ÃÛÌÒÊÓƵ

LU Computer Science and Industrial Engineering Faculty Secure Funding to Develop a Real-Time Preventative Maintenance Auditory System for Natural Gas Compressors

Dr. Jing Zhang, Associate Professor of Computer Science, and Dr. Maryam Hamidi, Assistant Professor of Industrial and Systems Engineering, have received funding for their project “Deep Learning-based Auditory Anomaly Detection and Classification for Natural Gas Compressors, Phase 1.” The project has been funded by ÃÛÌÒÊÓƵ University’s Center for Midstream Management and Science (CMMS) for a total of $50,000, and this funding was recently augmented by $40,000 by Well Checked Systems and CMMS.

jing-zhang-and-maryam-hamidi
Dr. Maryam Hamidi and Dr. Jing Zhang


“By pressurizing natural gas in pipelines, the compression system connects upstream gas production and downstream consumer use,” explained Dr. Hamidi. “Considering the installation cost of $1 million to $2 million for a compressor, failure of the equipment can be very expensive, both in repair and lost production.”

Traditional inspection for a compressor by human ear is challenging, given that the decibel level of a compressor exceeds 150 decibels, which is louder than a jet engine. Often, the excessive levels of noise prevent detection of new or abnormal sounds by the human ear. Drs. Hamidi’s and Zhang’s research aims to facilitate the development of an automated real-time preventive maintenance auditory system for natural gas compressors.

In the previous phases of this project, also sponsored by CMMS, compressors’ audio data, provided by Well Checked Systems International in collaboration with a large well-known gas producer, was successfully analyzed by the LU research team. They used the supervised and unsupervised training audio data to develop deep learning-based models for compressor anomalies.

The research team will continue collaborating with Well Checked Systems to extend the model in additional R&D phases. In these phases, the network performance will be improved by producing a long-term anomaly classification network that has an integrated feedback loop, on top of the anomaly detection method developed, to collect, classify, and retain anomalous audio data. Classifying sound anomalies will help the users to know the type of anomaly and problem indicated, enabling them to pursue required maintenance work.

Well Checked Systems International is developing automated solutions to monitor and identify failure anomalies of midstream compressors and components using data of microphones, installed on compressors. The analytics use a matrix of audio data to predict anomalies and alert operators to preventively maintain the component before it fails.

The research is also used in the development of a new Reliability course INEN 5305, as a part of a new CMMS certificate.