Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Machine learning based structure and material optimization of 3D printed artificial spinal disc
3D printed artificial spinal disc that has optimized structure and material owns better performance. To address the expensive computational cost, machine learning is used for response calculation. The goal is to test methods of machine learning and select the best one for spinal disc optimization.
Keywords: 3D printing; Artificial spinal disc; Machine learning; Structure optimization
Total disc replacements (TDR) is considered as an effective way to treat load back pain. The major design goal of an artificial spinal disc is to mimic the mechanical response of natural spinal disc and thus restore the natural spine movement. However, most of the currently available artificial spinal discs are based on ball-and-socket design, which leads to insufficient resistance to bending and rotation. In addition, the limited options of disc standard sizes lead to geometry mismatch between the human vertebrae and artificial disc, which further results in occurrence of complications such as facet joint overloading and disc subsidence. Considering the ability of 3D printing to manufacture complex geometry and structure efficiently, we aim to design and manufacture an artificial spinal disc that is optimized in terms of structure and material according to personal needs.
Due to the large number of coupled design variables and high computational cost of FEA (Finite Element Analysis) based optimization, machine learning such as Artificial Neural Network (ANN) is considered as an efficient way to improve the optimization efficiency. Based on representative training data, machine learning is able to build a regression model between the inputs (design variables) and outputs (optimization goals). A properly trained model can be used to replace FEA for the optimization process to significantly decrease computational cost.
Total disc replacements (TDR) is considered as an effective way to treat load back pain. The major design goal of an artificial spinal disc is to mimic the mechanical response of natural spinal disc and thus restore the natural spine movement. However, most of the currently available artificial spinal discs are based on ball-and-socket design, which leads to insufficient resistance to bending and rotation. In addition, the limited options of disc standard sizes lead to geometry mismatch between the human vertebrae and artificial disc, which further results in occurrence of complications such as facet joint overloading and disc subsidence. Considering the ability of 3D printing to manufacture complex geometry and structure efficiently, we aim to design and manufacture an artificial spinal disc that is optimized in terms of structure and material according to personal needs. Due to the large number of coupled design variables and high computational cost of FEA (Finite Element Analysis) based optimization, machine learning such as Artificial Neural Network (ANN) is considered as an efficient way to improve the optimization efficiency. Based on representative training data, machine learning is able to build a regression model between the inputs (design variables) and outputs (optimization goals). A properly trained model can be used to replace FEA for the optimization process to significantly decrease computational cost.
The aim of the project is to explore and determine the best machine learning method for the structure and material optimization of 3D printed artificial spinal disc. In the first step, training data will be generated by running FE model based on inputs (combination of design variables). Next, different machine learning methods (e.g.: Support Vector Machine, Artificial Neural Network…) will be tested and validated for its prediction error. The optimal regression model will then be used for the structure and material optimization of artificial spinal disc. Finally, the performance of 3D printed artificial spinal disc will be evaluated by mechanical testing machine.
The aim of the project is to explore and determine the best machine learning method for the structure and material optimization of 3D printed artificial spinal disc. In the first step, training data will be generated by running FE model based on inputs (combination of design variables). Next, different machine learning methods (e.g.: Support Vector Machine, Artificial Neural Network…) will be tested and validated for its prediction error. The optimal regression model will then be used for the structure and material optimization of artificial spinal disc. Finally, the performance of 3D printed artificial spinal disc will be evaluated by mechanical testing machine.