Real-Time Eccentricity Correction of Through-Tubing Cement Log Data Using Machine Learning

One of the major challenges associated with downhole Through-Tubing Cement Evaluation (TTCE) is the varying tubing or first pipe eccentricity (ECC) inside the casing or second pipe. As ECC varies, the signal or response from the cement behind the casing will change, therefore making it critical to have a method to measure and correct or compensate for the ECC effects in order to obtain a reliable cement bond log. The research study presented in this paper covers a real-time ECC correction algorithm for cement log data using neural networks. The scope covers the design, implementation, and evaluation of the ECC algorithm, as well as some of the potential benefits for the novel through-tubing evaluation services. The ECC algorithm uses a neural network that is trained for various logging environments. The neural network generates a weight matrix that is defined in a calibration file, which is then loaded into the acquisition software prior to the logging operation. The neural network structure consists of multiple layers of interconnected nodes, each with its own set of weights and activation functions. The algorithm works by reducing the effect of eccentricity on the bond index accuracy of through-tubing cement data. Inputs to the algorithm encompass both time and frequency domain signals from six azimuthal segments around the pipe to model the ECC effects. The neural network effectively demodulates the impact of tubing eccentricity and outputs ECC-corrected cement segment signals. The paper presents the overall approach and details of the process involved in designing, implementing, and evaluating the ECC algorithm during data acquisition and post-processing. The simulation and lab test results from this study demonstrate the effectiveness of the ECC algorithm in reducing the effect of eccentricity on through-tubing cement data. The real-time application of the ECC algorithm results in more accurate bond index measurements, which can improve well integrity evaluation. This technique is particularly useful in situations where eccentricity is an important factor affecting bond index, where the tool string and tubing positions in the well leads to a higher uncertainty in interpreting cement evaluation data. Overall, the developed algorithm adds significant value to the improved hardware and software solutions as part of the research effort towards a comprehensive TTCE solution for rigless operations. This paper presents a comprehensive description of the ECC correction algorithm's design, implementation, and evaluation, and demonstrates its effectiveness in improving bond index accuracy. The integration of machine learning tools to provide reliable real-time cement bond logging without the need to pull tubing, presents a major advance in well integrity evaluation that further improves safety, efficiency and cost.