Acoustic Field Analysis Of Transducer And Intelligent Identification Of Fracture-breakout In Ultrasonic Image Logging

In the exploration and development of oil and gas,the location and identification of fractures and breakouts is always a key problem for high-efficient exploitation of reservoirs safely.Ultrasonic imaging logging has been widely used in fracture-breakout detection since its birth due to its characteristics of non-contact,high resolution and full coverage of borehole.Aiming at the recognition and interpretation of wellbore geological features,the thesis studies the accurate recognition of geological features of ultrasonic imaging logging images from three aspects including high-resolution imaging,data generation and automatic detection of geological features systematically.Firstly,the acoustic field of ultrasonic transducer is analyzed to study the resolution of the system,which is directly related to the identification of the borehole geological features.The mathematical model of focused ultrasonic transducer is established,the formula of acoustic pressure distribution is derived,and the acoustic field characteristics of transducer with different frequency and geometric size is simulated by numerical simulation and finite element analysis respectively.The result shows that the focusing field of the focused ultrasonic transducer increases with the increase of the transducer frequency and curvature rate,and the corresponding lateral resolution also increases,which provides some guidance for geological exploration with different resolution requirements.In order to solve the problem of less ultrasonic borehole images and insufficient labeled data set,the thesis expanded the data set by artificial simulating and data augmentation.Considering the continuity of the left and right sides of the 2D wellbore image along the depth direction,the existing image texture synthesis algorithm is improved to characterize the continuity of the borehole image by add the continuous constraint of the two sizes.The method of using three Gaussian function superposition to simulate irregular breakout shape is proposed creatively.By randomly synthesizing fractures and breakouts,the data set of borehole ultrasonic amplitude image was greatly expanded and the data diversity was increased.A set of data augmentation process for borehole wall image was proposed,which can retain the actual data characteristics well and enhance the ability of data representation.The analysis shows that the recognition of geological features of borehole image is close to the semantic segmentation task of machine learning.The improved Unet model algorithm is adopted to simulated data and combined with the actual logging data of simulated data for training,which both achieve good results in the identification of geological features of actual imaging logging data.Especially,In the joint training with actual imaging logging data,the accuracy of model prediction and anti-noise performance has a certain improvement,but the prediction accuracy of complex distribution of microfractures and breakouts is not so good.In the actual interpretation of ultrasonic logging data,we could consider the machine model predicted results as reference and the relevant manual interpretation on the basis of the former as more accurate interpretation.