Identifying Abnormal Noise of Vehicle Suspension Shock Absorber Based on Deep Belief Networks

Considering the complexity and non-expandability of extracting abnormal noise features of shock absorbers by experience and manual work, applications of deep belief networks (DBNs) to identification of vehicle suspension shock absorber's abnormal noise are discussed, and a complete identification process of shock absorber abnormal noise is proposed by combining the shock absorber's road test with its rig test. The method only needs to take the vibration acceleration signal of the shock absorber piston rod as input, and then process the signal by learning layer-wise features in the DBNs model to classify the sounds of shock absorbers. In addition, the identification accuracy by DBNs is compared with that by the classical BP neural network, support vector machine, and other three traditional abnormal noise identification methods. The results show that when only the original signal is used as input, the classification accuracy by DBNs is 96.7%, which is higher than that by the other five methods. This illustrates the superiority of the DBNs algorithm in identifying the abnormal noise of shock absorbers and may imply a wide prospect in engineering application.