Natural frequency identification model based on BP neural network for Camellia oleifera fruit harvesting

Vibratory harvesting is an important means of mechanically harvesting tree fruit. The optimal excitation parameters are usually experimentally determined under complex conditions with different environments and machine configurations. Optimisation methods include tree modelling and dynamic analysis but experimental validation can take much time due to the complexity of tree structure and properties. A simple and appropriate identification model that could identify the natural frequencies of trees might simplify the process and promote the technology. A natural frequency identification model is proposed based on back propagation (BP) neural network to identifying the natural frequency of the tree based on its structure. Taking Camellia oleifera tree with its upright canopy as an example, the excitation parameters that can achieve better harvesting of fruit was determined here by orthogonal test. A dynamic model was established, and the tree structure variables were derived as the input layer of the model. The dataset of tree dynamics was established by finite element analysis and the effective natural frequency region was set as the model output layer. A natural frequency identification model was established based on TensorFlow, where the input and output parameters are fitted using a BP neural network. Application of the model was carried out after substantial training and testing. In the range of natural frequencies 6-7Hz, the mean square error between the natural frequency identification value and the measured value was only 0.0408, which verified the reliability of the model.