Parameter identification of a shell transfer arm using FDA and optimized ELM

To identify the unmeasurable parameters of a shell transfer arm, a virtual prototype of the shell transfer arm was built, and the built virtual prototype is regard as the source of the sample data. Considering the continuity and smoothness properties of the sample data, features of the curves were extracted by functional data analysis and functional principal component analysis, and the features and unknown parameters were used to train the extreme learning machine (ELM). At the meantime, the weight connecting the input layer and hidden layer and the threshold of the hidden nodes were optimized by particle swarm optimization (PSO) to improve the identification accuracy and generalization performance of ELM. At last, the presented method was verified by simulation data and test data. The identification results of the simulation data show that the optimized ELM has higher identification accuracy and better generalization performance. Also, the presented method is proved to be feasible and effective by comparing the real angular velocity and the angular velocity from the virtual prototype with respect to the test data identification results.