Research on characteristics of cab interior noise under different conditions by neural network algorithm.

This study aimed at establishing models to predict commercial vehicle's running conditions by neural network algorithms. Initially, experiments were carried out to collect cab interior noise data and a total of 420 samples were obtained after repeated tests. Then, each sample was intercepted and converted into six psychological acoustic metrics, which were sound pressure level (SPL), loudness, sharpness, roughness, articulation index and fluctuation strength. Finally, neural network algorithms were used to establish models between the running condition and the acoustic metrics. Through iterations, AC state, loading state, speed level, and road surfaces prediction models were figured out with accuracies of 0.903, 0.753, 0.978, 0.946, respectively. The results indicate that noise induced by speed variances greatly affects SPL, whereas road surfaces and AC states greatly influence sharpness and articulation index, separately. However, more evidence should be offered to verify the loading states have significant effects on interior noise.