Mit der Automatisierung in der Produktion wird oft versucht, den Menschen als mögliche Fehlerquelle zunehmend vom Prozess auszuschließen. Dabei besitzt der Mensch einzigartige und nützliche motorische, sensorische und kognitive Fähigkeiten. Innovative Technologien bieten nun die Grundlage, Automatisierung und menschliche Fähigkeiten ideal zusammenzuführen und somit die Effizienz von Produktionsprozessen deutlich zu steigern. Wir möchten Sie herzlich einladen, diese neuen Möglichkeiten mit uns zu diskutieren. Vertreter aus Forschung und Industrie werden aktuelle Strategien und Entwicklungen vorstellen. In der begleitenden Demo-Session finden Sie Gelegenheit, mit Experten zu sprechen und Technologien auszuprobieren.Ziel ist es, Ihnen einen ersten Einblick zu bieten und dadurch den Grundstein für eigene Anwendungsideen und -projekte zu legen.:1. Andre Schult (Fraunhofer IVV, Dresden): Begrüßung 2. Peter Seeberg (Softing Industrial Automation GmbH): KeyNote: Industrie 4.0 - Revolution durch Maschinelles Lernen 3. Andre Schult (Fraunhofer IVV, Dresden): Selbstlernende Bediener-Assistenzsysteme - Ein Update 4. Dr. Lukas Oehm (Fraunhofer IVV, Dresden): Ideenwerkstatt zukünftiger Projekte 5. Dr. Romy Müller (TU Dresden): Übervertrauen in Assistenzsysteme: Entstehungsbedingungen und Gegenmaßnahmen 6. Diego Arribas (machineering GmbH & Co. KG): Mehr Geschwindigkeit durch Digitales Engineering, Virtuelle Realität und Simulation 7. Sebastian Carsch (Fraunhofer IVV, Dresden): Informationsaustausch im interdisziplinären Entwicklungsprozess 8. Prof. Rainer Groh (TU Dresden): Das menschliche Maß der Interaktion 9. Fanny Seifert (Elco Industrie Automation GmbH): Smart Maintenance - Industrie-Apps als Grundlage für ein durchgängig integriertes Assistenzsystem 10. Markus Windisch (Fraunhofer IVV, Dresden): Cyber Knowledge Systems - Wissensbausteine für die digitalisierte Bauteilreinigung 11. Dr. Marius Grathwohl (MULTIVAC Sepp Hagemüller SE & Co. KG): IoT und Smart Services in agiler Entwicklung – Phasen der digitalen Transformation bei MULTIVAC 12. Andre Schult (Fraunhofer IVV, Dresden): Zusammenfassung und Abschlussdiskussion

In this paper we propose an approach for incremental learning of semi-supervised SVM. The proposed approach makes use of the locality of radial basis function kernels to do local and incremental training of semi-supervised support vector machines. The algorithm introduces a se- quential minimal optimization based implementation of the branch and bound technique for training semi-supervised SVM problems. The novelty of our approach lies in the in the introduction of incremental learning techniques to semisupervised SVMs.

Diese Dissertation verknüpft eisenbahnbetriebswissenschaftliche Grundlagen mit Methoden aus den Disziplinen Unternehmensforschung (Operations Research) und Maschinelles Lernen. Gegenstand ist die auf den mittelfristigen Zeithorizont bezogene Ressourcenplanung von Knoten des Schienengüterverkehrs, die sogenannte taktische Planung. Diese spielt eine wesentliche Rolle für eine wirtschaftliche und qualitativ hochwertige Betriebsdurchführung. Knoten des Schienengüterverkehrs stellen neuralgische Punkte in der Transportkette von Waren auf der Schiene dar. Sie dienen der Durchführung einer Vielzahl unterschiedlicher betrieblicher Prozesse zur Sicherstellung eines definierten Outputs an Zügen in Abhängigkeit eines jeweils gegebenen Inputs. Die Bereitstellung eines zu den Betriebsanforderungen passenden Ressourcengerüsts ist Teil der taktischen Planung und hat wesentlichen Einfluss auf die Qualität der Prozesse in den Knoten, im Speziellen, sowie auf die vor- und nachgelagerte Transportdurchführung im Allgemeinen. Die Bemessung des notwendigen Personals, der Betriebsmittel und der Infrastruktur für einen Betriebstag, die sogenannte Ressourcendimensionierung, ist in der Praxis geprägt durch einen erheblichen manuellen Aufwand sowie eine große Abhängigkeit von der Datenqualität. Vor diesem Hintergrund und zur Überwindung dieser Nachteile schlägt diese Dissertation ein neues Verfahren zur Ressourcendimensionierung vor. Exemplarisch wird der Fokus auf die großen Knoten des Einzelwagenverkehrs gelegt, die sogenannten Rangierbahnhöfe. In diesen werden Eingangszüge zerlegt, Güterwagen entsprechend ihrer Ausgangsrichtung sortiert und gesammelt, sowie neue Ausgangszüge gebildet und bereitgestellt. Nach dem Stand der Technik werden für die Ressourcendimensionierung mehrere Monate bis wenige Wochen vor der Betriebsdurchführung Rangierarbeitspläne erstellt. Diese umfassen einen detaillierten Arbeitsfolgenplan inklusive Terminierung von Prozessen sowie deren Ressourcenbelegung. Die Rangierarbeitspläne bilden die Grundlage für die Ressourcenanforderung. Aufgrund sich ändernder Nebenbedingungen vor dem Betriebstag und dem stochastischen Charakter der Betriebsprozesse sowohl im Netz als auch in den Knoten können die in der taktischen Planung erstellten Rangierarbeitspläne nur begrenzt für die Durchführung verwendet werden. Als Beispiele sollen das Einlegen von Sonderzügen, Unregelmäßigkeiten bei den Transporten und Witterungsauswirkungen angeführt werden. Der betriebene Planungsaufwand begründet sich in den komplexen Zusammenhängen zwischen den Betriebsprozessen und der größtenteils fehlenden EDV-Unterstützung, was eine Ermittlung der Ressourcendimensionierung bisher erschwert. Die Folge ist eine Diskrepanz zwischen der Datenqualität als Eingangsgröße für die Planung und der Präzision des Rangierarbeitsplans als Ausgangsgröße, was als Konsequenz eine Scheingenauigkeit der Planung und unter Umständen eine Über- oder Unterdimensionierung der Ressourcen mit sich bringt. Das zeigt, dass die Planung verkürzt werden muss und neue Hilfsmittel erforderlich