In the early phases of the product life cycle, developers are increasingly turning to topology optimization methods to achieve an ideal material distribution. This approach, whose enormous potential has so far been only partially exploited, is associated with high computational effort. At IKAT, a novel AI-based computational approach has therefore been developed to specifically address this bottleneck: the computationally intensive optimization process is shifted into the training phase of the neural network, enabling the generation of geometries in practical applications almost in real time.
The approach presented here combines a „predictor,“ which proposes geometries based on boundary conditions, with an „evaluator.“ The evaluator assesses these geometries according to predefined criteria and iteratively optimizes the network. Of particular interest is the application to compliant mechanisms, which, due to their monolithic structures, exhibit exceptional durability compared to classical assemblies. Since their movements rely on the elasticity of the material rather than the relative motion of separate bodies, their development poses a special challenge.
The presented project combines expertise in artificial intelligence with the evaluation criteria established at IKAT and aims to create a model that can automatically generate compliant mechanisms for any desired motion requirements. This has the decisive advantage of drastically reduced computation times. Initial implementations of this innovative approach, such as for stiffness optimizations and the synthesis of compliant mechanisms, are already showing promising results. In the future, it will be possible to use this model directly within CAD programs, thereby efficiently translating boundary conditions and objectives into optimal geometries.
You can experience this approach on your own.