Recognition of Different Operating States in Complex Systems by Use of Growing Neural Models

Abstract

This paper proposes a technology for numerical comparison of different operating states in construction machines and other complex systems, working in frequently changing modes and under variable loads. The results from the comparison can be used for detailed operations recognition and fault diagnosis. The raw data from each operation are represented in a compressed form by a neural model. A special "growing model learning" algorithm is proposed in the paper and compared with the standard "fixed model learning" algorithm. Results from a test example show the superiority of the growing learning algorithm in terms of computation time and its ability to guarantee the predetermined model accuracy. Two methods for numerical comparison of pairs of operations, which utilize the trained neural models, are also proposed in the paper. They use the center-of-gravity and the relative size of each operation. Finally, an application of the methods to the comparison and recognition of eight operating states of hydraulic excavator is given in the paper