sind. Motiviert durch diese Diskrepanz und den neuen Möglichkeiten, die die Methoden aus den Bereichen des Operations Research und des Maschinellen Lernens bieten, stellt diese Dissertation ein neues Planungsverfahren Parabola bereit. Parabola ermittelt mit geringerem Planungsaufwand und hoher Qualität relevante Kenngrößen für die Ressourcendimensionierung in Knoten des Schienengüterverkehrs. Dies beschleunigt den taktischen Planungsprozess, reduziert Scheingenauigkeiten bei der Ressourcendimensionierung vor der Betriebsdurchführung und orientiert sich daran, wann welche Entscheidungen zuverlässig und genau zu treffen sind. Folglich wird die Detailtiefe der Planung mit der Zuverlässigkeit der Daten in Einklang gebracht. Das in der Dissertation bereitgestellte Planungsverfahren Parabola analysiert eine ausreichend große Anzahl errechneter Rangierarbeitspläne und / oder historischer Betriebsdaten. Das dabei trainierte Regressionsmodell wird anschließend zur Bestimmung des Ressourcengerüsts genutzt. Die Kalibrierung der Regressionsmodelle erfordert hinreichend viele Rangierarbeitspläne. Für deren Erzeugung wird exemplarisch am Beispiel von Rangierbahnhöfen in dieser Dissertation ein ganzheitliches mathematisches lineares Programm entwickelt, das erstmalig sämtliche für die taktische Planung eines Rangierbahnhofs relevanten Entscheidungsprobleme vom Zugeingang bis zum Zugausgang abbildet. Dieses beinhaltet die Definition der Verknüpfung zwischen Eingangs- und Ausgangszügen, sogenannter Wagenübergänge, sowie die Terminierung sämtlicher Betriebsprozesse mit ihrer Zuweisung zu örtlichen Mitarbeitern, Betriebsmitteln und Infrastruktur. Die bestehenden mathematischen Modelle in der bisherigen Literatur beschränken sich lediglich auf Teile dieses Problems. Es folgt die systematische Erzeugung von Problemstellungen, sogenannten Instanzen, zur Generierung eines repräsentativen Testpools. Die Instanzen dieses NP-schweren Problems sind für generische, exakte Lösungsverfahren in akzeptabler Zeit nicht zuverlässig lösbar. Daher wird eine maßgeschneiderte Metaheuristik, konkret ein Verfahren der Klasse Adaptive Large Neighborhood Search (ALNS), entwickelt. Diese bewegt sich durch den Lösungsraum, indem schrittweise mittels mehrerer miteinander konkurrierender Subheuristiken eine vorher gefundene Lösung erst zerstört und anschließend wieder repariert wird. Durch unterschiedliche Charakteristika der Subheuristiken und einer statistischen Auswertung ihres jeweiligen Beitrags zum Lösungsfortschritt, gelingt es der ALNS, sich an das Stadium der Lösungssuche und an die jeweilige Problemstruktur anzupassen. Die in dieser Dissertation entwickelte ALNS erzeugt für realistische Instanzen eines Betriebstages Lösungen in hoher Qualität binnen weniger Minuten Rechenzeit. Basierend auf den erzeugten Rangierarbeitsplänen wurden für die Entwicklung des Planungsverfahrens insgesamt fünf Regressionstechniken getestet, die die Ausgangsgrößen der Pläne – Bedarf an Lokomotiven, Personal und Infrastruktur – prognostizieren. Die vielversprechendsten Ergebnisse werden durch die Methoden Tree Boosting sowie Random Forest erzielt, die in über 90 % der Fälle den Ressourcenbedarf für Personale und Lokomotiven exakt und für Infrastruktur mit einer Toleranz von einem Gleis je Gleisgruppe prognostizieren. Damit ist dieses Regressionsmodell nach ausreichender Kalibrierung entsprechend örtlicher Randbedingungen geeignet, komplexere Planungsverfahren zu ersetzen. Die Regressionsmodelle ermöglichen die Abstrahierung von Mengengerüsten und Leistungsverhalten von Knoten des Schienengüterverkehrs. Daher ist beispielsweise ein konkreter Fahrplan von und zu den Knoten nicht mehr notwendige Voraussetzung für die taktische Planung in Rangierbahnhöfen. Da das Regressionsverfahren aus vielen Rangierarbeitsplänen lernt, verringert sich die Abhängigkeit von einzelnen Instanzen. Durch die Kenntnis von vielen anderen Plänen können robustere Ressourcengerüste prognostiziert werden. Neben dem in dieser Dissertation ausgearbeiteten Anwendungsfall in der taktischen Planung in Knoten des Schienengüterverkehrs, eröffnet das vorgeschlagene neue Planungsverfahren Parabola eine Vielzahl an weiteren Einsatzfeldern. Die Interpretation des trainierten Regressionsmodells erlaubt das tiefgründige Verständnis des Verhaltens von Knoten des Schienengüterverkehrs. Dies ermöglicht ein besseres Verstehen der Engpässe in diesen Knoten sowie die Identifikation relevanter Treiber der Ressourcendimensionierung. Weiter können diese Modelle bei der Erstellung von netzweiten Leistungsanforderungen Berücksichtigung finden. Mit der in dieser Dissertation erfolgten Bereitstellung von Parabola wird durch Nutzung neuartiger Methoden aus dem Operations Research und Maschinellen Lernen das Instrumentarium der eisenbahnbetriebswissenschaftlichen Verfahren und Modelle sinnvoll erweitert. / This dissertation combines the knowledge of railway operations management with methods from operations research and machine learning. It focuses on rail freight facilities, especially their resource planning at a tactical level. The resource planning plays a crucial role for economical operations at high quality. The rail freight facilities represent neuralgic points in the transport chain of goods by rail. Their task is to carry out a multitude of different operational processes to ensure a defined output of trains, depending on a given input. Providing resource requirements appropriate to the amount of work has a significant impact on the quality of the processes in the facilities in particular and on the up- and downstream transport performance in general. The correct dimensioning of resource requirements, which include the necessary staff, locomotives, and