Detection For Filling Flow Of Viscous Food Based On Recurrent Neural Networks

In the field of automatic quantitative filling of liquid food,flow detection is one of the key link,which affects the filling speed and quantitative accuracy.Especially for viscous food,due to its inherent complex non Newtonian rheological properties,the accurate flow measurement in the filling process is very challenging.For this reason,this thesis carried out the research of filling flow detection for viscous food based on recurrent neural networks,the main research contents are as follows.(1)Aiming at the problem that it is difficult to accurately detect the filling flow of viscous food due to its high viscosity,and easy to be produced wire drawing and foaming,the idea of filling flow detection for viscous food based on recurrent neural network was proposed to realize the active sensing of filling flow.Firstly,the mechanism and characteristics of viscous food filling were analyzed,and its long-term correlation were described.Secondly,compared with feed-forward neural network,the structure and training method of recurrent neural network were elaborated in detail,and reasonable interpretations for its application in this problem were expounded.Finally,the workflow of this method was briefly described,and the possible shortcomings of recurrent neural network were discussed.(2)In order to solve the problem of gradient disappearance in the training process of recurrent neural network,a filling flow detection method of viscous food based on LSTM-Attention was proposed.Firstly,the data set of filling flow-related variables were described and processed to be converted into serialization data,which could be handled by supervised learning network.Then,the LSTM network based on the attention mechanism was trained and generalized,which adopted tensorflow back-end and the Adaptive moment estimation optimization algorithm,thus completed the construction of the filling flow detection model.Finally,the performance of the model was assessed on the test set by using the root mean square error and time of model training as the evaluation index,and compared with other network models on the same data set.The experimental results showed that the model can effectively detect the filling flow of viscous food and behaved well.(3)In view of the problem that the long-short term memory neural network needs to be improved in training speed of the model due to its complex gating structure,a filling flow detection method of viscous food based on SE-LBU-MGRU network is proposed.Firstly,the MGRU which only keeps the update gate was used as the hidden unit to build the linked bidirectional and undirectional minimal gated recurrent unit networks.secondly,the principle of SE-Net was analyzed and the attention module based on it was built,besides it was embedded in LBU-MGRU network,and then the complete model was well done by using pytorch framework and adding early stop optimization algorithm.finally,a comparative experiment on the same data set was conducted,and the network was proved that it not only ensures certain accuracy,but also improves the training speed of the model.