To enhance the reliability of solenoid valves across diverse applications, condition monitoring strategies are essential. Conventional monitoring approaches often rely on detailed physics-based models [1], which involve multiple uncertain parameters and frequently overlook various failure mechanisms. Alternatively, data-driven methods [2] may lack interpretability and can be sensitive to different operating conditions. This paper presents an innovative approach that combines physical parameters, identified online, with machine learning techniques. Specifically, an equivalent resistance and inductance are derived through an online parameter identification scheme, formulated as a least squares problem. These identified parameters serve as inputs to a classification Feed Forward Neural Network (FFNN), enabling accurate predictions of the solenoid valve's flow quality. Our dataset comprises 48 valves subjected to an accelerated lifetime test over 6 weeks, equivalent to approximately 1150 hours. The valves are switched on and off at a rate of 1 Hz. Flow values are sampled once every minute, resulting in 69000 data points per valve. For each minute, 10000 current and voltage measurements are collected to identify the equivalent electrical parameters. We investigate various types of classification architectures, varying in complexity, and evaluate their accuracy and performance. Our final implementation is a FFNN with one hidden layer with 10 nodes and achieves a remarkable prediction accuracy of 96.2%, which is comparable to other leading methods in the field [3]. Looking ahead, we aim to correlate the documented failure mechanisms of solenoid valves with variations in the identified parameters. Additionally, we will quantify the uncertainty in predicted flow quality, which is rooted in the uncertainty of these identified parameters, by employing a total least squares approach.