infrastructure for an operating day, is characterized by a considerable manual effort and a large dependency on the data accuracy. Against this background and to overcome these drawbacks, this dissertation proposes a new method for resource requirements. The focus is on the large facilities of single wagonload traffic, the so-called classification yards, in which inbound trains are disassembled, railcars are classified according to their outbound direction, and new outbound trains are formed. Nowadays, shunting work plans are created several months to a few weeks before operations. These operating plans comprise a detailed work sequence plan, including process scheduling, and resource allocation. The operating plans form the basis for resource requirements. Due to the changing constraints prior to operations, e.g., the addition of special trains, and the stochastic nature of the operational processes, for instance caused by weather conditions, shunting work plans can only be used for execution to a limited extent. This effort is made for planning due to the complex interdependencies between the operational processes and the predominant lack of IT support, which makes it difficult to determine resource requirements. The result is a discrepancy between the accuracy of the data as an input variable and the precision of the shunting work plan as an output variable. This leads to an illusory precision of the planning and possibly to an oversizing or undersizing of the resources. Hence, planning must be shortened and new tools are required. Motivated by this discrepancy and the new possibilities offered by methods from the _elds of operations research and machine learning, this dissertation provides a new planning method Parabola. Parabola determines with less planning effort and at high quality relevant parameters for resource requirements in rail freight facilities. This accelerates the planning process, reduces illusory precision before operations are carried out and enables decision-making with sufficient reliability due to the data accuracy. Consequently, the level of detail of the planning is harmonized with the reliability of the data. The planning procedure Parabola involves the analysis of numerous calculated operating plans and / or historical operating data. This trains a regression model that can then be used to determine the resource requirements. The calibration of the regression models requires many operating plans. For their generation, an integrated mathematical linear program is developed in this dissertation using the example of classification yards; for the first time, one program covers all relevant decision problems of tactical planning in a classification yard, from the train arrival to the train departure. This includes the definition of the connection between inbound and outbound trains, so-called railcar interchanges, as well as the scheduling of all operational processes with their assignment to local staff, locomotives, and infrastructure. All existing mathematical models in the literature are limited to parts of the problem. Thereafter follows a systematic generation of a test pool of problems named instances. The instances of this NP-hard problem cannot be reliably solved within an acceptable time frame with general-purpose solvers. Therefore, a tailored metaheuristic, namely an adaptive large neighborhood search (ALNS), is developed. It moves through the solution space by first destroying and then repairing a solution stepwise. Several competing subheuristics are available for this purpose. The ALNS combines multiple subheuristics, which have different characteristics and contribute to the solution progress, as determined by statistical evaluation. Consequently, the ALNS successfully adapts to the progress of the solution and to the problem structure. The ALNS, which is developed in this dissertation, generates high-quality solutions for realistic instances of an operating day in a few minutes of computing time. Based on the generated operating plans, five regression methods predicting the output variables of the operating plans – demand for locomotives, staff, and infrastructure – are tested. The most promising results are achieved by the methods tree boosting and random forest, which predict the resource requirements in over 90% of the cases for staff and locomotives accurately and for infrastructure with a tolerance of one track per bowl. Thus, a regression model can replace the more complex planning procedures after sufficient calibration according to local restrictions. The regression models allow the abstraction of quantity structures and performance behavior. Hence, for example, a dedicated timetable is no longer a prerequisite for tactical planning in classification yards. Since regression methods learn from many operating plans, the dependence on individual instances is reduced. By knowing many other plans, the regression model can predict robust resource requirements. In addition to the use case in tactical planning in rail freight facilities, the proposed new planning method Parabola opens a multitude of further _elds of application. By interpreting the trained regression model, the behavior of rail freight facilities can be understood in depth. Under certain circumstances, this allows a better understanding of the bottlenecks in these facilities and the relevant drivers of resource dimensioning. Furthermore, these models have potential applications in the design of network-wide performance requirements. By providing Parabola in this dissertation, the toolbox of railroad management science procedures and models is sensibly extended by using novel methods from operations research and machine learning.

