The purpose of this paper is to illustrate the application of neural network approach to analyze machine's behaviour quantitatively.
The model is developed based on real plant‐data from a variable speed drive centrifugal type pump. The best model settings are recorded and tested for another similar unit in the vicinity to check its generalization capabilities. Owing to the absence of faulty data, this model is tested against preventive maintenance data that show symptoms of abnormality that are seemingly undetected in existing monitoring and control systems. The paper systematically summarizes published literature and suggests suitable network architecture and its capabilities by illustrative example from oil export pumps from an oil and gas offshore production facility.
Artificial intelligent techniques provide a robust platform in providing useful information about system health and sub‐optimal performance.
In any industry, unexpected equipment downtime in principal questions the overall technical integrity of the platform raising major economical concerns. In the Oil & Gas sector, production platforms are in a 24/7 run mode, and thus undergoing major re‐engineering processes by improving existing surveillance and control techniques of their asset. Machine degradation and abnormalities gradually affect performance and in some cases these are not visible in existing condition monitoring (CM) schemes. Recently, there has been an increasing demand for testing and implementing intelligent techniques as a subsidiary to existing CM programs to monitor and assess system's health. Artificial neural networks have emerged as one of the most promising technique in this regard.
The proposed methodology highlights how healthy data from a system can be effectively modelled to identify significant abnormalities. This paper will be useful for experts working in the area of maintenance engineering to early identify state of the system performance.
