Filters

Settings

Now showing 1 - 10 of 10
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Preprocessing for Data-Driven Modeling with Probability Density Estimation
    (2026)
    Peter, Timm J.
    In engineering, the modeling of complex systems plays a central role. Increasing computing power and storage capacities as well as the trend towards deep neural networks are resulting in more and more data being stored. This dissertation addresses two main challenges arising from handling large amounts of data for data-driven modeling: Firstly, the choice of a subset that is representative of the dataset from which it is selected. Secondly, the handling of unbalanced datasets, i.e., datasets with regimes of higher and lower point density. The first challenge is addressed by developing a novel subset selection algorithm based on kernel density estimation. The method ensures that the selected subset is representative of the original dataset or any desired arbitrary distribution. A sophisticated yet simple approach to evaluating the estimated density allows to save computing time. The second challenge is addressed by introducing a data weighting method that extends the standard loss function. The weights for the individual data points are adjusted in such a way that data points from regions of sparser point densities are weighted higher and data points from regions of higher point densities are weighted lower in order to ensure a more balanced model performance. This approach is independent of model architecture and suited for any training algorithm. The effectiveness of the developed methods is demonstrated by using benchmark datasets and real-world application examples. Among others, the examples of thermal modeling of a permanent magnet synchronous motor and a cold forming process are used. The results show that the presented method for subset selection can effectively select representative datasets and is on par with state-of-the-art approaches to subset selection. Additionally, the presented method is able to select the subset to represent arbitrary desired pdfs which gives the user much freedom of design. The introduced method for data weighting typically results in significant performance improvements for dynamic models, especially for imbalanced training datasets. Overall, these contributions provide a valuable contribution to the further development of data-driven modeling methods and offer practicable solutions for real-world challenges.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/10875
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:10
      14  45
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Gray-Box System Identification using Continuous-Time Local Model Networks
    (2025)
    Glass, Leon
    Both first-principles-based and data-driven methods provide powerful tools for modeling physical systems. As such, there is a strong push for harnessing both approaches by combining them in the form of gray-box models. Automatic differentiation of ordinary differential equation (ODE) models and physical model parts enables flexible gray-box modeling in system identification tasks. Local model networks (LMNs) with their excellent interpretability and sparing use of model parameters are an ideal model class to provide data-driven elements in a gray-box model. Unfortunately, LMNs’ intricate training procedures currently do not allow for end-to-end training within larger model architectures. To remedy this, we propose a new LMN training algorithm and corresponding model called rectified linear unit model tree (ReLUMoT). ReLUMoT provides fully gradient-based training of LMNs, which permits us to train within gray-box ODE models. In essence, ReLUMoT pioneers the use of piecewise linear neural networks (PLNNs) as a training engine for LMNs. This is accomplished by distilling the locally linear structure of a PLNN into an LMN. We thoroughly review all theoretical concepts needed to train gray-box ODE models, ranging from state space modeling to initializing latent states during the training procedure. In particular, we give a critical review of sensitivity analysis methods, discussing their suitability for system identification tasks. We provide extensive experimental validation of ReLUMoT’s usefulness in practical modeling tasks. ReLUMoT is benchmarked against a variety of state-of-the-art modeling algorithms in learning both static and dynamic processes. In black-box settings, ReLUMoT matches or exceeds the performance of existing LMNs. Properly incorporating prior knowledge is shown to lead to vast improvements in model performance and size.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/10857
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:9
      9  67
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Versuchsplanungsmethoden für die experimentelle Modellbildung bestehender mechatronischer Prozesse
    (2024)
    Voigt, Tim 
    Existing mechatronic processes have to be continuously improved in order to meet current requirements for flexibility, quality and efficiency. For this purpose, models are needed that represent the process behavior and are suitable for analysis and optimization. More and more, data-based models are used for this purpose, whose quality strongly depends on the available data. Data from regular process operation are often concentrated on a few operating points and are therefore not well suited for modeling. In this work, experimental modeling is applied, in which the training data for a data-based model are systematically obtained in experiments. For this purpose, new design of experiments methods are developed that can be practically applied with little effort. Both in the design of experiments and in the execution of experiments, efforts are reduced by incorporating existing process knowledge and by exploiting process characteristics. With incremental modelling, a stepwise approach for experimental design and modelling is presented. In each step, the experimental design and the process model are extended by an additional input variable, which leads to an intuitive modeling approach due to the increasing complexity. By incorporating the most important input variables first, the main process behavior can be modeled in a few steps. Furthermore, a new experimental design method for composite units is described. In this type of processes, the values of the input variables are specified by the components used in the units. By creating optimal combinations of given components, cost-efficient experiments are enabled, e.g. for assembly processes. The developed methods are compared to reference methods using test processes. With these novel methods, equivalent model quality can be achieved even though greater practical applicability is given. The practical applicability is demonstrated by the means of three different real-world applications.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/10482