Feature selection or variable selection is an important step in different machine learning tasks. In a traditional approach, users specify the amount of features, which shall be selected. Afterwards, algorithm select features by using scores like the Joint Mutual Information (JMI). If users do not know the exact amount of features to select, they need to evaluate the full learning chain for different feature counts in order to determine, which amount leads to the lowest training error. To overcome this drawback, we extend the JMI score and mitigate the flaw by introducing a stopping criterion to the selection algorithm that can be specified depending on the learning task. With this, we enable developers to carry out the feature selection task before the actual learning is done. We call our new score Historical Joint Mutual Information (HJMI). Additionally, we compare our new algorithm, using the novel HJMI score, against traditional algorithms, which use the JMI score. With this, we demonstrate that the HJMI-based algorithm is able to automatically select a reasonable amount of features: Our approach delivers results as good as traditional approaches and sometimes even outperforms them, as it is not limited to a certain step size for feature evaluation.

With the advent of the Semantic Web and Semantic Technologies, ontologies have become one of the most prominent paradigms for knowledge representation and reasoning. The popular ontology language OWL, based on description logics, became a W3C recommendation in 2004 and a standard for modelling ontologies on the Web. In the meantime, many studies and applications using OWL have been reported in research and industrial environments, many of which go beyond Internet usage and employ the power of ontological modelling in other fields such as biology, medicine, software engineering, knowledge management, and cognitive systems. However, recent progress in the field faces a lack of well-structured ontologies with large amounts of instance data due to the fact that engineering such ontologies requires a considerable investment of resources. Nowadays, knowledge bases often provide large volumes of data without sophisticated schemata. Hence, methods for automated schema acquisition and maintenance are sought. Schema acquisition is closely related to solving typical classification problems in machine learning, e.g. the detection of chemical compounds causing cancer. In this work, we investigate both, the underlying machine learning techniques and their application to knowledge acquisition in the Semantic Web. In order to leverage machine-learning approaches for solving these tasks, it is required to develop methods and tools for learning concepts in description logics or, equivalently, class expressions in OWL. In this thesis, it is shown that methods from Inductive Logic Programming (ILP) are applicable to learning in description logic knowledge bases. The results provide foundations for the semi-automatic creation and maintenance of OWL ontologies, in particular in cases when extensional information (i.e. facts, instance data) is abundantly available, while corresponding intensional information (schema) is missing or not expressive enough to allow powerful reasoning over the ontology in a useful way. Such situations often occur when extracting knowledge from different sources, e.g. databases, or in collaborative knowledge engineering scenarios, e.g. using semantic wikis. It can be argued that being able to learn OWL class expressions is a step towards enriching OWL knowledge bases in order to enable powerful reasoning, consistency checking, and improved querying possibilities. In particular, plugins for OWL ontology editors based on learning methods are developed and evaluated in this work. The developed algorithms are not restricted to ontology engineering and can handle other learning problems. Indeed, they lend themselves to generic use in machine learning in the same way as ILP systems do. The main difference, however, is the employed knowledge representation paradigm: ILP traditionally uses logic programs for knowledge representation, whereas this work rests on description logics and OWL. This difference is crucial when considering Semantic Web applications as target use cases, as such applications hinge centrally on the chosen knowledge representation format for knowledge interchange and integration. The work in this thesis can be understood as a broadening of the scope of research and applications of ILP methods. This goal is particularly important since the number of OWL-based systems is already increasing rapidly and can be expected to grow further in the future. The thesis starts by establishing the necessary theoretical basis and continues with the specification of algorithms. It also contains their evaluation and, finally, presents a number of application scenarios. The research contributions of this work are threefold: The first contribution is a complete analysis of desirable properties of refinement operators in description logics. Refinement operators are used to traverse the target search space and are, therefore, a crucial element in many learning algorithms. Their properties (completeness, weak completeness, properness, redundancy, infinity, minimality) indicate whether a refinement operator is suitable for being employed in a learning algorithm. The key research question is which of those properties can be combined. It is shown that there is no ideal, i.e. complete, proper, and finite, refinement operator for expressive description logics, which indicates that learning in description logics is a challenging machine learning task. A number of other new results for different property combinations are also proven. The need for these investigations has already been expressed in several articles prior to this PhD work. The theoretical limitations, which were shown as a