Unprecedented levels of manufacturable complexity can be achieved by new additive manufacturing technologies. Traditional design methods are being superseded by automated design methods, such as Topology Optimization, that can fully exploit these new capabilities. The paper presents the structural optimization of two existing turbine blade designs and a performance assessment of an open-source topology optimization algorithm. The low-pressure T106C and high-pressure C3X (cooled) turbine blade models were redesigned to minimize the material volume. The Solid Isotropic Material with Penalization (SIMP) and Sequential Element Rejections and Admissions (SERA) methods were used and compared. The SIMP method was applied with the commercial optimization code in the Abaqus Topology Optimization Module and the SERA method with an open-source optimization code. The open-source code was modified for implementing geometric constraints, extending its utility to the structural optimization of structures with multidisciplinary design considerations. The resulting optimal geometries from SIMP and SERA were predominantly similar, revealing hollow, thin-walled blade topologies with axially distributed material reinforcements at the leading edge and suction side, the locations of maximum bending and tensile stresses. Despite this similarity, considerable differences in the intricacy and complexity of the designs were observed. The SIMP-derived topologies were characterized by smooth boundaries and an even material distribution across the camber line. The SERA-derived topologies were composed of intricate features with unrealistic connectivity, reflecting the finite element discretization applied on the original geometry. The SERA method was demonstrated to be less computationally efficient, consistently requiring about 2.6 times longer than SIMP to achieve convergence but produced better-performing geometries, with 24.7% and 37.9% lower objective functions for the T106C and C3X geometries respectively. The effect of Laplacian mesh smoothing applied by the ATOM SIMP optimizer on the performance and optimal design is believed to play a major role and therefore commented in the paper.

This paper shows the current research to move towards the full digital design of a gas turbine. In the last years new manufacturing technologies, such as additive manufacturing, become more common for gas turbine applications, allowing greater flexibility in the design space. There is a need to fully exploit this flexibility and to design and validate in a digital environment new solutions. This work shows how optimization methods, mainly based on topology optimization strategies, requires more accurate estimator for critical applications, such as high temperature components of high pressure stages. For this reason a comparison of recent Gene Expression Programming and Neural Networks in topology optimization are shown. In particular it is shown how a RANS estimator in fluid topology optimization is capable of obtaining predictions compatible to high fidelity DES.