Cracks in concrete and rusting reinforcing steel can pose a critical threat to the stability of buildings. Therefore, reliable detection of these damages as part of maintenance plays an important role in ensuring the safety of civil infrastructure. Conventional image processing-based methods are set to one damage type and lack the adaption to new environments in real world-situations. By using images from mobile devices this thesis presents a comparison of three convolutional neural network architectures to locate and classify the mentioned damage types. Two datasets were created for this task and varied by applying Data Augmentation. For classifying images with a 227×227 pixel resolution a 98.03% f1-score was achieved. In a second step, the networks were transformed to fully convolutional networks and applied to images of different aspect ratios. Thereby, the proposed method has successfully achieved a f1-score of 96.28% in classifying cracks and corrosion. The segmentation of the images on a grid level can further be used for the digitalization of the damages.
On construction sites workers are exposed to several dangers. Especially cranes are considered dangerous, because they can potentially cause accidents by being struck by crane load. In the worst case they cause fatal injuries. It is necessary for crane operators to know the location of nearby coworkers. By improving communication among workers and crane operators it is possible to reduce the risks of accidents. An automated system specialized in warning and signalizing potential dangers could prevent further casualties.
By using cameras that can be attached to cranes this thesis presents a method for tracking workers. Considering the implementation of methods in the fields of computer vision and machine learning the tracking system is established. The tracking process is structured in three phases. To isolate objects and persons the first phase separates for- and background using a background subtraction. Using the soft cascade method trained to detect worker the second phase classifies isolated regions in the foreground. Finally, using the kalman-filter workers will be tracked.
Results are evaluated by analyzing roc-curves to detect a nearly optimal classifier. Manually marked images are used to compare against the detection and valued. Those results can then further be used to evaluate the efficiency of the tracking system.
Augmented Reality (AR) is a novel technology, which is mainly in the state of research and rarely used in practice, especially in construction. In order to point out the capability of deployment and the potential fields of AR technology in the construction industry, this master thesis focuses on the technical experience of construction engineers. Different points of views and improvement suggestions are provided through interviews with those experts, who have worked on various construction projects. The challenges from construction projects and suggestions of deployment with AR are pointed out. Based on the evaluation of the experts’ suggestions, an application prototype with AR is developed to solve a practical problem and provides an example of the usage of BIM data with AR. The test of the prototype is carried out by the interview partners. The judgment of the prototyp as well as the usage of AR in the construction practice is worked out from the feedback of the test persons. Finally, the possibilities and challenges of the usage of AR technology in the surroundings of construction are discussed.
The ability of indoor pose estimation is an essential key component for a wide range of applications in the construction industry, such as progress monitoring of finished works or augmented reality-based maintenance and navigation of indoor facilities. There are numerous indoor localization methods based on different technologies, with an image-based approach via a mobile device being the most economical. This work aims to investigate an existing image-based approach using convolutional neural networks trained on synthetic images to estimate the pose of real images. Four datasets of indoor scenes from two buildings with various trajectory properties were created and independently analyzed. The average position accuracy was 11m, while an accuracy about 1m could be achieved by training and testing the artificial neural networks with a dataset from the same domain. Moreover, the networks using cross-domain data localized all test images in a confined area and estimated the orientation mostly as the same one.The ability of indoor pose estimation is an essential key component for a wide range of applications in the construction industry, such as progress monitoring of finished works or augmented reality-based maintenance and navigation of indoor facilities. There are numerous indoor localization methods based on different technologies, with an image-based approach via a mobile device being the most economical. This work aims to investigate an existing image-based approach using convolutional neural networks trained on synthetic images to estimate the pose of real images. Four datasets of indoor scenes from two buildings with various trajectory properties were created and independently analyzed. The average position accuracy was 11m, while an accuracy about 1m could be achieved by training and testing the artificial neural networks with a dataset from the same domain. Moreover, the networks using cross-domain data localized all test images in a confined area and estimated the orientation mostly as the same one.
