Wellbore stability prediction method based on intelligent analysis model of drilling cuttings logging images

At present, drilling sites usually rely on rock mechanics analysis results to predict wellbore stability, which takes a long time. Therefore, this study attempts to use real-time drilling cuttings logging image data to characterize the results of rock mechanics analysis, so that drilling cuttings logging has the function of predicting wellbore stability. The study established an image sample library consisting of 16 types of drilling cavings shapes and lithology, and improved ShuffleNetV2 network as the basic architecture to form an intelligent prediction model. In order to enhance the network's attention to the iconic feature information of drilling cavings images, XConv convolutional kernel parallel branches and SimAM attention mechanism modules were introduced into the Shuffle unit. In order to preserve key features of drilling cavings contours, a multi-channel feature fusion algorithm was designed for Stage2, Stage3, and Stage 4 stages in ShuffleNetV2 network. The final improved ShuffleNetV2 network model has a recognition accuracy of 90.56 % for the shape and lithology of drilling cavings. The effectiveness of the on-site application of Fenggu ∗ Well has verified the reliability of this method. The time from input of returned cuttings images to output of results is less than 1 s, and the recognition and prediction results are basically consistent with geological data and construction process conditions. This fully demonstrates that this method can meet the needs of rapid perception and prediction of wellbore stability on site.