result of these investigations, provide clear criteria for the design of refinement operators. In the analysis, as few assumptions as possible were made regarding the used description language. The second contribution is the development of two refinement operators. The first operator supports a wide range of concept constructors and it is shown that it is complete and can be extended to a proper operator. It is the most expressive operator designed for a description language so far. The second operator uses the light-weight language EL and is weakly complete, proper, and finite. It is straightforward to extend it to an ideal operator, if required. It is the first published ideal refinement operator in description logics. While the two operators differ a lot in their technical details, they both use background knowledge efficiently. The third contribution is the actual learning algorithms using the introduced operators. New redundancy elimination and infinity-handling techniques are introduced in these algorithms. According to the evaluation, the algorithms produce very readable solutions, while their accuracy is competitive with the state-of-the-art in machine learning. Several optimisations for achieving scalability of the introduced algorithms are described, including a knowledge base fragment selection approach, a dedicated reasoning procedure, and a stochastic coverage computation approach. The research contributions are evaluated on benchmark problems and in use cases. Standard statistical measurements such as cross validation and significance tests show that the approaches are very competitive. Furthermore, the ontology engineering case study provides evidence that the described algorithms can solve the target problems in practice. A major outcome of the doctoral work is the DL-Learner framework. It provides the source code for all algorithms and examples as open-source and has been incorporated in other projects.

In den letzten Jahren hat sich maschinelles Lernen als hilfreiches Werkzeug zur Vorhersage von simulierten Materialeigenschaften erwiesen. Somit können aufwendige Berechnungen mittels Dichtefunktionaltheorie umgangen werden und bereits bekannte Materialien besser verstanden oder sogar neuartige entdeckt werden. Eine zentrale Rolle spielt dabei der Deskriptor, ein möglichst interpretierbarer Satz von Materialkenngrößen. Diese Arbeit präsentiert einen Ansatz zur Auffindung von Deskriptoren für periodische Multikomponentensysteme, deren Eigenschaften durch die genaue atomare Anordnung mitbeinflusst wird. Primäre Features von Einzel-, Paar- und Tetraederclustern werden über die Superzelle gemittelt und weiter algebraisch kombiniert. Aus den so erzeugten Kandidaten wird mittels Dimensionalitätsreduktion ein geeigneter Deskriptor identifiziert. Zudem stellt diese Arbeit Strategien vor bei der Modellfindung Kreuzvalidierung einzusetzen, sodass stabilere und idealerweise besser generalisierbare Deskriptoren gefunden werden. Es werden außerdem mehrere Fehlermaße untersucht, die die Qualität der Deskriptoren bezüglich Genauigkeit, Komplexität der Formeln und Berücksichtung der atomaren Anordnung charakterisieren. Die allgemeine Methodik wurde in einer teilweise parallelisierten Python-Software implementiert. Als konkrete Problemstellungen werden Modelle für die Gitterkonstante und die Mischenergie von ternären Gruppe-IV Zinkblende-Legierungen "gelernt", mit einer Genauigkeit von 0.02 Å bzw. 0.02 eV. Datenbeschaffung, -analyse, und -bereinigung werden im Hinblick auf die Zielgrößen als auch auf die primären Features erläutert, sodass umfassende Analysen und die Parametrisierung der Methodik an diesem Testdatensatz durchgeführt werden können. Als weitere Anwendung werden Gitterkonstante und Bandlücken von binären Oktett-Verbindungen vorhergesagt. Die präsentierten Deskriptoren werden mit den Fehlermaßen evaluiert und ihre physikalische Relevanz wird abschließend disktutiert. / In the last years, machine learning methods have proven as a useful tool for the prediction of simulated material properties. They may replace effortful calculations based on density functional theory, provide a better understanding of known materials or even help to discover new materials. Here, an essential role is played by the descriptor, a desirably interpretable set of material parameters. This PhD thesis presents an approach to find descriptors for periodic multi-component systems where also the exact atomic configuration influences the physical characteristics. We process primary features of one-atom, two-atom and tetrahedron clusters by an averaging scheme and combine them further by simple algebraic operations. Compressed sensing is used to identify an appropriate descriptor out from all candidate features. Furthermore, we develop elaborate cross-validation based model selection strategies that may lead to more robust and ideally better generalizing descriptors. Additionally, we study several error measures which estimate the quality of the descriptors with respect to accuracy, complexity of their formulas and the capturing of configuration effects. These generally formulated methods were implemented in a partially parallelized Python program. Actual learning tasks were studied on the problem of finding models for the lattice constant and the energy of mixing of group-IV ternary compounds in zincblende structure where an accuracy of 0.02 Å and 0.02 eV is reached, respectively. We explain the practical preparation steps of data acquisition, analysis and cleaning for the target properties and the primary features, and continue with extensive analyses and the parametrization of the developed methodology on this test case. As an additional application we predict lattice constants and band gaps of octet binary compounds. The presented descriptors are assessed quantitatively by the error measures and, finally, their physical meaning is discussed.

