Duration
Start: 01-03-2021
End: 28-02-2025
Staff
- Aneesh Ravikumar, MSc.
- dr.ir. Javad Hazrati Marangalou
- dr. ir. Wouter Hakvoort
- prof.dr.ir. Ton van den Boogaard
Description
Manufacturing industries, nowadays, focus on building reliable, flexible, agile and robust control systems to steer towards zero defect manufacturing. Multi-stage metal forming processes are popularly used for high volume production in the manufacturing industries. Over the years, predictive models of these processes have gained significant importance to replace intensive physical trials in the real world. These predictive models are developed through several real time measurements to serve as eligible digital counterparts of the manufacturing processes. In the physical world, there are several sources of uncertainties that cause variation in the geometry, appearance and physical properties of the product. Accounting for these variations is important to estimate the accuracy and robustness of these virtual models. The virtual models are high fidelity (HF) finite element (FE) models subjected to approximations known as discretization. With an appropriate choice of model structure and features, these models can estimate very accurate results in comparison with the sensor measurements, but they are computationally expensive. High dimensional design of experiments have to be solved to gain knowledge on the relative importance of the input parameters subjected to noise, fluctuations or variations. The ongoing battle of accuracy and efficiency has seen resolution through the development of surrogate models or metamodels. Surrogate models are developed through model order reduction based on Singular value Decomposition (SVD) and interpolation techniques (like using Kriging and Radial Basis Functions) applied on the HF models making them extremely efficient. The surrogate models are highly adaptive that evolve to become more robust. These models are purely based on mathematical assumptions of data distribution (of input parameters and system response) and stochastic approaches. But due to the lack of synergy between the existing data driven approaches and the HF models to apply or link the underlying physics, the estimation of these models are still being questioned. Thus, this research project aims at building methodologies to combine model based Engineering with physically inspired data driven approaches to predict the product attributes and estimate the occurrence of failure in the manufacturing process.
The research project is a part of the Integration of Data-drIven and model-based enGIneering in fuTure industriAL Technology With value chaIn optimizatioN (DIGITAL TWIN) which is a consortium of six major Dutch Universities and some of the leading manufacturing industries in the country.