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:8
      445  225
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Machine learning with nonlinear state space models
    (2022)
    Schüssler, Max
    Contemporary automation systems require accurate models for analysis, design, and control. Oftentimes, it is not possible to derive satisfactory models by first principles. This is often the case in process engineering or in mechatronic systems, where the overall process is just too complex, or the development of those first-principle models would just be too expensive. Thus, there is a strong demand for data-driven modeling approaches. Additionally, there are enormous successes in machine learning concerning the generation of models solely by the use of data. Driven by the need for accurate models as well as the successes and advances in machine learning, a novel class of model structures and associated training algorithms for building data-driven nonlinear dynamic models is developed. The new identification procedure and the resulting model is called local model state space network (LMSSN). It fuses nonlinear state space models with local model networks (LMNs). The LMSSN is designed as a user-friendly and deterministic approach with warm-start initialization (with a linear model), incrementally growing complexity, favorable and robust extrapolation behavior, easy interpretation, and straightforward incorporation possibilities for prior knowledge. Furthermore, recurrent neural networks (RNNs) and their similarities to nonlinear state space models are elaborated on. Popular RNNs like the long short-term memory (LSTM) and gated recurrent unit (GRU) models are translated into controls perspective and a comprehensive study comparing different RNN structures is carried out. The overall outstanding performance of the LMSSN is demonstrated on various applications. It is shown that the LMSSN can accurately model a wide variety of processes by consistently computing expressive yet compact models. The LMSSN is benchmarked against state-of-the-art nonlinear system identification algorithms and achieves similar or superior results. The practical usefulness and applicability of LMSSN are strikingly demonstrated on two real-world processes.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/10164
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:7
      1312  1814
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Analyse gesamtsystemischer Zusammenhänge von hybriden Antriebsstrangkonzepten bezüglich kundenrelevanter Anforderungen
    (2020)
    Mayer, André
    System-level design of hybrid electric vehicles focuses on the analysis and optimization of hybrid powertrains within a high-dimensional solution space with regard to the variety of variants and requirements. Existing methodical approaches are based on the identification of optimal system designs with regard to predefined powertrain requirements. In this thesis, the possibility of substituting certain engine-power classes of the product portfolio with hybrid concepts whose system performance is largely defined by the electric drive system is investigated. A significant reduction of the combustion engine power within the overall system compared to conventional drives requires a verification of the original powertrain requirements in the context of customer relevance. In contrast to previous investigations, the method presented in this thesis identifies for the first time the demands in power and energy sizing of the electric drive system and the electric energy storage, which result from the fulfillment of original powertrain requirements. The additional influence on driving comfort spans a region of conflict between powertrain efficiency, performance and driving comfort, in which the relevant overall systemic relationships for decision-making are analysed. For this purpose, systemic influencing factors such as a reduction of the maximum velocity requirement are identified and their influence on a customer-relevant powertrain design is evaluated. In the course of this work, visualization techniques from the field of machine learning provide a comprehensible and transparent basis for substantiated decision-making. Furthermore, well-known optimization-based operating strategy approaches are extended by methods of fuzzy logic with regard to the solution of the problem. The latter is mainly enabled by considering a timeframe-based analysis for the complete characterization of the performance requirements of the electrical components.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/9974
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:6
      654  503
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Machine learning with finite impulse response models
    (2021)
    Münker, Tobias
