The Research Of Defects Signal Automatically Identification In Bottom Plates Of Oil Tanks Based On Ultrasonic Testing

Oil tanks as pressure vessel are very essential to the storage and transportation process on oil and gas.With the technological development,the scale of oil tanks have been increasing all the time due to the increase of their needs for oil and gas usage in all industries and fields.If we want to make sure that oil tanks can be worked safety within their life periods,one of the best methods is to use non-destructive testing technology,i.e.,ultrasonic testing and acoustic emission et al.,for oil tanks evaluation in irregular intervals and take measures to keep oil tanks integrated.By this way,we can have a basic knowledge about how oil tanks’ integrity conditions are being and make a detailed plan to do something about maintaining for them in order to keep them all the time in good working conditions within a deserted environment.The methods we commonly use for non-destructive testing are ultrasonic wave,acoustic emission,magnetic flux leakage testing,magnetic powder inspection,radiographic testing and eddy current testing.Ultrasonic wave testing with some advantages,i.e.,wide detection range,high detection accuracy and accurate orientation,is very suitable to be used for the inspection of defects in the oil tanks bottom plates.While,traditionally,the ultrasonic testing was accustomed to be implemented by manual,leading to low testing efficiencies.Besides,the result of ultrasonic wave testing is depended on the inspector,in the other word,if the inspector has a rich experience of understanding how to conduct ultrasonic wave testing process properly,the performance of detecting could be good enough.Otherwise,the result of that would be unreliable.Responding to the problem mentioned above,the efficiency of ultrasonic testing was made a great progress in the paper by adopting machine learning method to automatically classify different types of ultrasonic defect signals.In order to achieve the function of automatic identification,the first thing we need to do is for preprocessing signals which is a kind of way to extract characteristic values.Wavelet packet decomposition by decomposing time domain signal into different frequency components for signal preprocessing has been utilized.Using average value,maximum value,entropy value and variance value represented every frequency component as characteristic values to be trained and identified by machine learning turned out not to be a good idea.Through classification accuracy of machine learning,we realized that wavelet packet decomposition is not suited to process signal this time.Time domain ultrasonic signal and Pearson correlation coefficient method used for calculating correlation coefficients of two types of defect signals in each part have been investigated in the paper.By comparison among correlation coefficients in different part,we narrowed the range of extracting characteristic values down by searching signal in low correlation coefficients part and found the defect features in time domain signal within a short time.A mathematical algorithm called separability algorithm also has been developed to calculate the separability values of characteristic values extracted from time domain signals.By this way,the performance of separability in characteristic values can be quantified by the results calculated by separability algorithm.By demonstration of classification results of 96.66%(29/30),it shows that using pearson correlation coefficient method and separability algorithm are efficient ways of preprocessing signals.