Hybrid Pressure Anomaly Detection With Physical-Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Accurate detection of pressure abnormalities in drilling equipment is hindered by unreliable and expensive monitoring systems, prone to errors and requiring extensive data sets that are impractical or cost-prohibitive to obtain.
Innovation Solution
A hybrid machine learning approach using a limited set of training data supplemented by a physical model to train a machine learning system for predicting pressure measurements, allowing for quicker training and real-time estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive training data is used to train machine learning systems for pressure prediction, then prediction accuracy is improved, but data collection cost and time requirements increase significantly
Solution Approach 1:
A physical model serves as an intermediary to generate supplemental training data. The physical model takes operational parameters as input and outputs predicted pressure measurements, filling the gap between available sensor data and required training data. This mediator enables the system to train with limited real sensor data while maintaining prediction accuracy.
Solution Approach 2:
The patent merges machine learning models with physical models to create a hybrid system. The ML model processes real sensor data while the physical model provides supplemental training data. This combination allows the system to achieve high prediction accuracy without requiring extensive real sensor data, as the physical model compensates for data deficiencies.
2Reliability
If extensive training data is collected to improve model accuracy, then prediction reliability is improved, but system cost increases due to monitoring infrastructure requirements
Solution Approach 1:
The physical model acts as a cost-effective intermediary that generates training data without requiring expensive monitoring infrastructure. Instead of deploying extensive sensor networks to collect training data, the system uses the physical model to synthesize realistic pressure scenarios, significantly reducing infrastructure costs while maintaining data quality for reliable training.
Solution Approach 2:
The physical model creates virtual copies of real pressure measurement scenarios. By simulating various operating conditions and anomaly patterns through the physical model, the system obtains training data that replicates real-world conditions without needing to physically monitor every scenario, thereby reducing system complexity and cost.
3Measurement precision
If traditional monitoring systems are used for pressure detection, then measurement capability is provided, but system reliability deteriorates due to errors and false alarms
Solution Approach 1:
The system incorporates feedback mechanisms where predicted pressure values from the hybrid model are compared against actual sensor readings. This feedback loop allows the machine learning model to continuously refine its predictions, correcting errors and reducing false alarms by learning from discrepancies between predicted and actual measurements, thereby improving monitoring reliability.
Solution Approach 2:
The patent creates a composite monitoring system combining machine learning algorithms with physical models. This composite approach leverages the strengths of both methodologies: the ML model provides adaptive learning from data patterns while the physical model ensures physical consistency. The combination mitigates the weaknesses of individual approaches, improving overall system reliability by cross-validating predictions.
Data Source
AI summary
Methods, computing systems, and computer-readable media for training and using a machine learning system to predict equipment pressure measurements, of which the method includes inputting a set of training data including a first set of equipment pressure measurements, inputting a set of supplemental data. The supplemental data is obtained from a physical model that estimates a second set of equipment pressure measurements. The method includes training the machine learning system based on the set of training data and the set of supplemental data to generate a trained machine learning system, receiving real-time operational data, inputting the real-time operational data into the trained machine learning system, predicting a real-time equipment pressure measurement based on the inputting the real-time operational data into the trained machine learning system, and executing a computer-based instruction based on the predicting the real-time equipment pressure measurement.