Research On Defect Inspect And Pattern Recognition Of Sucker Rod

Sucker rod is an important part for pumping oil equipment with rods, and it is large-scale to be used. Sucker rod works under cyclic tensile stress, and it has impact and extrusion load in assembling and transporting to in the pit process, so it is apt to crack, corrosion and so on in the poor working conditions. If sucker rod defects are not found in time, will happen downhole rod fracture in the incident, causing great economic losses, so right sucker rod defects on the surface of feature extraction and the correct identification, and take corresponding measures, These will reduce broken poles from the accident, and has low production costs.At present, the detect is mainly focuse on crack inspecting for sucker rod automation detecting, but corrison and partial wear is the direct cause to crack, so discoving as soon as early the non-crack defects and renovating the suck rod, then avoid to coming to being crack and prolong the suck rod service life.The main object is developing the defects inspect and recognition system applying for sucker rod, so the detecting method, signal denosing, feature extraction and pattern recognition and so on are studied.For the defect inspecting method, applying eddy-current testing (ET) and magnetic flux leakage (MFL) for sucker rod, and the inspecting result shows that the MFL is better than ET for sucker rod inspecting, influencing factor is few and sensitiveness is very high.In order to eliminate noise during the detection, the denoise way with improving threshold function is proposed, and the improved threshold function is the combine betweent the hard-threshold and the soft-threshold, so has the continuous of soft-threshold, and choose the vivid parameter a,so can obtain a valid way. The experimental result shows that the way is valid.With the way of feature extraction, a method is presented to extract frequency band energy feature by using wavelet package decomposition. The way can make the not obvious signal frequency feature to come as the sub-spaces energy with different resolution, and compares the normal output, in order to extract feature with system fault. The experimental result shows the validity of this method. In the meantime, to extract the peak-peak value in the time-domain and make the mixed feature vector. Applying the separable criterion based on within-class and between-class distant to improve the mixed feature has the better separable feature, and increase validity of classification of feature greatly. With the way of pattern recognition, compares the BP Neural network,PNN neural network,"one-versus-one" support vector machine(SVM) and improved "one-versus-one" svm at sucker rod defect inspection, and the result shows that improved "one-versus-one" svm has more high recognition rate, and be suitable to the sucker rod defect recognition. The improvd "one-versus-one" svm reduces the influence of undivided area, and uses the distant to solve the recognition in the undivided area, so can enhance the defect recognition rate.Based on above study, developed the sucker rod defect inspection and recognition system based on wavelet and improved svm. The system has been applied for one lifting oil factury, and the applied result shows the system can reduce the lifting cost and prolong the pump inspection cycle.At last, in the paper, because of the low accuracy for micro-crack inspecting, so using a new way of magnetic flux leakage inspection under stretching sucker rod by liquid force in the material the flexible scope. A 45 steel specimen with artificial crack of 0.1 mm and 0.3mm and with fatigue crack was used to verify the method. The results show that the way not only can insure NDT, but also can improve inspecting accuracy and sensitivity. It is valid path for improving sucker rod crack inspected in oil field, and also give a refert for micro-crack inspecting of bar.