The Bosch Rexroth company in Dortmund has various test benches to test different types of gears used in agricultural machinery. The test benches are managed by different employees, so that an order must be created for the use. To get an overview of the orders, test stands and gearboxes used, an application that contains a suitable user interface and a data storage system must be developed. This bachelor thesis describes the conception and implementation of such an application, which we call the Data Aggregation Tool (DAT). Important aspects of planning in combination with agile project management, mockups and prototypes, as well as the implementation of different components, are discussed using the Framework Angular. Requirements for the application are defined with use cases and user stories. Different orders are divided into different categories, with one being considered complex enough for all other orders to build on. To test the implementation of the application, a limited evaluation is performed and evaluated.
Up to the present, only analogous applications for construction permit (ACPs) are accepted in Germany. Yet the submission of a digital ACP provides several advantages, especially when making use of existing federal standards in the areas of construction and planning. Further, the direct integration of Building Information Modeling (BIM) into a digital ACP process provide further potential. Within the scope of the ongoing research project "BIM-basierter Bauantrag" (BIM-based ACP), which aims to integrate BIM into the digital ACP process, one objective is the development of a software prototype suitable for realizing a BIM-based ACP process. An existing Desktop-based prototype already provides basic functionality. A web-based solution would offer further advantages, such as platform independency and centralized data environments. This work examines the migration of the existing Desktop application into a web-based environment. In order to be able to perform such a migration, this work provides an extensive requirements analysis. The migration's primary objective is to carry over as many software components into the new system as possible to minimize code changes. In order to achieve that goal, an appropriate strategy has been designed to develop a web-based client which reuses parts of the existing software as server-components. The resulting web-based software system has been evaluated using a real-life sample project.
Project management and construction informatics had begun to blend and guide the construction industry internationally the last decade. Technological advancement gave naturally birth to the innovative Building Information Modelling method on the construction scene. With the introduction of a BIM Execution Plan, a company develops an agenda that describes all the steps of generating a project, clarifying the project goals and its coordination throughout the whole life-cycle of the structure. The expansion of the thesis is twofold. Namely, in the first part describes in detail the pure theoretical background of BIM Execution Plan and its adaptation on the company by a development of a Master BEP Format and a second task of a computational implementation method to automatize the quality control of buildings. The main aim of this scientific work within a Consulting Design in Construction Management business is to deliver the knowledge and enlighten the co-workers for the applied management tools of BIM that the market offers nowadays, as well as to advance the work-flow on BIM Uses that constitute a BIM Execution Plan’s input. The master thesis has been achieved by an elaborate literature and regulations review, whereas a continual use of the company’s provided software was assigned in order to realize the shortcomings on a genuine project. Moreover, an intense practice in Integrated Development Environments and programming languages was attained and ultimately a quality control tool for Revit Autodesk was implemented.
As self-driving technologies are implemented in autonomous vehicles an increase in the reliability on automotive electronics hardware is increasingly becoming a requirement. Current methods have statistical significance; this mean that they remain population-relevant and are prone to outliers. A new way to increase the reliability of automotive electronics systems and getting it closer to a 100% is to implement PHM. This would imply the presence of degradation sensors included in the silicon of the packages, enabling a more direct measure of quantities related to the degradation of the system. This thesis focuses on the development of three Machine Learning models for the prediction of degradation of electronic components. A data set is created from ETVs subject to accelerated ageing. This data is transformed to a compressed understandable form with various preprocessing techniques and then fed to the ML models. After tuning, the best model is picked and applied to new datasets. Expert knowledge from Bosch is used to validate the results at every stage. After having proved that the concept is valid, new models are developed using only a small dataset of ANSYS thermo-mechanical simulations data performed by Bosch and provided to the author and then used for prediction of real-world experimental stresses.
