Damage Identification Of Thin Plate Structures By Using Nonlinear Ultrasonic Features And Support Vector Machine

Since thin plate-like structures inevitably suffer from structural damage after being subjected to various external loads during their service period,the early damage identification of thin plate-like structures is of great significance.Lamb wave takes place nonlinear ultrasonic effect at the structural damage,in response to the issue of its optimal excitation frequency of that can reflect the size of structural damage usually changes as the size of structural damage increases,and the early damage characteristics of the structure are weak and difficult to identify.Therefore,this paper proposes a nonlinear ultrasonic method based on frequency-swept signal,which extracts and fuses the nonlinear ultrasonic features related to the damage,and classifies and identifies the damage of thin plate-like structures by using support vector machine.The main work is followed as below:(1)Based on the nonlinear ultrasonic theory,the nonlinear ultrasonic characteristics under different structural damages are analyzed.It shows that Lamb wave takes place nonlinear ultrasonic effect when it travels through the damage,and an additional second harmonic component of the excitation signal is also observed.Therefore,nonlinear ultrasonic energy features,bispectral features and statistics features in frequency-domain are extracted to reflect its nonlinear ultrasonic characteristics,and support vector machine is used to classify and identify the damage of thin plate-like structures.(2)To increase the efficiency of damage identification,single-feature based support vector machine is used to classify and identify the single damage of thin plates.The experimental results of the single damage identification on aluminum plate by using different features including nonlinear ultrasonic energy features,bispectral features and frequency domain statistics features are compared and analyzed from three aspects of recognition accuracy,training time and generalization ability.The experimental results show that the dual-spectrum fuzzy feature among the dual-spectrum features has the highest recognition accuracy for single damage(damage size 0.3mm～0.7mm),which can reach 89.44%.(3)Since the recognition accuracy of multi-damages by using a single feature is low,it is proposed to fuse nonlinear ultrasonic features including nonlinear ultrasonic energy features,bispectral features and frequency domain statistics and they are inputted into SVM for multi-damages classification.Moreover,the damage identification performance of an aluminum plate by fusing different features is compared and analyzed.The experimental results of multi-damages show that the multi-damages recognition accurary can reach 100% when 8 nonlinear ultrasonic features are fused from nonlinear ultrasonic energy and bispectral features to frequency domain mean in frequency domain statistics.