Your web browser does not support JavaScript, but it does not affect browsing through the rest of the web site.

ITRI Introduces Gas Leakage Automatic Recognition Technology (GLART), World’s First Auto-Recognition Software for Gas Leakage Detection

Font size:
Small
Medium
Large

ITRI Introduces Gas Leakage Automatic Recognition Technology (GLART), World’s First Auto-Recognition Software for Gas Leakage Detection

ITRI introduces Gas Leakage Automatic Recognition Technology (GLART), the world’s first and only auto-recognition software for gas leakage detection. GLART significantly improves the detection rate and shortens the inspection time especially of minor gas leaks when compared to manual inspection, thereby reducing the risks of plant pipeline disasters. GLART detects minor leaks with 85-97 percent accuracy, and can be combined with robots, tracks or unmanned aerial vehicles (UAVs) for fully automated inspections. GLART software is portable and may be applied to infrared (IR) thermal images created by any leakage detection device; therefore, GLART can enhance the leakage inspection accuracy and speed of existing IR thermal imagers even if operated manually. GLART is available for licensing, and currently is a finalist for the 2018 R&D 100 Awards in the United States.

 

“Early detection of small leaks is crucial for the prevention of accidents; even minor leaks of a few grams of hydrocarbons per hour can lead to disastrous results if an ignition source is present. A large-scale industrial trial of GLART for pipeline inspection is currently underway at a petrochemical plant in Taiwan. The deployment of GLART technology will drastically improve public and industrial safety worldwide and save thousands of lives each year,” said Dr. Ming-Shan Jeng, Division Director of ITRIs’ Green Energy and Environment Research Laboratories.

 

Before ITRI developed GLART, there was no good solution for minor leakage detection. Most conventional leakage-monitoring devices that monitor pipeline pressure, flow rate, or acoustic signals are unable to detect small leaks. These fixed devices work well only when the leakage rate exceeds one percent of the total transport amount. Manual labor can detect small leaks, but it is far too slow for large-scale plants and pipelines, and often results in delayed discovery of leaks. When using only an IR camera for manual visual inspection of pipelines, operators often miss smaller leaks due to visual fatigue under long working hours. Small leaks are the main target of inspection and missing them poses a serious threat to pipeline safety. Manual inspection of elevated pipelines for minor leaks is especially laborious, but GLART, when combined with UAVs, eliminates the inspection obstacles for elevated pipelines. Most pipeline gases feature special IR absorptivity signatures that GLART recognizes; with GLART, minor leaks of pipeline gases, formerly not detectible by conventional leakage-monitoring devices or by manual methods, can now be identified with high precision.

According to statistics in petrochemical plants, the detection accuracy rate for manual inspection using an IR camera is 50 percent; GLART increases that detection rate to between 85 and 97 percent while significantly decreasing inspection time. When combined with plant robots, tracks, or UAVs, inspection time may be shortened by more than 90 percent. For example, all valves of a small petrochemical plant with 100,000 valves could be inspected for leaks in approximately five calendar days if GLART is combined with robots, tracks, or UAVs operating continuously, rather than the 100 to 500 working days required using only manual inspections. GLART dramatically increases the possible scope and frequency of inspections in addition to dramatically reducing inspection time.

How GLART works

GLART utilizes machine learning and image processing technologies for auto-recognition of minor gas leakage. GLART enhances the leakage image and greatly extends the lower limits of IR detection. Its algorithm not only improves the inspection effectiveness of IR cameras but also enables better operation of drone-carried IR cameras by enhancing the accuracy of long-distance leakage detection. GLART consists of two parts: gas leakage IR image enhancement technology, and gas leakage auto-recognition technology.

Gas leakage IR image enhancement technology

The gas leakage IR image enhancement technology utilizes an image stabilizing compensation method to solve jittering issues in recording. Unlike images from color cameras, grey-scale IR images recorded by a moving IR camera are full of noisy signals, making them more difficult to process. GLART solves this problem by identifying multiple moving features in the images to determine a final motion vector. After that, images are analyzed pixel-by-pixel for differences among timewise consequential images. These differences are weighted over multiple frames, filtered to exclude abnormality, and treated by various feature-dependent transfer functions. The results are superposed on the original image. Even minor traces of leakages become identifiable after image enhancement. The sensitivity of the detecting limit depends on the combined effect of the specially designed weighting functions, filtering methods, and feature-dependent transfer functions, which have proved to be extremely effective in gas leakage detection once the system is well tuned. On average, two-thirds of originally invisible leakage images become apparent after applying ITRI’s enhancement algorithm. 

Gas leakage auto-recognition technology

To make gas leakage detection fully automatic, ITRI developed an image auto-recognition technology to replace manual technologies for gas leakage detection. Because IR image features for gas leakage are extremely diverse and conventional image recognition techniques are not suitable, GLART adopts a unique approach. By analyzing numerous IR leakage images compiled by ITRI, ITRI’s R&D team found gas leakages can be described using seven image characteristics derived from grey pattern and texture analysis. Based on this finding, these seven characteristics are used in training seven machine learning models individually. Furthermore, the outputs of the seven models are treated by an ensemble and regression process to generate the final output. The core of recognition under this scheme depends on the characteristics derived solely from static images. With enough learning, usually a single frame can provide sufficient information for auto recognition of gas leakage in GLART’s detecting system. This is particularly useful during real-time auto inspection when the camera needs to move quickly. In practice, however, dynamic motion analysis is added in the end to further ensure the robustness of the algorithm and reduce false alarms. In a gas leakage where manual visual IR image inspection has a detectability of 50 percent, GLART’s auto-recognition algorithm can improve the overall detectability to 85-97 percent after video database training.

GLART (Gas Leakage Automatic Recognition Technology) video.