Deep Learning Model for Classifying Cutting Volume at Shale Shakers in Real-Time via Video Streaming

Abstract

Abstract A real-time deep learning model is proposed to analyze the volume of cuttings from a shaker on an offshore drilling rig. The model is able to extract features and perform real-time classification in relatively good accuracy compared to the traditional video analytics method, which consumes enormous time on the feature extraction. Our approach mainly consists of two parts: 1) a multi-thread engine for decoding, processing, and encoding real-time video streams. The video streaming is offered by a modularized service called RSVP (Rig-Site Virtual Presence). RSVP enables aggregating, storing, transrating/transcoding, streaming, and visualization of video data from the rig. 2) a convolutional neural network (CNN) for the quantitative analysis. The CNN is pre-trained with videos collected from previous drilling operations. Additional noises are added to the video during the training stage to improve the robustness of the model. Each video frame needs to be properly normalized and re-scaled to a fixed size before fed into the CNN. The CNN will generate output to classify the volume of cuttings in real-time. The classification result includes category labels, ("ExtraHeavy", "Heavy", "Light", and "None"), as well as the probability for each label. The model and the workflow have been tested on videos streamed directed from an offshore drilling rig. The video stream has a bitrate of 137Kbps, 5.84 frames/s, and a frame size of 720x486. The CNN is trained with an Nvidia GeForce 1070 GPU. The deep learning model and the workflow is tested on a machine without GPU support. Due to the implementation of the multi-threaded video processing engine, we can handle decoding, encoding, video preprocessing and classification in real-time. This allows us to receive a real-time video stream and display the classification results with encoded frames on the user-side screen at the same time. In order to evaluate the result, we build the confusion matrix to facilitate the visualization of the performance of the model. Compared to results manually labeled by engineers, our model achieves more consistent results with improved confidence. The novelty of the project is the capability of monitoring and classifying the cutting volume at shale shakers in real-time via the video streaming.