Gassonic Observer i Ultrasonic Gas Leak Detector with ANN
Gassonic i Ultrasonic Gas Leak Detector
The Variable Background Noise Problem
The Artificial Neural Network Solution in Acoustic Gas Leak Detectors
To overcome these limitations, we have developed a breakthrough intelligent UGLD based on artificial neural network (ANN) intelligence. This new UGLD design distinguishes the broadband ultrasound produced by pressurized gas releases from other artificial and natural sources. Such design can be implemented with or without a threshold-crossing scheme, enhancing the scalability of gas leak detection. For convenience, the detector can operate with the threshold – crossing scheme or one of several enhanced modes that rely on ANN to classify ultrasound signals.
Furthermore, testing results show that the UGLD with the ANN algorithm achieves a longer detection range(radius ~ 28 m) and a shorter response time in the presence of ultrasonic background noise.An artificial neural network operates in a manner very similar to how the human brain handles the constant flow of information.
When we meet a person, the brain receives a massive amount of visual information through the eyes, and over time this substantial amount of information is used to recognize this person years later or even to identify further family members. When the brain has received visual information about other family members, it is easier for it to distinguish between family and non-family members.
In other words, the more we train our brain to recognize familiarity the better we will be able to recognize or deny a person’s face. The brain does not look for an exact match, it looks for familiarity, and so does the ANN.
Like the brain, however, the neural network needs to be trained first. A UGLD does not have to recognize different people.It needs to recognize the sound signature from a gas leak effectively while at the same time rejecting sound signatures from acoustic background noise that are not related to gas leaks.
Acoustic noise from a real gas leak source normally ranges from 10kHz and up to the 60 to 70 kHz range. Acoustic noise from a false alarm source, such as a gas compressor for example, can easily generate high level frequencies in the range of 100Hz to 20 kHz.
Earlier generations of UGLDs were designed with electronic filters to screen out and ignore noise below 20 kHz, which eliminated false alarms resulting from most normal plant background noise such as gas compressors, but also limited detection of smaller real gas leaks.
With most real gas leaks generating a sound typically with a frequency of 10 to 60 Khz, detecting a higher frequency range of 75 kHz or greater doesn’t add any meaningful capability.
Adding ANN intelligence capability to UGLD’s provides process and plant engineers with a new tool that can help them detect more real gas leaks and avoid nuisance false alarms. The OBSERVER-i UGLD with its capability to reliably detect leaks below 20 kHz provides another layer of safety to the industry, protecting people, equipment and facilities from hazardous gas leaks.
Reference: MSA White Paper, UGLDs with Artificial Neural Network