Advanced signal processing and pipeline leak detection algorithms development

Synthetic well log generation and software development

Developed and commercialized the methods and procedures for robust downhole fluid analysis using Concatenated Optical Computing Networks (COCN) that combine optical sensor data transformation networks and fluid characterization networks. The new COCN developments enable two-directional forward and reverse progressive modeling to improve the calibration of transformation networks.Developed and commercialized a novel technique to generate full-range fluid optical spectra from optical sensor measurements with neural networks. This implementation expands the functionality of existing real-time optical tool data processing software, and improves the quality of downhole fluid characterization, especially in fluid typing and contamination analysis.Developed and commercialized a novel technique applied to pooled optical sensor calibration with neural networks. This methodology can be used to ruggedize the calibration for both optical sensor data transformation networks and sensor-based fluid characterization networks, and minimize the uncertainty in real-time multi-sensor tool data processing.Developed and implemented a novel method for predicting the chemical additive concentrations in a fracturing fluid when the water is supplied from a non-traditional water source. This method uses a combination of neural networks and genetic algorithms for fluid design optimization based on a given fluid chemistry and expected viscosity. Developed a novel method for MOE (Multivariate Optical Element) design optimization and performance evaluation. This new method uses genetic algorithm to evolve each individual film thickness given the fixed layer structure, and apply constrained multi-objective performance function to generate diverse candidate designs for further fabricability study.

Developed pressure pulse testing optimization method which can be applied to very low permeability reservoir testing with wireline formation tester for quick stabilization in determining multiple reservoir parameters. The integrated workflow includes pre-design optimization, test execution and automation, calibration transfer and inverse processing.Developed inversion methods applied to formation testing with analytical flow models. The methods are implemented with deterministic, probabilistic and evolutionary approaches. These methods can be implemented into future formation testing simulators as faster regression algorithms to infer reservoir parameters from measurement data.Developed and implemented a software module in C# for HFDT (High Frequency Dielectric Tool) logging data interpretation. The module has capability of processing complex number, integrating measurements from multiple tools to generate robust solution with reduced uncertainty. Implemented an analytical simulator using spherical flow equations for near borehole pressure transient analysis applied to well testing. The simulator can be used to evaluate flow-line storage and mud skin effect, and optimize pretest and pulse test design with minimized operational cost.

Developed an integrated solution method in determining minerals of complex shale reservoirs from geochemical inputs. Compared with using conventional normative analysis alone, the new method using multi-disciplinary model fusion (intelligent linear programming, feed-forward neural network and normative analysis) reduces mineral prediction error up to 30 percent on the testing shale reservoirs(Woodford Shale and Haynesville Shale). Automated PNT (Pulsed Neutron Tool) data pre-processing such as depth matching, anomaly detection and full-set decay curve visualization/normalization for cased-hole interpretation (CHI) software. Optimized information utilization with CHI by designing and implementing methodology of multi-well neural network ensembles. Improved prediction on open-hole formation parameters and reduced cost in data processing demonstrated through extended field testing.

Played a vital role in development of Halliburton CHI Modeling System for predicting open-hole triple-combo data (porosity, density and resistivity) using cased-hole pulsed neutron inputs with neural networks. Invented and commercialized a variety of multi-objective genetic algorithms (MOGA) in selecting member neural networks to construct surrogate model ensemble with respect to its fidelity, complexity and member diversity. The CHI Modeling System using MOGA as a key functional component was honored by 2006 Hart’s Meritorious Engineering Innovation Award. Designed and tested an earth deformation models (EDM) in conjunction with genetic algorithm to estimate reservoir orientation, volume and boundary parameters based on the measured surface displacement due to the formation pore/bulk pressure change caused by injection/production. Feasibility demonstrated on the cyclic steam stimulation (CSS) induced deformation data recorded by Cold Lake InSAR image using linear superposition of multiple point-source prolate-spheroid models. Designed and implemented a cooperative optimization algorithm to reduce the data dimensionality with minimized information loss. The positioning of the low-dimension training samples is optimized with evolutionary computation followed by PSO (particle-swarm-optimization), and conversion algorithm is implemented with neural networks. The method has demonstrated its application in data visualization, lithological characterization, and petrophysical rock typing. Developed and implemented a method to construct computational-efficient surrogate model ensemble based on high-fidelity FEA/CFD simulation data, and utilized the generated ensemble in conjunction with evolutionary optimization to perfect engineering design of completion tools. Applications validated through simulations and lab testing in optimizing perforated hole patterns of the expandable well screens and geometric parameters of expandable liner hanger.

Completed feasibility and permanent magnet life studies for development of Halliburton DepthStar magnetic actuator for the use of subsurface safety valves. The product received 2004 Woelfel Best Mechanical Engineering Achievements, OTC Spotlight on New Technology, and Energy Institute Technology Awards.Developed a dynamic calibration method for quartz pressure transducers which can be used to generate adequate thermal transient data in the lab for neural network based temperature compensation design. This method is an extension of the standard static calibration routine, and can be implemented without any additional hardware.Developed a novel resistivity inverse modeling method with neural networks in processing high resolution array induction logs applied to dipping beds. A salient feature of this approach is its fast execution time (i.e., approximately 100 feet per second) which makes it ideal for real-time processing at the well site.

Developed a number of novel neural network training algorithms using the second derivative information of validation data, validation-set-based Bayesian regularization, and re-trained early stopping. Completed a solution book for neural network class which was published with revised text book.Conducted joint research with Cummins Engine on soft sensor design for engine emission control using knowledge based fuzzy rule and neural networks.Completed joint research with Halliburton on environmental correction of neutron logs and neural network inversion of induction logs.Completed velocity control of a DC motor using mixed programming languages (C and assembler), and sliding –mode based two-link control of a rotational inverted pendulum.Developed an algorithm in C for simulating the throughput of heavily loaded Ethernet local area networks (LAN), which had been used as teaching material on telecommunication class for several years. Proposed a research on ATM trunk sizing and user QoS for Frame Relay/ATM interworking .Applied sonic sensor technology to maturity measurement of agricultural product, co-authored several ASAE papers based on this study.Completed a number of signal processing projects on object recognition, target tracking, stereo vision and speech analyzer and synthesizer.