Abstract: The time delay neural network, extensively applied in speech recognition, is introduced to identify random dynamic loads. Combining the "memory" property of the time delay neural network with the causal Finite-Impulse-Response (FIR) system theory and the steady response solution of the vibration theory, we propose a reversed time sequence dynamic load identification method. Experimental verification of the proposed method is conducted using an aircraft rudder model excited by two-point random loads. The results demonstrate that the root mean square errors between the time histories of the identified and real dynamic load samples on the two loading points are 0.635 4 and 2.543 7, respectively, and the correlation coefficients are 0.9657 and 0.8262, respectively. The curve of the power spectral density function between the identified and real dynamic loads on the two loading points coincides fairly well. The proposed dynamic load identification method has the advantage of high precision and requires no structural modelling.

Key words: load identification, time delay neural networks, random dynamic loads, reversed time sequence identification, causal finite-impulse-response systems