Pipeline Leak Detection Using Pressure Patterns and IoT Gateways
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Upstream pipeline networks lack effective, cost-efficient leak detection methods due to the absence of SCADA/telemetry infrastructure, operational complexities, and challenges from emulsions, bi-directional flow, and thermal changes, making conventional methods like CPM and RTTM impractical.
Innovation Solution
A system utilizing IoT Gateways and cloud-based recurrent neural networks (RNNs, specifically LSTM autoencoders) to analyze high-frequency pressure data from pressure sensors, extracting key parameters and reconstructing patterns to detect leaks through multi-threshold filtering, eliminating the need for costly instrumentation and infrastructure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional leak detection methods (CPM, RTTM) are used, then leak detection capability is improved, but device complexity and cost increase due to requirement of SCADA/telemetry infrastructure, flow meters, and pressure sensors
Solution Approach 1:
The patent extracts the essential leak detection functionality from complex SCADA/telemetry systems by using only pressure sensor data processed through edge computing devices and cloud-based machine learning models, eliminating the need for flow meters, SCADA infrastructure, and other expensive instrumentation
Solution Approach 2:
The patent creates a virtual model of pipeline behavior through machine learning algorithms that replicate the expected pressure patterns, allowing leak detection through comparison of actual versus predicted pressure data without requiring physical instrumentation at every pipeline point
2Measurement precision
If flow meters and pressure sensors are installed for real-time monitoring, then measurement precision is improved, but loss of energy and operational costs increase due to continuous power requirements and maintenance
Solution Approach 1:
The system uses existing pipeline pressure data that is already being collected for operational purposes, eliminating the need for additional powered instrumentation. The edge computing device processes this existing data locally, and the cloud-based model provides leak detection as a self-service function without requiring continuous energy input at remote pipeline locations
Solution Approach 2:
The patent introduces an intermediary cloud-based machine learning model that receives periodic pressure data transmissions and performs the computationally intensive leak detection analysis in the cloud, allowing minimal energy consumption at the pipeline site while maintaining high measurement precision through advanced algorithms
3Loss of information
If SCADA/telemetry communication is implemented at remote sites, then data acquisition capability is improved, but device complexity and project costs increase
Solution Approach 1:
The patent segments the data processing function into two parts: simple data collection and transmission at the remote pipeline site, and complex leak detection analysis in the cloud. This segmentation allows the use of simple, low-cost communication devices at remote locations while maintaining sophisticated leak detection capabilities through cloud-based machine learning models
4Area of stationary object
If multiple instrumentation devices are deployed across extensive pipeline networks, then coverage area is improved, but loss of time for installation and maintenance increases
Solution Approach 1:
The patent creates a universal leak detection system that can monitor multiple pipeline segments across extensive networks through a single cloud-based platform. The machine learning model is designed to handle data from multiple pressure sensors simultaneously, providing network-wide coverage without requiring separate instrumentation systems for each pipeline segment, thereby reducing installation and maintenance time
Data Source
AI summary
A computer implemented method is provided to detect pipeline leaks in a pipeline utilizing one or more sensor devices, each located adjacent the pipeline. The method comprises capturing, from each of the one or more sensor devices, discrete pipeline pressure measurements at discrete instances in time at each sensor location; extracting, by a processor of an IoT Gateways associated with each of the one or more sensor devices, a plurality of key parameters from each of the discrete pipeline pressure measurements; transferring, via the IoT Gateways, the plurality of key parameters to a cloud system; processing in the cloud system, the plurality of key parameters by using a recurring neural network (RNN) to calculate and reconstruct patterns of the key parameters; comparing the reconstructed patterns to historically collected patterns from pipeline operations to determine a reconstruction error in the reconstructed patterns; and evaluating the reconstruction error by applying multi-threshold filtering to enhance anomaly detection robustness to determine presence of a pipeline leak. A computer system and a nontransitory computer-readable medium containing program are also provided.