| Titanium (Ti) alloy is an emerging structural material,which is widely used in many fields such as aerospace,marine engineering,civil engineering and biomedical engineering because of its superior performance.Among all kinds of Ti alloy products,sheet is the most widely used,accounting for about 50% of the total output.However,during the rolling process of Ti sheets,some quality issues may occur,such as uneven thickness,the same plate difference excess the permissible error,poor surface quality poor,including cracks,pits or other defects.Eddy current testing(ECT)is a nondestructive testing(NDT)technology based on electromagnetic induction,which is widely used in the NDT and evaluation of metal materials,and has the advantages of high efficiency,low cost,no pollution compared with other NDT technologies.The use of ECT for Ti alloy sheets is still facing some problems,such as the detection signal is easily disturbed by noises,the difficulty of thickness inversion during the online measurement,and the accuracy of classification and quantitative evaluation for defects need to be further improved.In this paper,after discussing the quality issues that may occur during the rolling process of Ti ally sheet,based on the current research status of ECT at home and abroad,the methods on eddy current NDT and quantitative evaluation of Ti alloy sheets were carried out.The main research work is as follows:(1)To solve the problem of noise reduction pre-processing of eddy current(EC)detection images of titanium sheets,a Principal Component Analysis(PCA)combined with Local Linear Embedding(LLE)method for noise reduction of EC images was developed.The PCA method was firstly used to initially process the eddy current detection images to remove the large noise.Then,the LLE algorithm was used to automatically find the nearest neighbor points within the variable neighborhood of the pixel,and reconstruct the pixel with a locally optimal linear combination of the nearest neighbor points,which can maintain the local geometry of the image while further fine noise reduction.Experimental results have shown that under the conditions of the input signal-to-noise ratio(SNR)of 18 d B?33d B,the output SNR can reach 30.96?42.36 d B,and the EC images SNR is increased by 12.71 d B on average,which confirmed the validity of this method in denoising pre-processing the EC images of Ti alloy sheet defect.(2)To solve the problem that traditional EC thickness measurement methods are difficult to obtain the sheet thickness directly by the detection signal,an EC analytical model of Ti alloy sheet and an improved EC thickness measurement method were proposed.Combining the low conductivity and non-ferromagnetic properties of Ti alloy,under specific measurement conditions,a new analytical model was deduced to describe the relationship between the thickness and the real part of differential voltage in the EC probe.Finally,based on the deduced analytical model,a fast and accurate EC measurement method for Ti alloy sheet thickness was proposed.The thickness of Ti sheet can be calculated directly from the calibration signal by using the method,without the need to obtain experimental records or predict response for comparison by preexperiment or solving forward models respectively.The experimental results have shown that the measurement error of cold-rolled Ti sheet within 6mm thickness is no more than ±2.86%,which is much smaller than the current industry standard of ±7.0%.(3)To solve the problem that traditional machine learning methods are difficult to accurately and effectively classify EC detection image information of Ti alloy sheet defects with high dimensional and nonlinearity under different working conditions,a Stacked Sparse Denoising Autoencoder deep neural network model for classification of EC images of Ti sheet defects was proposed.Sparsity restriction was introduced into the Denoising Autoencoder(DAE),and the deep network model including multiple hidden layers was constructed by stack combining multiple DAEs.The trained network model can effectively classify and identify Ti sheet defects under different working conditions,has good robustness.The results of the Ti sheet defects classification experiments have shown that when the input SNR is greater than 26 d B,the classification accuracy of the method is higher than 97%;when the input SNR gradually decreases,the classification accuracy can also be maintained at a high leve,which has the advantages of antiinterference and stability in the complex environment of the Ti sheet production line.(4)To solve the problem that EC detection images are difficult to design features manually and conventional regression methods are need to design complex objective function in quantitative evaluation of Ti alloy sheet defects,a quantitative evaluation method for EC image of Ti sheet defect based on Deep Belief Network(DBN)and Least Squares Support Vector Machine(LSSVM)was proposed.Based on the correct defect classification,the DBN network was used to extract effective features from the EC images,and the extracted feature vector and LSSVM algorithm were used to establish a multi-objective regression model to achieve the quantitative evaluation of Ti sheet defects.The experimental results have shown that the relative error and repeatability standard deviation of the defect evaluation are less than 4.1% and 0.12 mm respectively,which is more accurate and repeatable than other conventional methods.(5)The design and development of an eddy current NDT system for Ti alloy sheet were completed.According to the NDT system requirements,the overall architecture of the NDT system was designed,and the software and hardware functions of the NDT system were modularly implemented.Finally,the thickness measurement model,classification identification and quantitative evaluation intelligent algorithm of the above researched were integrated into the NDT system.The application examples of eddy current detection,classification and quantitative evaluation of Ti alloy sheet defects confirm the effectiveness and practicality of the developed system. |
PDF Full Text Request