Non-destructive defect detection in solid objects using the photoacoustic method

Photoacoustic (PA) is a technique that utilizes acoustic waves generated by the conversion of light energy into thermal energy to detect material defects. The PA method is advantageous since it allows for non-destructive testing (NDT), thereby enabling defect detection without causing any damage to the material, and it can be applied to a wide variety of materials. In this method, laser light is sent into the material and the light absorbed by the material is converted into thermal energy. The thermal energy results in thermal expansion, which in turn results in the generation of sound waves. The sound waves are then detected by the acoustic sensors for analysis. PA signals show different characteristics from intact and defective materials. These characteristics can be described by parameters such as signal amplitude, frequency, duration and time-frequency characteristics. Machine learning (ML) methods are used for classification of these parameters to detect defects. Photoacoustic has become an increasingly popular method for material defect detection in recent years due to its advantages. In this study, we propose a framework for material defect detection based on Empirical Mode Decomposition (EMD) based feature extraction in time and time-frequency domain for PA signals obtained from aluminum, iron, plastic and wood materials using a laser, microphone and data acquisition board. Within each material group, a total of 240 samples (120 intact samples and 120 defective samples) and a total of 960 samples were used. With 14 features extracted from these samples, k-nearest neighbor (k-NN), decision tree (DT) and support vector machine (SVM) classifiers were used to classify defective and intact materials. Materials were classified both within the same class and within the class including all material groups. With the leave-one-out cross-validation method, the defective and intact detection rates for all materials and within the class were 100% for SVM and 97.77% for k-NN, respectively. The method proposed in this study makes a significant contribution to improve the effectiveness of the PA method for surface defects in different materials. With EMD, the noise in the PA signal is filtered out, the effect of the baseline drift is reduced and at the same time a useful signal is obtained for feature extraction. The proposed method effectively improves the signal processing by eliminating the baseline drift in the PA signals while performing two different tasks at the same time. In this way, the accuracy rate of defect detection in different materials is significantly improved. In contrast to the material-specific studies in the literature, with the proposed method, a common classification framework was developed for different materials with different defects.