Machine Learning Subsurface Stress Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in predicting and warning of subsurface stress criticality and fluid-driven failures in terrestrial media, such as those occurring during hydraulic fracturing or geothermal energy extraction, due to difficulties in characterizing internal conditions and providing timely alerts for potential failures.
Innovation Solution
The use of machine learning models, trained on data from sensors and remote sensing measurements, to analyze signals from microscopic processes within the terrestrial medium, allowing for the estimation of macroscopic parameters indicative of future fluid-driven failures, including time-to-failure, type, size, and energy release, and to guide engineering remediation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional earthquake forecasting methods relying on catalogs of past events are used, then historical data can be analyzed to determine periodic behavior or historic event magnitudes, but the internal conditions of terrestrial media cannot be accurately characterized and failures cannot be predicted in real-time
Solution Approach 1:
The patent replaces traditional mechanical/catalog-based earthquake forecasting with acoustic emission monitoring and machine learning algorithms. Acoustic sensors detect microscopic signals from subsurface processes, and ML models analyze these signals to predict failures in real-time, enabling both accurate characterization of internal conditions and timely warnings.
Solution Approach 2:
The patent introduces acoustic emission signals as an intermediary between subsurface processes and surface observations. These acoustic signals serve as messengers that carry information about internal stress states and failure processes, allowing remote characterization of conditions deep within terrestrial media without direct physical access.
2Reliability
If early warning systems analyzing the first few seconds of seismic signals are used, then warnings can be provided during earthquakes, but the warning time is minimal (only a few seconds) and the system cannot distinguish between larger magnitude events
Solution Approach 1:
The patent applies preliminary action by detecting acoustic emission signals from microscopic processes that occur before major failure events. The system identifies precursory acoustic signals that indicate building stress and impending failure, providing advance warning time before the actual earthquake or failure occurs, rather than waiting for the seismic signal to arrive.
Solution Approach 2:
The patent substitutes traditional seismic wave analysis with acoustic emission monitoring and machine learning classification. The ML models analyze acoustic signal patterns to distinguish between different types of events and predict magnitude, providing both earlier warning and better discrimination capability than conventional seismic analysis.
3Measurement precision
If sensors are deployed to detect acoustic signals from microscopic processes, then real-time monitoring of subsurface stress can be achieved, but the complexity of the monitoring system increases
Solution Approach 1:
The patent applies universality by designing a multi-functional monitoring system where acoustic emission sensors serve multiple purposes: detecting different types of subsurface processes, characterizing stress states, predicting failures, and providing warnings. The machine learning models perform multiple analysis functions simultaneously, reducing the need for separate specialized systems.
Solution Approach 2:
The system applies self-service through automated machine learning classification that processes acoustic signals without requiring constant human intervention. The ML models automatically detect patterns, classify events, and generate warnings, reducing the operational complexity burden on personnel while maintaining high detection precision.
4Measurement precision
If machine learning models are trained on data from sensors and remote sensing measurements, then accurate estimation of macroscopic parameters can be achieved, but the amount of training data and computational resources required increases
Solution Approach 1:
The patent applies parameter changes by transforming acoustic emission signals into meaningful features and parameters that capture the essential characteristics of subsurface processes. The machine learning models work with these transformed parameters rather than raw sensor data, reducing the effective data volume needed while maintaining estimation precision.
Solution Approach 2:
The patent applies local quality by focusing training data and computational resources on the most critical and informative aspects of the monitoring data. Rather than processing all sensor data uniformly, the system identifies and emphasizes local patterns and features that are most predictive of failure, reducing overall data requirements while maintaining high estimation accuracy.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate prediction and timely warning of fluid-driven failures, surpassing conventional methods by utilizing machine learning to classify acoustic and other signals from the subsurface, thereby mitigating physical damage and enhancing safety in subsurface engineering activities.
Implementation Method 1
Signals representative of microscopic processes within the terrestrial medium are received
Data Source
AI summary
Machine-learning methods and apparatus are disclosed to determine critical state or other parameters related to fluid-driven failure of a terrestrial locale impacted by anthropogenic activities such as hydraulic fracturing, hydrocarbon extraction, wastewater disposal, or geothermal harvesting. Acoustic emission, seismic waves, or other detectable indicators of microscopic processes are sensed. A classifier is trained using time series of microscopic data along with corresponding data of critical state or failure events. In disclosed examples, random forests and artificial neural networks are used, and grid-search or EGO procedures are used for hyperparameter tuning. Once trained, the classifier can be applied to live data from a fluid injection locale in order to assess a frictional state, assess seismic hazard, assess permeability, make predictions regarding a future fluid-driven failure event, or drive engineering solutions for mitigation or remediation. Variations are disclosed.