Structural load estimation of the wheel loader for customer usage profile monitoring.

This paper aims to estimate the structural load applied to the attachment and frame of a wheel loader by using pressure and acceleration, which have a load correlation, instead of strain. First, we propose a data-driven modeling methodology for load estimation. Load features as model input are derived based on cylinder pressure and frame acceleration. The relationship between features and the load is expressed through correlation coefficients and generalized by using the supervised machine learning technique. Next, to experimentally collect the large data set for learning, two wheel loaders of different classes are selected to build sensor cars. Then, the measurements on both wheel loaders are performed repeatedly with four experienced operators according to the test cases. Using conventional learning algorithms, model selection is performed for five parts that make up the attachment and frame. Then, the model learning and evaluation are performed. As a result of an out-of-sample test, the average estimation error for each part is approximately 5 % and the proposed methodology is experimentally verified that it is effective in estimating load for CUP monitoring. [ABSTRACT FROM AUTHOR]