    An accurate and reliable model is the foundation for analysis, design, and control of a modern automation system. It is often too complicated, too expensive, or too inaccurate to develop these models based on first principles. In recent years, machine learning has made tremendous progress through the usage of data to generate models. Nevertheless, a key drawback is that for the identified models there is no stability guarantee. Thus, within this thesis, methods for identifying both linear and nonlinear systems based on finite impulse response (FIR) models are investigated. These avoid feedback and thus ensure stability. Linear FIR models offer a compelling advantage due to the interpretability of the impulse response and inherent stability. Recently, novel methods for regularization based on a specifically designed Tikhonov regularization have been proposed. In this contribution these approaches are extended to allow for a better incorporation of existing prior knowledge. The developed method can be employed for order selection and gray-box identification. Its feasibility is demonstrated on numerical benchmarks and on a laboratory example. Local model networks allow for the extension of linear identification methods to nonlinear systems. Here, the regularization mechanism applied to linear systems is employed to identify local FIR models. Especially for identification problems with a low signal to noise ratio, the method performs significantly better than local model networks with feedback of the output. Finally, machine learning with convolutional neural networks is investigated. The relation between these and block-oriented nonlinear systems is analyzed. Then, a specific structure of a deep neural network containing FIR models as building blocks is proposed. This structure is equipped with a regularization scheme adopted from linear FIR models for the impulse responses comprised in the neural network. It is shown that the bias-variance trade-off is influenced positively on a benchmark example and achieves state-of-the-art results while additionally the internal impulse responses are smoothed and the stability of the system is guaranteed.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/10041
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:5
      711  1523
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Safe and dynamic design of experiments
    (2019)
    Schillinger, Mark
    In the calibration of modern combustion engines, stationary model-based approaches are common. For future developments, the use of dynamic data-based modeling is an important and required step to further increase the efficiency and quality of the calibration process. In order to achieve high modeling quality, these models require informative measurement data. This necessitates a transient variation of the input signals during the measurement process. At the same time the safety of the system under test needs to be ensured during the measurements by supervising critical output signals. This thesis covers different methods to tackle this task, namely for stationary and dynamic design of experiments. Therefore, two strategies are presented and analyzed: on the one hand the use of an offline generated design of experiment, on the other hand the use of active learning. The first strategy is threefold. It starts with the identification of the stationary safe system boundary. Thereafter a dynamic design of experiment is created within the stationary safe boundary. Finally, a dynamic measurement is pursued, while the system under test is safeguarded by a newly introduced supervising controller. The second strategy uses a dynamic safe active learning approach. Thereby, the dynamic model is not anymore learned offline, after the whole measurement was completed, but online, in parallel to the measurement process. Instead of a predefined design of experiment, the queried trajectories are optimized iteratively based on the learned model. The optimization also considers dynamic safety constraints of the system under test. Special emphasis is put on the application of these strategies to real-world measurements in the combustion engine domain. The various design of experiments, measurement, and active learning algorithms are compared and their specific advantages and disadvantages are discussed.
    Source Type:Doctoral Thesis
    DOI:10.25819/ubsi/36
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:4
      1203  1109
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Fighting the curse of dimensionality with local model networks
    (2018)
    Belz, Julian
    This thesis is settled in the field of data-based modeling (identification) and specifically focuses on the weakening of the effects of the curse of dimensionality with local model networks (LMNs). The methods for fighting the curse of dimensionality originate from the fields of input selection and design of experiments (DoE). The model type of LMNs allows the distinction in two input spaces - the rule premises input space and the rule consequents input space. The developed input selection techniques exploit this unique property of LMNs and the possibility to assign potential inputs to each input space individually. It is shown that this additional freedom enables input selection methods using LMNs to find a better bias/variance trade-off. Furthermore, one of the two arising input spaces is directly connected to the nonlinear effects of the model, which allows for more detailed interpretations. The DoE contributions of this thesis concern passive and active learning schemes. Recommendations for passive experimental designs are given based on extensive simulation studies with the help of a function generator. The following three questions are addressed. Should corners (vertices) be measured? Which space-filling experimental design is likely to yield the best model accuracy? What order of experimentation leads to the best model quality in early stages of the measurement process? Eventually, the contribution regarding active learning is an extension of the already existing hierarchical local model tree for design of experiments (Hilomot-DoE), addressing three important issues of modeling simultaneously: (I) optimality, (II) model bias, and (III) model variance. All developed methods demonstrate their abilities on selected application examples, including the prediction of the fuel consumption of cars, the data-based modeling of the air-mass flow of combustion engines, the metamodeling of fans, and the generation of a dynamic model of a heating, ventilating, and air conditioning system for control design.
