Numerical Methods

Due to their high specific mechanical properties, fiber-reinforced plastic composites (FRP) offer high lightweight construction potential for use in automotive components. FRP composites with a thermoplastic matrix are characterized by short processing cycles, which makes them particularly suitable for large-scale automotive production. Semi-finished products, such as organic sheets or laminates, can be processed into shell-shaped structural components in the thermoforming process, whereby unwanted wrinkling often occurs in more complex components, which reduces the component quality. In order to counteract this deformation mechanism, it is important to adjust many process parameters to prevent wrinkling in structurally relevant component areas.

New tool technologies for thermoforming deal with the use of segmented mold inserts. The appropriate partitioning and sequencing (initial deflection of the segments in relation to each other) of the die segments can achieve an additional reduction in wrinkling. However, this opens up a new, large parameter space that cannot be mapped by purely experimental tests.

Numerical simulations provide a detailed insight into the thermoforming process and the process parameter settings. However, these models are characterized by a very high level of complexity and can take several hours to run. The use of machine learning and artificial intelligence (AI) approaches therefore offers a further step towards shortening the computing time while at the same time achieving comprehensive optimization. With the help of an FEM database, these can be enabled to efficiently solve the optimization problem of suitable segmentation and sequencing of the shape segments.

Within the scope of student work, various subtasks of the presented subject area are to be worked on. The aim is to be able to numerically map the draping behaviour of FRP in the thermoforming process and to make optimization problems regarding the setting of process parameters more time-efficient through the use of AI. The current processing status requires a more theoretical elaboration in the area of simulation and programming. In addition, experimental tasks for the validation of developed numerical models may also arise. The scope and duration of the work depends on the type of student work. The focal points of the subject area presented are listed below, from which student work can be derived in consultation:

Modelling und simulation: Further development of a parameterized simulation model for thermoforming of FRP materials in LS-Dyna. Development of thermo-mechanical models; development of multi-layer models; development of FEM database of forming geometries; basic consideration of material guidance system; basic consideration of segmentation and sequencing.

Development of AI-algorithms: Development and modeling of AI algorithms (e.g. neural networks) based on simulation data.

Validation and optimization: Validation of the simulation model using real thermoforming processes. Comparison of the virtual results with the actual results. If the simulation model is sufficient, the AI algorithms adequately represent reality.

______________________________________________________

Field of study: CSE, mechanical engineering or comparable

Start of your thesis: immediately

Contact: Jan Middelhoff

Metal forming is a pivotal manufacturing process involving the deformation of metallic materials to attain desired shapes and structures. These material as later on considered as the basic building blocks in major industries such as  automotive, aerospace, and machinery manufacturing, which make the importance of  the forming process even more evident. Not only that, but also the material can be made stronger and uselful with the help of these processes. However due to the complexity of the application, these processes can also prove to be very challenging at the same time with the conventional trial and error approaches.

Simulation plays a crucial role in the field of metal forming due to its ability to model and predict the complex behaviors of metals under various forming processes. By employing simulation tools, manufacturers can virtualize the processes before physical production begins. This not only reduces the need for costly and time-consuming prototyping but also helps in minimizing material waste. At the same time Simulation helps identify optimal conditions, refine designs, and detect potential defects, enhancing the overall reliability of metal forming processes.

Neverthess, Simulations can also be constrained with respect to time and effort with increasing degree of complexity, especially when they involve delicate interations between material, tooling and manufacturing enviroment leading to a multibody, multiphysics system overall. Simulation coupled with AI can address this issue. AI-powered simulations can rapidly analyze vast datasets and complex variables, accelerating the optimization of metal forming processes. This speeds ups the overall prozess by reducing the time spent in trial and error approaches.

Considering the huge domain of simulation and its coupling with advanced algorithms for complex manufacturing processes like forming, there is possibility to devise and strategize a variety of research themes in the form of Study projects. The scope and duration of the work depend on the nature of each student project. A study project in the field of Metal forming can be conducted in conjunction with following broadened Areas:

Coupled Multiphysics Simulations: Investigate the integration of multiple physics simulations, such as thermal and mechanical aspects, to provide a more comprehensive understanding of the metal forming process.

High-Performance Computing: Study the application of high-performance computing (HPC) and parallel processing to accelerate metal forming simulations, allowing for larger and more detailed models.

Data-Driven Approaches: Examine the integration of data-driven approaches, such as machine learning, into metal forming simulations to improve prediction accuracy and enable self-learning models.

Optimization: Focus on the development and application of optimization algorithms to improve metal forming processes, considering factors like energy efficiency, cycle time reduction, and material utilization

Digital Twin: Explore the concept of a digital twin for metal forming processes, integrating real-time data with simulation models to create a virtual representation for monitoring, analysis, and optimization.

______________________________________________________

Field of study: CSE, mechanical engineering or comparable

Start of your thesis: immediately

Contact: Syed Sarim Ali

The utility of neural networks has been widely demonstrated in a range of engineering applications. Neural networks act as excellent function approximations, even for extremely non-linear relationships between corresponding input and output data.  In standard implementation of such networks usually a large training data set is required, in order to adequately train the network to produce accurate predictions. The training data set is generated through either experiments or simulations.

In physics based neural networks, the physical equations underlying a certain process can be directly incorporated into the loss function of the network. These physical equations usually present themselves in the form of partial differential equations (PDEs) and are derived from the governing physical laws. To incorporate them into the loss function, the outputs of network are differentiated with respect to the given inputs and the partial derivatives obtained are used to construct the governing equation.

For any input, the outputs and partial derivatives are calculated based on the values of the weights and biases of the network. The loss function is then trained to minimize the error between the expected and predicted output of the PDE. Such physics based neural networks can be additionally supplemented by adding training data and their corresponding terms to the loss function. The main advantages of such physics based approaches is that the physical phenomena is encoded into the network and they can be used when a large standard training data set is unavailable.

The main focuses of the presented subject area are listed below, from which student work can be derived in consultation:

Physics-based process modeling: Derivation of differential equations from the physics underlying the physical processes. Solving the differential equations using neural networks (PINNs) to enable faster computation compared to numerical approaches such as FEM.

Development of AI algorithms: Generation of training data through simulations and development of AI algorithms based on simulation data.

Validation and optimization: Validation of the simulation model using data from real processes. Comparing virtual results with actual results, to check if the simulation model is sufficient and that the AI algorithms adequately reflect reality.

______________________________________________________

Field of study: CSE, mechanical engineering or comparable

Start of your thesis: immediately