Support Vector Machines (SVMs) are popular machine learning methods, especially for binary classification of data. Computational complexity of kernelized (Non-linear) SVMs becomes a limiting factor when dealing with large-scale machine learning problems. This thesis presents an online adaptive version of Dual coordinate ascent that honours a budget constraint and restricts the number of support vectors used to represent the model. This new algorithm is coined as Adaptive coordinate frequency budgeted
stochastic coordinate descent (ACF-BDCA). The Budget methods have proven to be effective for reducing the training time of kernel SVM while retaining high accuracy. In addition to that, instead of fixing selection frequencies of a general Coordinate ascent with uniform random selection of coordinates, the Adaptive coordinate frequencies (ACF) method removes the need to estimate optimal coordinate frequencies beforehand, and it automatically reacts to changing requirements during an optimization run. The ACF-BDCA algorithm has demonstrated the ability to significantly speed-up the BDCA method. Also, a new robust stopping criterion was successfully designed to halt this non-convex optimization problem, while achieving good performance levels.
The integration of mechanical and electrical building equipment may cause issues and leads unquestionably to an increased complexity of the entire building process. Especially the planning of openings for technical equipment is a very complex procedure. The usage of classic methods for the coordination of openings between all the involved parties of a construction process causes inconsistencies and intransparency, as simultaneous work is not possible. Building Information Modeling (BIM) aims at providing appropriate means to tackle the challenges, thereby improving the overall transparency and quality. Coordination and communication during the planning of openings is possible because suitable BIM-based tools are already available. Despite its many benefits, however, the processes are too complex and deficient, especially when different data formats of opening-models have to be reconciled. This master thesis develops suitable processes for a significant improvement of the model-based planning and coordination of openings. Hereby the loss of information through different data formats will be compensated. By using the open BIM Collaboration Format (BCF), the transparency and traceability of coordination and communication processes is ensured.
As a result of the mechatronization and the exponential growth in the number of product variants, the calibration effort of heaters is expanding intensely and demands new, more efficient approaches other than experimental laboratory testing. Within the scope of the present master thesis, an application tool was developed, the tool which is modular in design and was implemented based on the UI/UX concept so that the user can operate it intuitively, can be systematically used by a device development engineer to read, visualize and process measured data in order to automatically calculate functional parameters of a heating device.
In the recent years a lot of new models have been developed in machine learning. One of such new model classes is Deep Recurrent Gaussian Process. In this thesis we de scribe the Deep Recurrent Gaussian Process with Variational Sparse Approximation, in particular the one presented by R. Föll, and make an attempt to provide a time and memory efficient implementation for it, in order to ease its practical application. We implement this model in two different packages, TensorFlow and PyTorch, supply as a result three implementation versions and indicate some issues that have arisen in the meanwhile. We further evaluate performance of our code and compare all three implementations in terms of system memory requirements and running time, estimating the most balanced one to be about 4.2 times faster and with no increase of system memory consumption, compared to initial implementation presented by the model’s author. Also, a brief estimation of model’s accuracy based on comparison with other Gaussian Process based models is made.
There are several service building components (SBC) like fire extinguishers and emergency exit sings in public buildings. Since they usually possess a prominent appearance and their locations are known, they can be used for indoor navigation. This thesis deals with a color-based preprocessing of images containing SBCs for a subsequent detection. The goal is to find relevant image patches, that may contain SBCs and, this way, reducing the search space within the image that needs to be examined for further object recognition. To achieve this, a dataset has been created which consists of fire extinguishers, fire alarm boxes, smoke funnel boxes and emergency exits. Since these SBCs possess salient colors, an object segmentation by color is reasonable to facilitate detection. For segmentation, each component's color range is determined by averaging over several samples. Segmentation is, then, performed in different color spaces in order to find the optimal color space by comparing the results. The HSV color-space demonstrated acceptable results across all objects but emergency exits. The LAB color-space shows the best results for emergency exits, however requires a vastly bigger runtime for the segmentation.