Background: Acute promyelocytic leukemia (APL) is considered a hematologic emergency due to high risk of bleeding and fatal hemorrhages being a major cause of death. Despite lower death rates reported from clinical trials, patient registry data suggest an early death rate of 20%, especially for elderly and frail patients. Therefore, reliable diagnosis is required as treatment with differentiation-inducing agents leads to cure in the majority of patients. However, diagnosis commonly relies on cytomorphology and genetic confirmation of the pathognomonic t(15;17). Yet, the latter is more time consuming and in some regions unavailable. - Methods: In recent years, deep learning (DL) has been evaluated for medical image recognition showing outstanding capabilities in analyzing large amounts of image data and provides reliable classification results. We developed a multi-stage DL platform that automatically reads images of bone marrow smears, accurately segments cells, and subsequently predicts APL using image data only. We retrospectively identified 51 APL patients from previous multicenter trials and compared them to 1048 non-APL acute myeloid leukemia (AML) patients and 236 healthy bone marrow donor samples, respectively. - Results: Our DL platform segments bone marrow cells with a mean average precision and a mean average recall of both 0.97. Further, it achieves high accuracy in detecting APL by distinguishing between APL and non-APL AML as well as APL and healthy donors with an area under the receiver operating characteristic of 0.8575 and 0.9585, respectively, using visual image data only. - Conclusions: Our study underlines not only the feasibility of DL to detect distinct morphologies that accompany a cytogenetic aberration like t(15;17) in APL, but also shows the capability of DL to abstract information from a small medical data set, i. e. 51 APL patients, and infer correct predictions. This demonstrates the suitability of DL to assist in the diagnosis of rare cancer entities. As our DL platform predicts APL from bone marrow smear images alone, this may be used to diagnose APL in regions were molecular or cytogenetic subtyping is not routinely available and raise attention to suspected cases of APL for expert evaluation.

Predicting evaporation is essential for managing water resources in basins. Improvement of the prediction accuracy is essential to identify adequate inputs on evaporation. In this study, artificial neural network (ANN) is coupled with several evolutionary algorithms, i.e., capuchin search algorithm (CSA), firefly algorithm (FFA), sine cosine algorithm (SCA), and genetic algorithm (GA) for robust training to predict daily evaporation of seven synoptic stations with different climates. The inclusive multiple model (IMM) is then used to predict evaporation based on established hybrid ANN models. The adjusting model parameters of the current study is a major challenge. Also, another challenge is the selection of the best inputs to the models. The IMM model had significantly improved the root mean square error (RMSE) and Nash Sutcliffe efficiency (NSE) values of all the proposed models. The results for all stations indicated that the IMM model and ANN-CSA could outperform other models. The RMSE of the IMM was 18, 21, 22, 30, and 43% lower than those of the ANNCSA, ANN-SCA, ANN-FFA, ANN-GA, and ANN models in the Sharekord station. The MAE of the IMM was 0.112 mm/day, while it was 0.189 mm/day, 0.267 mm/day, 0.267 mm/day, 0.389 mm/day, 0.456 mm/day, and 0.512 mm/day for the ANN-CSA, ANN-SCA, and ANN-FFA, ANN-GA, and ANN models, respectively, in the Tehran station. The current study proved that the inclusive multiple models based on improved ANN models considering the fuzzy reasoning had the high ability to predict evaporation.

Driven by the need to use mineral resources more sustainably, and the increasing complexity of ore deposits still available for commercial exploitation, the acquisition of quantitative data on mineralogy and microfabric has become an important need in the execution of exploration and geometallurgical test programmes. Hyperspectral drill-core scanning has the potential to be an excellent tool for providing such data in a fast, non- destructive and reproducible manner. However, there is a distinct lack of integrated methodologies to make use of these data through-out the exploration and mining chain. This thesis presents a first framework for the use of hyperspectral drill-core scanning as a pillar in exploration and geometallurgical programmes. This is achieved through the development of methods for (1) the automated mapping of alteration minerals and assemblages, (2) the extraction of quantitative mineralogical data with high resolution over the drill-cores, (3) the evaluation of the suitability of hyperspectral sensors for the pre-concentration of ores and (4) the use of hyperspectral drill- core imaging as a basis for geometallurgical domain definition and the population of these domains with mineralogical and microfabric information.:Introduction Materials and methods Assessment of alteration mineralogy and vein types using hyperspectral data Hyperspectral imaging for quasi-quantitative mineralogical studies Hyperspectral sensors for ore beneficiation 3D integration of hyperspectral data for deposit modelling Concluding remarks References

