Machine learning-based prognosis for valves in Carbon Capture applications

In this paper a possible approach based on machine learning techniques is proposed to prevent failures in electro mechanical valves, which are usually used in carbon capture applications. These systems will be essential in the near future to reduce air pollution and the effects of climate changes. The main objective of this work is to propose a classification method capable of preventing catastrophic failure by processing the signals acquired by two different sensors: uniaxial accelerometer and pressure sensor. The identification of the valve status allows the estimation of the residual useful life and, consequently, the organization of the maintenance operations, thus avoiding critical situations and reducing recovery times. To achieve this goal, two unsupervised machine learning methods are proposed and experimental measurements extracted from a real testbench are processed. The whole procedure can be divided into two steps. At first, machine learning methods must distinguish measurements extracted from valves under nominal conditions from those obtained after artificial degradation of two components. The classification results obtained are in the range of (95-100)% depending on the level of opening applied to the valve. Subsequently, the possibility of recognizing the progressive degradation of the valves due to their use is investigated.