machine learning, and transfer learning concepts are then explained and discussed. Time-frequency fingerprinting, as well as feature engineering, is used in conjunction with principal component analysis (PCA) to prepare vibration signals to be clustered by a gaussian mixture model (GMM). This process allows for the intelligent referral of data towards algorithms that have performed well on similar datasets and favours the re-use of domain-specific tasks. An algorithm is then proposed that promotes generalization in convolutional neural networks (CNNs) and simplifies the hyperparameter tuning process to allow machine learning models to be applied to a broader set of problems. The machine learning process is then automated as much as possible through meta learning and ensemble models: data similarity measurements are used to evaluate the data fit for transfer and propose training guidelines. Throughout the thesis, three open-source bearing fault datasets are used to test and validate the hypotheses. This thesis focuses on developing and adapting current deep learning models to succeed in challenging domains and real-world scenarios, while improving performance with unsupervised learning and transfer learning.

Identifying faults in machinery before they cause critical failure is the core purpose of condition monitoring. This report gives a background to condition monitoring and outlines the current state of research in the field, and its most important theoretical components. It also describes parallels between sustainability goals and condition monitoring. Further, a method for creating a statistical model to predict faults in machines is described. The proposed model is machine specific and is evaluated on three cases. The model’s predictions is then compared to general limit values provided by an ISO-standard. The model successfully detected faults in time for repair in two of the three cases where the ISO-standard did not. The third case was a control and featured a machine with no issues. Neither our model nor the ISO-standard falsely predicted a fault on the control. From the results of the three cases it is concluded that the proposed machine specific approach is required for reliably predicting faults. / Syftet med tillståndsövervakning är att identifiera fel före de orsakar yterligare fel. Denna rapport ger en bakgrund till tillståndsövervakning samt redogör för den aktuella forskningen och de mest centrala teoretiska grunderna inom området. Rapporten beskriver även hur tillståndsövervakningen bidrar till de globala hållbarhetsmålen samt föreslås en konkret metod för tillämpning av tillståndsövervakning. Den föreslagna modellen är maskinspecifik och grundar sig på statistiska avvikelser av vibrationsdata som samlas från maskiner i ett välfungerande tillstånd. Modellen appliceras på tre olika maskiner och resultaten jämförs med ISO-standarden som har definierat generella gränsvärden för flera maskintyper. Den föreslagna modellen visar lovande resultat genom att upptäcka fel som ISO-standarden missade. Av resultaten från fallen dras slutsatsen att en generella gränsvärden inte är tillräckligt, utan en maskinspecifik metod krävs för att, på ett pålitligt sätt, detektera fel.