This dissertation evaluates the possibilities and limitations of rock type identification in rock cutting with conical picks. For this, machine learning in conjunction with features derived from high frequency cutting force measurements is used. On the basis of linear cutting experiments, it is shown that boundary layers can be identified with a precision of less than 3.7 cm when using the developed programme routine. It is further shown that rocks weakened by cracks can be well identified and that anisotropic rock behaviour may be problematic to the classification success. In a case study, it is shown that the supervised algorithms artificial neural network and distributed random forest perform relatively well while unsupervised k-means clustering provides limited accuracies for complex situations. The 3d-results are visualised in a web app. The results suggest that a possible rock classification system can achieve good results—that are robust to changes in the cutting parameters when using the proposed evaluation methods.:1 Introduction...1 2 Cutting Excavation with Conical Picks...5 2.1 Cutting Process...8 2.1.2 Cutting Parameters...11 2.1.3 Influences of Rock Mechanical Properties...17 2.1.4 Influences of the Rock Mass...23 2.2 Ratios of Cutting Force Components...24 3 State of the Art...29 3.1 Data Analysis in Rock Cutting Research...29 3.2 Rock Classification Systems...32 3.2.1 MWC – Measure-While-Cutting...32 3.2.2 MWD – Measuring-While-Drilling...34 3.2.3 Automated Profiling During Cutting...35 3.2.4 Wear Monitoring...36 3.3 Machine learning for Rock Classification...36 4 Problem Statement and Justification of Topic...38 5 Material and Methods...40 5.1 Rock Cutting Equipment...40 5.2 Software & PC...42 5.3 Samples and Rock Cutting Parameters...43 5.3.1 Sample Sites...43 5.3.2 Experiment CO – Zoned Concrete...45 5.3.3 Experiment GN – Anisotropic Rock Gneiss...47 5.3.4 Experiment GR – Uncracked and Cracked Granite...49 5.3.5 Case Study PB and FBA – Lead-Zinc and Fluorite-Barite Ores...50 5.4 Data Processing...53 5.5 Force Component Ratio Calculation...54 5.6 Procedural Selection of Features...57 5.7 Image-Based Referencing and Rock Boundary Modelling...60 5.8 Block Modelling and Gridding...61 5.9 Correlation Analysis...63 5.10 Regression Analysis of Effect...64 5.11 Machine Learning...65 5.11.2 K-Means Algorithm...66 5.11.3 Artificial Neural Networks...67 5.11.4 Distributed Random Forest...70 5.11.5 Classification Success...72 5.11.6 Boundary Layer Recognition Precision...73 5.12 Machine Learning Case Study...74 6 Results...75 6.1 CO – Zoned Concrete...75 6.1.1 Descriptive Statistics...75 6.1.2 Procedural Evaluation...76 6.1.3 Correlation of the Covariates...78 6.1.4 K-Means Cluster Analysis...79 6.2 GN – Foliated Gneiss...85 6.2.1 Cutting Forces...86 6.2.2 Regression Analysis of Effect...88 6.2.3 Details Irregular Behaviour...90 6.2.4 Interpretation of Anisotropic Behaviour...92 6.2.5 Force Component Ratios...92 6.2.6 Summary and Interpretations of Results...93 6.3 CR – Cracked Granite...94 6.3.1 Force Component Results...94 6.3.2 Spatial Analysis...97 6.3.3 Error Analysis...99 6.3.4 Summary...100 6.4 Case Study...100 6.4.1 Feature Distribution in Block Models...101 6.4.2 Distributed Random Forest...105 6.4.3 Artificial Neural Network...107 6.4.4 K-Means...110 6.4.5 Training Data Required...112 7 Discussion...114 7.1 Critical Discussion of Experimental Results...114 7.1.1 Experiment CO...114 7.1.2 Experiment GN...115 7.1.3 Experiment GR...116 7.1.4 Case Study...116 7.1.5 Additional Outcomes...117 7.2 Comparison of Machine Learning Algorithms...118 7.2.1 K-Means...118 7.2.2 Artificial Neural Networks and Distributed Random Forest...119 7.2.3 Summary...120 7.3 Considerations Towards Sensor System...121 7.3.1 Force Vectors and Data Acquisition Rate...121 7.3.2 Sensor Types...122 7.3.3 Computation Speed...123 8 Summary and Outlook...125 References...128 Annex A Fields of Application of Conical Tools...145 Annex B Supplements Cutting and Rock Parameters...149 Annex C Details Topic-Analysis Rock Cutting Publications...155 Annex D Details Patent Analysis...157 Annex E Details Rock Cutting Unit HSX-1000-50...161 Annex F Details Used Pick...162 Annex G Error Analysis Cutting Experiments...163 Annex H Details Photographic Modelling...166 Annex I Laser Offset...168 Annex J Supplements Experiment CO...169 Annex K Supplements Experiment GN...187 Annex L Supplements Experiment