Scholarlink

<p>To improve the pass rate and work efficiency of Flex-PLI knee static calibration, the overall displacement curve of the medial collateral ligament (MCL), anterior cruciate ligament (ACL), and posterior cruciate ligament (PCL) can be made to approach the median value of the upper and lower limits by optimizing the elongation of the knee spring, which means that the knee mechanical performance reaches the ideal optimal state. Aiming at the uncertainty of the nonlinear relationship between spring elongation and corresponding ligament elongation, a mechanical performance optimization method of Flex-PLI knee based on improved Keras neural network model is proposed in this paper. The mechanical performance analysis of the knee is carried out to obtain the elongation error of three groups of ligament as input for the Keras neural network. The consistency of experimental results is ensured by setting random seeds. Using Optuna to optimize hyperparameters, the optimal hyperparameter configuration is obtained and applied to the main part of Keras neural network to improve the efficiency of model parameter adjustment. At the same time, dropout regularization technology is used to reduce the overfitting phenomenon of the model and improve its generalization ability. The training and testing samples of the model are extended through K-fold cross validation, and the training effect of the model is evaluated according to the average test error. Finally, the elongation of four rows of springs at the ideal optimal state of MCL, ACL, and PCL displacement curves are obtained through training. The experimental results show that the improved Keras neural network prediction model has a significant optimization effect on the mechanical performance of Flex-PLI knee within the current range of experimental parameter. The overall MAE of the validation group decreases significantly compared with that of the initial group and the experimental optimal group. By this method, the mechanical performance of the knee can be precisely regulated within the target parameter range, which provides a good theoretical and practical basis for more efficient and reliable static and dynamic calibration of the Flex-PLI knee.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mechanical performance optimization of Flex-PLI knee based on improved Keras neural network model

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Rundong Huang,
  • Jian Yang,
  • Haipeng Hou,
  • Ye Pan,
  • Li Zhang,
  • Dongjie Zhu,
  • Mingsheng Jin

摘要

To improve the pass rate and work efficiency of Flex-PLI knee static calibration, the overall displacement curve of the medial collateral ligament (MCL), anterior cruciate ligament (ACL), and posterior cruciate ligament (PCL) can be made to approach the median value of the upper and lower limits by optimizing the elongation of the knee spring, which means that the knee mechanical performance reaches the ideal optimal state. Aiming at the uncertainty of the nonlinear relationship between spring elongation and corresponding ligament elongation, a mechanical performance optimization method of Flex-PLI knee based on improved Keras neural network model is proposed in this paper. The mechanical performance analysis of the knee is carried out to obtain the elongation error of three groups of ligament as input for the Keras neural network. The consistency of experimental results is ensured by setting random seeds. Using Optuna to optimize hyperparameters, the optimal hyperparameter configuration is obtained and applied to the main part of Keras neural network to improve the efficiency of model parameter adjustment. At the same time, dropout regularization technology is used to reduce the overfitting phenomenon of the model and improve its generalization ability. The training and testing samples of the model are extended through K-fold cross validation, and the training effect of the model is evaluated according to the average test error. Finally, the elongation of four rows of springs at the ideal optimal state of MCL, ACL, and PCL displacement curves are obtained through training. The experimental results show that the improved Keras neural network prediction model has a significant optimization effect on the mechanical performance of Flex-PLI knee within the current range of experimental parameter. The overall MAE of the validation group decreases significantly compared with that of the initial group and the experimental optimal group. By this method, the mechanical performance of the knee can be precisely regulated within the target parameter range, which provides a good theoretical and practical basis for more efficient and reliable static and dynamic calibration of the Flex-PLI knee.