| Lithium batteries are widely used in new energy vehicles,such as drones,cars and aircraft,etc.,with the popularity of new energy vehicles,non-destructive testing of lithium batteries has become one of the hot directions of major battery manufacturers and research institutions.In the scheme of non-destructive testing with lithium battery as the non-destructive testing object,the detection scheme for the weld defects of lithium battery lug plates is an important branch of lithium battery non-destructive testing.Therefore,this thesis aims at the weld defects of the avionic lithium battery lug plates,and uses the change characteristics of the eddy current signal as the basis for defect classification: firstly,a non-destructive testing experimental platform is built to verify the feasibility of eddy current detection;Then,combined with machine learning algorithms,the model is trained by using the sample data collected by the experimental platform to achieve accurate classification of defects.There are three main aspects of this contribution:(1)Focusing on the characteristics of aviation lithium batteries such as larger size,larger capacitance and more welding places of lug plates,and aiming at the problem that the induced voltage signal detected by traditional eddy current sensors cannot remove excitation noise,a nondestructive testing experimental platform for the weld of aviation lithium battery lug plates is designed.Firstly,a single-frequency probe is designed according to Faraday’s law of electromagnetic induction,and an experimental platform for strongly excitation eddy current detection is constructed by using the designed single-frequency probe,which verifies the feasibility of eddy current detection.Then,on the basis of feasibility,facing the problem of many welding places in the tabs of aviation lithium batteries,the detection structure based on the measured induced voltage signal data of multi-frequency differential eddy current sensor array is proposed by using differential detection coil,and the differential detection sensor structure is used to design a multi-frequency array differential eddy current sensor,and a nondestructive testing platform for aviation lithium batteries is constructed.This platform realizes the acquisition of induced voltage values of the electrode plate weld based on eddy current,and solves the problem of removing excitation noise.(2)Based on the above experimental detection platform,in order to meet the needs of inspection data visualization defect detection,this thesis further focuses on the test data output by the experimental platform to achieve frequency sweep tomography of the tabs and plates.Swept tomography can analyze the eddy current distribution at different skin depths,so that the presence of defects in the tabs can be visually judged from a visual angle.(3)After obtaining the corresponding induced voltage data,traditional machine learning methods,including SVM,KNN,etc.,are first used to classify and train the detected induced voltage signal data to obtain the model,but the experimental conclusion shows that the accuracy of defect classification does not meet the actual requirements.In order to solve this problem,based on the ideas of decomposition and decoupling and HRNet,this thesis proposes a discriminant method based on multi-scale fusion strategy,and compares it with the existing neural network-based algorithms,including Res Net,Dense Net,Google Net,etc.,and the experimental conclusion shows that the proposed method based on multi-scale fusion strategy is superior to other methods in classification recognition rate compared with other methods,and has certain practical application value and significance. |