As a result of digitization, an increasing number of small and medium sized companies are forced to design and optimize internal operating processes. This is being accomplished with the help of today's intelligent technologists according to the vision of Industry 4.0. The present scientific work analyzes the state of digitalization in the craft industry. In this context, PDS, a craft specialized ERP System, is evaluated in Elomech Elektroanlagen GmbH as a practical example of how the software links internal processes with each other and thus optimizes the processes in the company. The focus of analysis is on the mobile applications of PDS Service App and PDS Zeit App, which enable the mobile administration of service orders and the digital recording of hours. Despite the gain of efficiency, the analysis shows that there are still existing challenges and risks for the company. In particular, issues regarding the privacy aspect of fitters needs to be a critical consideration.
Real-time performance of fuel cell plant model developed in Simscape is evaluated and tested in this thesis. Various simulation models for fuel cell system are developed by different departments at Robert Bosch GmbH in various simulation platforms for e.g. Simscape, GT-Power. A matrix is developed for comparison of these simulation models based on different criteria like model type (0-D or 1-D), accuracy, validation, parameterization, parameter identification and modeling physics (physics based, data based and hybrid). Besides exploring the configurations of model like solver type, step size and usage of local solvers, built in performance packages of Simulink (e.g. Simulink profile report, Performance Advisor) are also used to further check the computationally intensive parts of the model in Simulink environment. It is observed that the execution time can be optimized by selecting a local solver and fixed-cost runtime consistency iterations on Simscape model. Later on different experiments were performed on Simscape fuel cell model on ETAS Labcar RTPC in Hardware-in-the-Loop environment to test the real time performance.
This thesis deals with the application of the Deep-Q-Learning method for a collision-free drone flight. Building on Reinforcement Learning, the aim is to use a drone as an "agent" by interacting with the environment using rewards or punishments. The goal is to teach a behavior to enable a collision-free movement through 3D space. To explore this, the realistic simulation environment AirSim is used where the real world is adequately represented. The drone can be controlled via a Python programming interface to interact with the environment according to physical laws. Depth images of a virtual camera attached to the front of the drone map the environment and are passed as input to an artificial neural network.
Traditional working practices of quantity and cost determination, which are still common in the construction industry, are characterized by a susceptibility to errors and time intensity due to their manual approach. Concerning this matter, the working methodology Building Information Modeling (BIM) offers a process optimization, which promises the avoidance of laborious and error-prone work by a consistent further use of digital information, resulting in an increase in quality and productivity. In the construction industry, engineering structures e.g. bridge and tunnel construction have different requirements compared to classical building structures, which ultimately also affect the implementation of BIM. Within the scope of this work, it was possible to develop a concept for an optimized BIM-based cost calculation in this field. As a result, workflow processes were used to describe the necessary work steps, roles and responsibilities as well as exchange formats for three application-specific scenarios relating to the building model and the cost structure. Finally, the feasibility of the concept was prototypically demonstrated using selected software products.
Transactions help to create reliable software and bring consistency and integrity to the system. Each transaction has to be completed successfully or aborted in the case of failure. Distributed transactions have the same properties of transactions and are a set of two or more transaction branches, which can span multiple data sources. The Vendo project is the process of booking Deutsche Bahn train tickets online per website or per “DB Navigator”, or via ticket vending machine. The project consists of various RESTful microservices. For each request of the user, a transaction will be generated. Each transaction has a transactionId and will be continued until the end. Implementation of transactions in such a large project can be tricky and presents some challenges. The goal of the thesis is to identify these challenges of implementing the transactions in Vendo project, find appropriate solutions for them, and monitor the transaction at the end to ensure that they work.
This work examines the various aspects in the development of a software application based on the architectural model of microservices und the expected benefits. Particular attention should be paid to compliance with data consistency in relational databases, through which transaction security must be addressed. For revision, a suitable prototype has been implemented. The architectural pattern "microservices" requires an unusual amount of communication over an insecure network (the Internet). Therefore, the risk of a network failure must be considered.
