The noise from the machinery is often repetitious, while the signals generated by the explosives tend to be impulsive in nature.Working in MATLAB®, Schultz developed an adaptive, predictive nonlinear neural network filter that cleanses the signals of the contaminating repetitive noises, leaving only the impulsive components—which include the signal generated by the subsurface explosion.
A MATLAB user for the past eight years, Schultz knew that MATLAB was the best tool for this project: “The real beauty of MATLAB is that you can do really fast matrix manipulation. Neural networks are formulated in terms of matrices, so it’s a perfect fit. Not only that, but almost any math tool that you want to use is right there. It’s a really wonderful instrument.”
He based his neural network code on models included in Deep Learning Toolbox™. To develop the filtering algorithm, he took data files, digitized them, and used MATLAB to perfect the structure for the neural network.The interactive MATLAB environment made this fine-tuning easy. He was then able to create a standalone application that could be used on a PC at the well site.
Schultz relied on MATLAB Compiler™ to quickly compile and execute the application on the desktop. Before he used MATLAB Compiler, he recalls, “getting the algorithm working was only the first step. In order to create an application, I would then have to retrace the functions I’d used, start a Visual Basic® program, type in code, and install debugging software. With MATLAB and MATLAB Compiler, I can devote more of my time to fine-tuning the algorithm.”
He adds, “It’s really significant that I can take the math functions that are available in MATLAB and compile those into a complete graphical program that includes user interfaces and plots as well as math functions. In fact now I don’t worry about writing programs in Visual Basic or C. I haven’t written a C program in months!”
Once he had an executable program, Schultz was able to use the sound functions in MATLAB to play the filtered signal over the sound system on his computer at the well site. He explains, “I record the data in the noisy environment, bring it back to my office and filter it with my filtering program, then listen to it using the sound function in MATLAB.” This capability has proved particularly useful when—as is often the case—he wants to hear something in a noisy environment.
Following successful trial tests, the adaptive neural network filter is being used as the basis for other projects using adaptive neural networks, and Halliburton has initiated patent protection for the technology.