Machine Learning for Monitoring of the Solenoid Valves Coil Resistance Based on Optical Fiber Squeezer

Abstract

Solenoid valves represent indispensable elements in various engineering systems. Their failure can lead to unexpected problems. This failure may be caused by fluctuations in the coil resistance of the electromagnetic solenoid (EMS) which actuates these solenoid valves. Hence the need to monitor this parameter for a preventive maintenance of these actuators. The proposed method consists to use supervised machine learning to monitor coil resistance of the EMS valve. The EMS valve is coupled to an optical fiber squeezer which, acts as a force sensor. The solenoid armature applies a mechanical force to the optical fiber and changes the polarization state of the light that travels through the optical fiber and then this force infects the power of the light. A Simulink model is used to determine the open loop system step response. The identification of the system allows obtaining its transfer function, which depends on the parameters of the EMS and in particular on its coil resistance. By varying the coil resistance while fixing the other physical parameters of the EMS, we generate a database whose elements are the coefficients of the transfer function of the solenoid open loop and the electrical resistance of its coil. The generated database is used for training several supervised machine learning models whose predictors are the elements of the transfer function; the response is the coil resistance. The Gaussian process for regression allows to predict the variations of the coil resistance with the smallest relative error although it takes a relatively long time for the training compared to the other models used.