26 - Applications of AI and possibilities for process control

Abstract

Multiphase flows occur in the oil and gas industries, e.g., gas/oil/water/sand (g/o/w/s) in all possible combinations, drilling mud with cuttings and g/o/w/s, in the storage and transport of wet or dry particulates, e.g., fluidized beds, slurries and sedimentation, e.g., as in dredging, in the nuclear power industries, e.g., entrained air, and steam at supercritical temperatures in cooling water in pressurized water reactors. Most of these processes have different flow regimes with varying distributions of the different materials/phases, flowing at different speeds and spread over the cross-section of the conduit supporting the flow, an important topic in CFD studies and software development. These processes are monitored with a plethora of sensors, continuously gathering vast amount of valuable data from various locations with many control loops distributed in the processes with a dedicated overall process control using different strategies, which recently have AI and machine learning techniques in their portfolio. The data from the sensors are valuable in data fusion not only for deterministic mechanistic modeling but also for exploratory data analysis (EDA), a growing branch of AI-based industrial machine learning. Data from process tomography/tomometry using nonintrusive and noninvasive sensing provide big data in real time, useful in identifying various flow phenomena, such as flow regimes, tunneling flow in silos, infiltration of sand in pipes, unusually high presence of gas bubbles in cooling water, etc. In this chapter, some applications of ECT and EIT in the above three process industries are presented. This chapter focuses on the sensor arrays and protocols used along with excitation and sensing methods used in the ECT/EIT modules and then presents some results from EDA as applied in the growing field of industrial machine learning. In all the three branches of process industries mentioned above, interesting results are presented showing possibilities of flow regime identification based on the distribution of the phases involved with the possibilities of integrating ECT/EIT in model free adaptive control of these processes.