Reliability of Load Detection using Multi-Layer Perceptron and Convolutional Neural Network

Purpose: The detection of external and internal forces is important for the safety evaluation of mechanical structures. In this study, the results of using a multi-layer perceptron (MLP) and convolution neural network (CNN) to improve the detection reliability were analyzed.BRMethods: Load data collected from structural analysis of a crawler crane was used for training the MLP and CNN. A brief procedure for setting the learning conditions and functions for the MLP and CNN was presented and verified with quantitative figures.BRResults: The appropriate conditions for the functions used by the MLP and CNN for learning and recognizing the loads on the crawler crane were identified. Under these conditions, the maximum errors of the MLP and CNN were 0.24 and 1.45 tons, respectively. Thus, the MLP was more reliable at learning and recognizing the loads.BRConclusion: Based on the results, MLPs and CNNs can effectively detect loads on mechanical structures and systems. Thus, load-detection methods based on MLPs and CNNs could improve the safety evaluation of mechanical structures.