    Source Type:Doctoral Thesis
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:3
      1294  746
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Lokale Modellnetze zur Identifikation und Versuchsplanung nichtlinearer Systeme
    (2013)
    Hartmann, Benjamin
    This thesis proposes new approaches for experimental, data-based modeling (identification) and for experimental design of nonlinear systems based on local model networks. On the one hand, the proposed algorithms are based on a detailed analysis and further development of the linear optimization for locally weighted, polynomial models. This concerns parameter estimation, regularization, validation and subset selection. On the other hand, nonlinear optimization methods are required in order to adapt the validity regions of the local models to the underlying nonlinear process. This is achieved by the development of heuristic partitioning strategies that incrementally subdivide the input space. These methods enable an efficient adaptation of local model networks to the given data. The basic idea of the investigated approaches is the application of flexible axes-oblique partitioning strategies that are well-suited for higher-dimensional problems. In addition, a strategy is proposed that applies a structure trade-off where the model is improved either by performing an axes-oblique split or by increasing the number of regressors to be selected for the local sub-models. The partitioning as well as the local selection of the significant polynomial terms are carried out simultaneously. Furthermore, the newly developed "hierarchical local model tree" (HILOMOT) algorithm is extended for the application as an experimental design approach. Next to an offline strategy for design of experiments (DoE) the algorithm is enhanced with active learning strategies in order to perform online measurements of processes where no prior knowledge is available. The active learning algorithm is able to interact with the process and can meet the requirements to efficiently place the measurement points. The proposed algorithms are verified on real system applications where their performance has been demonstrated. That incorporates the modeling of the fuel consumption and emissions of a modern diesel engine as well as the active learning of damage locations in the framework of a structural health monitoring system. Moreover, an adaptive online experimental design is implemented on an engine test bed for the calibration of a diesel engine.
    Source Type:Doctoral Thesis
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:2
      1333  642
  • Some of the metrics are blocked by your 
    consent settings
    PublicationAccess status: Open Access ,
    Entwicklung eines haptisch unterstützten Bedienkonzepts für Hydraulikbagger
    (2012)
    Hayn, Henning
    In mobile hydraulic machines such as excavators, haptic human-machine interfaces are currently not in use. Today, the machines are operated by mechanical-hydraulic joysticks. Each joystick axis controls a single hydraulic actuator. This leads to operational concepts which are not very easy to use. Due to the fact that electro-hydraulic systems with electronic joysticks are available for serial applications, alternatively, more intuitive operational concepts have become feasible. These concepts can be enhanced further by adding haptic driver-assistance systems. The aims of haptically enhanced operational concepts are to increase the efficiency of the human-machine interaction by providing driver-assistance systems, reducing operating errors, and shortening the time required for inexperienced drivers to learn how to operate the machine. In this thesis, a haptically enhanced master-slave operational concept for a hydraulic excavator is proposed. The concept includes a conceptual design for intuitive, active operating-elements as an alternative to conventional joysticks. It also includes bilateral control-architecture for the complete system, consisting of the hydraulic excavator and the active operating-device. The device resembles the kinematic chain of the excavator’s manipulator. This allows for intuitive operation of the machine. The operating device is actuated to implement position control of the device. This is necessary to permanently synchronize the operating device and the position of the hydraulic manipulator. The main idea entails position controllers for both the excavator and the operating device where both control loops output the reference position to one another. More specifically, two position-controlled systems are placed in a feedback loop with each other. The human operator acts as an input disturbance to the operating device. This approach allows stable teleoperation of the machine and gives the operator haptic feedback of the manipulator’s inertia. The Internal Model Control (IMC) methodology is employed to design the position controllers for both the hydraulic manipulator and the operating device. Therefore, a simplified design rule for the IMC filter and a novel anti-windup approach for models with pure integrators are proposed. A second aspect is the utilization of the active operating-device’s actuators to implement haptically enhanced driver-assistance systems. These systems support the driver as he performs his working tasks (e. g. leveling assistance, slope cutting, and workspace limitation) through tactile forces fed back by the operation element. In order to verify the operational concept and the haptic assistance systems, they were exemplified on a virtual-reality excavator simulator and on an experimental excavator equipped with electronic pilot-valves, rapid-control prototyping hardware, and sensor systems. To test the concept on the prototype, commercially available devices, namely a SensAble Phantom Omni and a 3Dconnexion SpaceBall 5000, were used. Measurement results are given. Experiments with test drivers are performed to evaluate the proposed operational concept and the haptically enhanced driver assistance systems. The experiments confirm the usability of the operational concept and the benefit of the haptic assistance systems.
    Source Type:Doctoral Thesis
    Series:Schriftenreihe der Arbeitsgruppe Mess- und Regelungstechnik - Mechatronik, Department Maschinenbau
    Volume number:1
      1855  1249