Light weight cement slurries with foam additions are being used in multiple applications including onshore and offshore deep wells installed in varying geological formations with low strength rocks. Also, the foam cement has reduced thermal conductivity making it a better insulator. It is also important to model the behavior of the foam cement for real time monitoring with the artificial neural network (ANN) models for adaptation in machine learning for various applications. The investigation was on the performance of highly sensing foam added smart cement and verified with various behavior models. Added foam was characterized based on stability, half-life, electrical resistivity, oxidation-reduction potential (ORP), and pH. Highly sensing smart oil well cement with water to cement (w/c) ratio of 0.38 was modified by adding 5% and 20% foam by weight and was first characterized using the impedance–frequency response to identify the critical electrical property for monitoring. Based on the Vipulanandan Impedance Model, electrical resistivity was the monitoring property. Also rheological properties, fluid loss and piezoresistivity (slurry) were investigated. With the addition of 20% foam the density reduced by 45 the thermal conductivity reduced by 65%.With the addition of 20% foam the initial resistivity increased by 94% indicating a potential quality control parameter for monitoring in the field. The thermal conductivity and the initial resistivity were related to the density using the Vipulanandan Correlation Model. The slurries investigated were piezoresistive, when pressure was applied the electrical resistivity changed. The rheological behavior of the smart foam cement slurries have been quantified using the new Vipulanandan rheological model and compared with Artificial Neural Network (ANN) Model. The total fluid loss with the addition of 20% foam, reduced by about 90% and the Vipulanandan fluid loss model and the ANN model were used to predict the behavior. The slurry piezoresistivity model was used to predict the piezoresistive behavior and compared it to an ANN model. The accuracy of the all the model predictions were compared using the statistical parameters.
Read full abstract