Deep Neural Network Structure Design for Equipment Failure Prediction in Smart Factory

초록

A smart factory is a system in which all equipment is connected through the Internet of Things (IoT) and all facilities are operated through a virtual physical system. In the factory, failure prediction for facilities maintenance combined with deep learning has recently attracted much attention to researchers. To perform deep learning-based prediction, a neural network model is built and trained using the learning dataset. The training performance varies depending on the structure of the constructed neural network design. In this study, an efficient design of a neural network is explored that is expected to have the best training performance using a dataset of vibration and temperature. To do this, different neural network models are implemented, and their performance was compared by evaluating loss, accuracy, and execution time. In the experiment, the neural network model was configured to have one input layer, two or three hidden layers, and one output layer, and performance metrics values were measured according to the internal node numbers in every hidden layer. Experimental results show that losses do not increase in spite of that node number becomes large. In general, as the structure of a deep neural network model is complex and the number of nodes inside it increases, it is expected that the loss value will be smaller. For accuracy, there was almost no significant difference between all models. In the case of execution time, it took more time as the number of internal nodes of the model increased.

키워드