GR...191 Annex M Preliminary Artificial Neural Network Training...195 Annex N Supplements Case Study (CD)...201 Annex O R-Codes (CD)...203 Annex P Supplements Rock Mechanical Tests (CD)...204 / Die Dissertation evaluiert Möglichkeiten und Grenzen der Gebirgserkennung bei der schneidenden Gewinnung von Festgesteinen mit Rundschaftmeißeln unter Nutzung maschinellen Lernens – in Verbindung mit aus hochaufgelösten Schnittkraftmessungen abgeleiteten Kennwerten. Es wird auf linearen Schneidversuchen aufbauend gezeigt, dass Schichtgrenzen mit Genauigkeiten unter 3,7 cm identifiziert werden können. Ferner wird gezeigt, dass durch Risse geschwächte Gesteine gut identifiziert werden können und dass anisotropes Gesteinsverhalten möglicherweise problematisch auf den Klassifizierungserfolg wirkt. In einer Fallstudie wird gezeigt, dass die überwachten Algorithmen Künstliches Neurales Netz und Distributed Random Forest teils sehr gute Ergebnisse erzielen und unüberwachtes k-means-Clustering begrenzte Genauigkeiten für komplexe Situationen liefert. Die Ergebnisse werden in einer Web-App visualisiert. Aus den Ergebnissen wird abgeleitet, dass ein mögliches Sensorsystem mit den vorgeschlagenen Auswerteroutinen gute Ergebnisse erzielen kann, die gleichzeitig robust gegen Änderungen der Schneidparameter sind.:1 Introduction...1 2 Cutting Excavation with Conical Picks...5 2.1 Cutting Process...8 2.1.2 Cutting Parameters...11 2.1.3 Influences of Rock Mechanical Properties...17 2.1.4 Influences of the Rock Mass...23 2.2 Ratios of Cutting Force Components...24 3 State of the Art...29 3.1 Data Analysis in Rock Cutting Research...29 3.2 Rock Classification Systems...32 3.2.1 MWC – Measure-While-Cutting...32 3.2.2 MWD – Measuring-While-Drilling...34 3.2.3 Automated Profiling During Cutting...35 3.2.4 Wear Monitoring...36 3.3 Machine learning for Rock Classification...36 4 Problem Statement and Justification of Topic...38 5 Material and Methods...40 5.1 Rock Cutting Equipment...40 5.2 Software & PC...42 5.3 Samples and Rock Cutting Parameters...43 5.3.1 Sample Sites...43 5.3.2 Experiment CO – Zoned Concrete...45 5.3.3 Experiment GN – Anisotropic Rock Gneiss...47 5.3.4 Experiment GR – Uncracked and Cracked Granite...49 5.3.5 Case Study PB and FBA – Lead-Zinc and Fluorite-Barite Ores...50 5.4 Data Processing...53 5.5 Force Component Ratio Calculation...54 5.6 Procedural Selection of Features...57 5.7 Image-Based Referencing and Rock Boundary Modelling...60 5.8 Block Modelling and Gridding...61 5.9 Correlation Analysis...63 5.10 Regression Analysis of Effect...64 5.11 Machine Learning...65 5.11.2 K-Means Algorithm...66 5.11.3 Artificial Neural Networks...67 5.11.4 Distributed Random Forest...70 5.11.5 Classification Success...72 5.11.6 Boundary Layer Recognition Precision...73 5.12 Machine Learning Case Study...74 6 Results...75 6.1 CO – Zoned Concrete...75 6.1.1 Descriptive Statistics...75 6.1.2 Procedural Evaluation...76 6.1.3 Correlation of the Covariates...78 6.1.4 K-Means Cluster Analysis...79 6.2 GN – Foliated Gneiss...85 6.2.1 Cutting Forces...86 6.2.2 Regression Analysis of Effect...88 6.2.3 Details Irregular Behaviour...90 6.2.4 Interpretation of Anisotropic Behaviour...92 6.2.5 Force Component Ratios...92 6.2.6 Summary and Interpretations of Results...93 6.3 CR – Cracked Granite...94 6.3.1 Force Component Results...94 6.3.2 Spatial Analysis...97 6.3.3 Error Analysis...99 6.3.4 Summary...100 6.4 Case Study...100 6.4.1 Feature Distribution in Block Models...101 6.4.2 Distributed Random Forest...105 6.4.3 Artificial Neural Network...107 6.4.4 K-Means...110 6.4.5 Training Data Required...112 7 Discussion...114 7.1 Critical Discussion of Experimental Results...114 7.1.1 Experiment CO...114 7.1.2 Experiment GN...115 7.1.3 Experiment GR...116 7.1.4 Case Study...116 7.1.5 Additional Outcomes...117 7.2 Comparison of Machine Learning Algorithms...118 7.2.1 K-Means...118 7.2.2 Artificial Neural Networks and Distributed Random Forest...119 7.2.3 Summary...120 7.3 Considerations Towards Sensor System...121 7.3.1 Force Vectors and Data Acquisition Rate...121 7.3.2 Sensor Types...122 7.3.3 Computation Speed...123 8 Summary and Outlook...125 References...128 Annex A Fields of Application of Conical Tools...145 Annex B Supplements Cutting and Rock Parameters...149 Annex C Details Topic-Analysis Rock Cutting Publications...155 Annex D Details Patent Analysis...157 Annex E Details Rock Cutting Unit HSX-1000-50...161 Annex F Details Used Pick...162 Annex G Error Analysis Cutting Experiments...163 Annex H Details Photographic Modelling...166 Annex I Laser Offset...168 Annex J Supplements Experiment CO...169 Annex K Supplements Experiment GN...187 Annex L Supplements Experiment GR...191 Annex M Preliminary Artificial Neural Network Training...195 Annex N Supplements Case Study (CD)...201 Annex O R-Codes (CD)...203 Annex P Supplements Rock Mechanical Tests (CD)...204