The goal of this project is to develop an autonomous, real-time methane leak detection technology, the Smart Methane Emission Detection System (SLED/M), which applies machine learning techniques to passive optical sensing modalities to mitigate emissions through early detection. Phase 1 leveraged previous Southwest Research Institute (SWRI) research and experience to develop the prototype methane detection system with integrated optical sensors and the embedded processing unit. Phase 2 focused on integration and field-testing of the prototype system along with a demonstration to the Department of Energy (DOE) within a representative controlled environment. Phase 3 focused on adapting the system developed under previous phases for use on a mobile aerial drone platform. Phase 4 is focused on quantifying detected methane and building a commercialization pathway. The system will target the following key features:

Table 1. Key Features of the Smart Methane Emission Detection System

SwRI, San Antonio, TX 78238
Falcon Inspection (camera in-kind contributor)
IRCameras (camera in-kind contributor)
FLIR Systems, Inc. (camera in-kind contributor) 
Sierra-Olympic Tech. Inc. (camera in-kind contributor) – Phase 3 and 4
Heath Consultants (camera in-kind contributor) – Phase 4

While much has been accomplished in developing optical sensors for imaging methane leaks, limited work has been accomplished in developing methane emission detection technologies that meet five key critical criteria for effective methane emission mitigation:

• Autonomy (no need for a human to be in the loop)
• High-reliability (low false alarm rates)
• Real-time performance (ability to detect emissions within a few minutes once it starts)
• Extensibility to multiple platforms (stationary, mobile, ground, aerial)
• Quantification estimation

This project focuses on these five key aspects and significantly advances the state-of-the-art in methane emission detection using optically based sensing technologies.

An autonomous, real-time methane leak detection system facilitates the early detection of emissions before they become a larger problem. Compressor station operators will be able to identify failing equipment in aging infrastructure and replace faulty components expediently, resulting in methane emissions being reduced significantly through early detection of non-compliant equipment. By adding the capability to estimate leak flow rates in conjunction with visual inspections, operators will be able to identify and stratify which components to replace first. This project produced the following outcomes and/or impacts:

• A system that reliably, accurately, and autonomously identifies methane leaks at critical midstream sections of the natural gas distribution network in real-time for the purpose of mitigating methane emissions
• A system that adds a high degree of automation to the process of methane leak detection to minimize sources of human error, minimize response time to a leak event, and maximize midstream visibility
• Assists in the quantification process by providing a means of collecting temporal and spatial image data of a leak event
• A system the provides operators with leak detection capabilities as well as quantification estimation capabilities in a single sensor
• Reduces operational costs of emissions detection technologies by significantly minimizing the need for operator involvement
• Provides a solution that is scalable, cost-effective, and non-intrusive
• Reduces methane emissions through early real-time, autonomous detection of methane leaks

• SLED/MQ achieved the following metrics over test and validation data
  • Detection precision of 96.6%
  • Detection false positive rate of 2.22%
  • Quantification average percentage prediction error: 12.3%
  • Quantification accuracy ±50scfh: 97.78%
• SLED/MQ Phase 4 Algorithm Delivered
• The SLED/M system went through several adaptations in order to enable it to detect and quantify fugitive methane emissions at higher fidelity.
• One (1) full week of data collection took place, recording methane emissions with known leak rates, and concentration profiles, through a variety of weather conditions using the FLIR G300a in conjunction with LwIR, LiDAR, Visible cameras.
• SLED/M showed initial quantification algorithm results, with the capability accurately categorize a methane flow above or below 100 scfh 82% of the time.
• One (1) full week of data collection took place, recording methane emissions with known leak rates, and concentration profiles using the FLIR G300a and Heath Consultants EyeCGas 2.0, in conjunction with LwIR, LiDAR, Visible and methane concentration sensor for ground truth.
• SLED/M Phase 3 Algorithm delivered.
• The SLED/M model is going through several adaptations in order to enable it to detect methane from a drone platform.
• The SLED/M pipeline is going through adaptations to enable enhanced real-time performance from an embedded platform.
• Several data collection events took place using a DJI Matrice 600 as the drone platform and the FLIR G300a and the Sierra Olympic Ventus Optical Gas Imagers (OGIs).
• Planning started for the integration and deployment of SLED/M onto a drone platform.
• SLED/M Phase 2 Algorithm refined and delivered.
• Field testing is being conducted.
• Data has been collected for the final round of testing in order to establish the limits of the SLED/M detection capabilities.
• The SLED/M system went through initial testing and successfully detected methane leaks from the live Mid-Wave Infrared (MWIR) camera stream in real-time.
• The embedded detection algorithm (running on an NVIDIA Tegra TX1 chip) was successfully integrated with the real-time MWIR camera stream.
• The algorithm passed all initial laboratory validation tests with target performance measures.
• The embedded system was verified to drive the cameras at high framerate and execute the detection algorithm in real-time (~1Hz).
• Sensor (cameras) setup, configuration, and calibration was completed
• Initial test data were labeled for classifier training. An initial convolutional neural network (CNN) was configured for training.
• The following equipment for the budget period was ordered and received: MWIR camera from IRCameras, MWIR camera from Falcon Inspection, and a thermal camera (SwRI in-kind).
• Tests and data acquisition for budget period 1 are underway.
• Feature extraction and analysis tasks were completed.
• Development of initial leaks and non-leak events detection algorithm was completed.

The final report for this project can be viewed at https://www.osti.gov/.

$1,811,107

$453,057

NETL – Joseph B. Renk III (joseph.renk@netl.doe.gov or 412-386-6406)
SwRI – Heath Spidle (heath.spidle@swri.org  or 210-522-6717)

Smart Methane Emission Detection System Development (Oct 2020)
Presented by Heath Spidle, Southwest Research Institute, Natural Gas Infrastructure Project Review Meeting, October 28, 2020