The focus of this thesis is the development of a LOD concept for 4D and 5D analysis using BIM for bridge structures. Such a LOD concept determines the geometric and semantic structure, where the levels of detail are to be defined and specified by specific parameters. Relevant requirements flow into the development of the LOD concept as well as the rules for scheduling and cost planning. Subsequently, a prototype implementation of the LOD concept using a workflow is presented.
The roads and streets of Germany extend over a length of approximately 230,000 km must be continuously monitored and maintained. Both the age of many roads, dating back to the 1970s and 1980s, as well as the high traffic load of today force the road authorities to plan conservation measures using digital tools. In particular, the easy exchange of data required for road maintenance from the start of a repair project to the final phase between all involved is an important aspect. The benefits anticipated from the use of digital planning methods in road maintenance lead to increased efforts to further development of digital computer-aided methods.
Due to the technological advances of computer-aided work processes, the method Building nformation Modeling (BIM) as a digital planning method based on 3D-models by now basically determines the developments in the construction industry. Since at least the publication of the graduated scheme „Digitales Planen und Bauen“ by the Federal Ministry of Transport and Digital Infrastructure, and therefore the requirement of a solid BIM-level for new infrastructure construction projects after 2020, standards and guidelines for the implementation of the BIM-methodology are urgently needed in bridge construction. The BIM-applications in different project phases differ both in terms of their requirements for model detailing and in the implementation of subject-specific analyzes. This is where the concept of the Level of Development (LoD) comes into play, which defines the level of detail of the digital building model in different requirement stages and ensures sufficient nformation content for the specific analyzes.
Road maintenance and restoration have undoubtedly the largest need of funding within the road infrastructure. Therefore, the careful planning of maintenance measures for roadways has extraordinary relevance. In particular, BIM systems have gained significant importance in engineering disciplines because they allow the linking of data and their dependencies in a data model. On the one hand, a high volume of data can be processed while, on the other hand, information can be specified in great detail. In maintenance planning, the OKSTRA standard is currently being used. As a result, a BIM-suitable data model for road construction based on the OKSTRA is discussed in this thesis.
This thesis presents an efficient best-effort approach to improve the simulation performance of a given virtual prototype of Electric Steering Column Lock Assembly (ESCLA) to make it real-time capable. This prototype is modeled and simulated in SimulationX software. While reducing the model, balance between accuracy and performance is maintained. Since it involves mainly mechanical and electrical components, special importance is given to capturing the associated physical events as far as possible. Subsequent use of such kind of reduced simulations is for the purpose of Model in Loop (MiL) testing of the components.
The need for Building Information Modeling (BIM) in the field of infrastructure engineering has never been greater. BIM is slowly but surely spreading worldwide and can be the game changer in the AEC (Architecture, Engineering & Construction) industry. The federal ministry of transport and digital infrastructure in Germany wanted to ensure, that digital planning and building will be established for the German AEC industry as well. Therefore, they introduced a roadmap for Digital Planning and Construction (Digitales Planen und Bauen) back in 2015, that aims at the standardization of Building Information Modeling within the infrastructure until 2020. Currently, research and experience from real projects are showing, that unified guidelines and standards should be defined for a successful use of the BIM method. For example, a definition of a geometric and semantic information depth of a digital model is for the building section with terms like Level of Development (LOD) internationally established. During this study, a concept for the description of Level of Development of tunnel structures with mechanical propulsion is developed.
Numerical simulation for fluids, commonly called Computational Fluid Dynamics(CFD), has become standard tool for a wide range of industries. With the expansion of computing power demands for the physical accuracy, spacial and temporal boundaries of fluid simulations have increased. The common bottlenecks for numerical modeling fluids, include: large run times, computational power and expertise required. One novel solution that has been proposed to reduce the large run time is through the generation of black-box CFD surrogate models. This paper focuses on exploring the concept using Machine Learning methods to generate a CFD surrogate model to replicate entire simulation results. The novel concept of CFD surrogates was first reviewed by looking at the prior research. From an in depth review common strategies were determined and evaluated through experimentation.
