Gas turbine availability improvement based on long short-term memory networks using deep learning of their failures data analysis

Practically, a maintenance operation is performed on industrial equipment after scheduled planning that depends on the average useful life of this equipment (Mean Time Between Failures or Mean Time to Failure). Hence, in the industry, the use and the processing of data certainly improve productivity. But they induce a complexity of the industrial system caused by the different misconduct and measurements. This requires significant expenses on the safety, reliability, and availability of this type of machine. In this work, a new approach is proposed to determine the degradation indicators of a GE MS 5002B gas turbine installed on the Hassi R'Mel gas field in southern Algeria. The proposed approach is based primarily on Long Short-Term Memory LSTM networks, using in-depth learning of operating data. We are starting with the study of their reliability and their prognosis to validate and improve their performance, by optimizing their life cycle costs through good operating, repair, and maintenance planning. The objective is to remedy the problems mentioned by the processing of conventional data and predict their evolution and progression during the lifetime of the examined turbine. By combining actual reliability tests with predictions based on their failure rates to ensure good operating safety, and availability of the turbine system by controlling aging and degradation indices with satisfaction in environment and yield of this rotating machine.