The rapid growing interest in working conditions with Building Information Modeling (BIM) raises the question, which processes can be realized with the available BIM-tools. To answer this question, this bachelor thesis will observe a geometric presentation of a tunnel which was created by a shield driving, to make a belated valuation of the installation quality. The geometric presentation is based on a modeled ring which will be placed on a three-dimensional track by using a parametric model. The modelling and the parametric model will be done by Revit and Dynamo, two programs from the company Autodesk. With the help of the ring modelling and the implementation with the model, a qualitative assessment methodology is put up to judge the situation exactness of tunnel driving afterwards. The feasibility and suitability of the methodology is checked with the help of a real use case.
In order to improve existing 3D models of urban environments the structure of facade elements and windows in particular is of high interest. Therefore, the problem of automated extraction of window locations from facade images is to be solved using methods of computer vision. In this approach a cascaded classifier is implemented and trained with various window examples. This classifier is used to detect windows in any presented image without requiring user intervention. The evaluation of it proved the trained classifier functioning properly. Further optimizations to allow for a more precise training, reconsiderations concerning the training set as well as possible postprocessing steps are inevitable for a practical detector based on this approach.
Microfluidic flows are prevalent everywhere in modern technologies. From lab-on-a-chip devices, to inkjet printing, to system biology, and in wetting and de-wetting phenomena for coating technologies, their application is ubiquitous. In this study we are interested in developing a novel framework with reduced dimensionality for open microfluidic problems. More specifically, we are interested in studying thin film flows using depth averaged viscous shallow-water equations with capillary forces and fluid-wall interactions included. We adopt the mesoscopic lattice Boltzmann method (LBM) to solve for the shallow water system.
The introduction of digital planning methods is a constant, fast-changing process.For engineers, it is important to always be up-to-date to meet the requirements of the clients.In the course of this master thesis, existing processes in the planning phase and in the engineering office itself are to be identified which can be optimized by the use of BIM.Through a fundamental research on the current dissemination status of BIM, nationally and internationally, as well as through the comparison of conventional planning and BIM planning, the foundation for an implementation strategy is to be laid.A specific example is carried out on the basis of a completed infrastructure project done at an engineering office.
Recently, advances in the field of the so-called wire robot technology offer new prospects for the automation of large-scale processes. This thesis presents a simulation-based approach to analyze the technical and economic feasibility of wire robots for automated construction in future investigations. Masonry buildings are considered as an appropriate application case due to repetitive construction procedures and high demands concerning quality as well as accuracy of construction. A simulation model representing the fundamental mechanics of a wire robot is created including process-relevant objects and interactions. Special focus lies on creating collision-free motion profiles which can be exported to the robot control system and may serve as a basis for path planning. Building information models provided in the IFC format are used to set-up the simulation model and to prepare the required input data.
The aim of this thesis is to investigate the way to create a digital 3D-Subsoil-Model by using borehole data. Further, the usability if the used methods and possible benefits to the building industry should be analyzed. In the beginning, the thesis focuses on building information modeling and borehole technology to show the possible ways of generating 3D-Model by using borehole data. Following, the thesis examines the used add-ons for AutoCAD Civil-3D to create and to use a 3D-Subsoil-Model. Concluding, the model quality as well as the modeling efficiency is analyzed and evaluated.
This master thesis illustrates the importance of the Building Information Model for the German construction industry. The analysis of the conventional approach and the BIM based approach are compared to show the advantages and disadvantages of both methods. Using the BIM based software iTWO, a prototypical implementation is created to calculate quantity determination according to given construction guidelines. Furthermore, a strategy on the use of BIM is proposed for production planning and execution.