A Novel Online Detection Method of Linear Refrigeration Compressor Stroke Based on Artificial Neural Network

Abstract Linear compressors offer excellent capacity control, enhancing system efficiency and adapting to diverse operating conditions. However, achieving efficient and reliable operation requires addressing two main issues: nonlinear gas forces causing resonant frequency variations and unrestricted piston strokes potentially causing collisions with the discharge valve. Addressing these issues requires obtaining the value of the piston stroke. Measuring piston stroke without displacement sensors can reduce costs and facilitate stroke control. This paper presents an innovative artificial neural network (ANN) method for online piston stroke measurement. Unlike other methods, the proposed model does not assume constant key parameters and requires only six easily obtainable inputs: voltage, current, frequency, suction pressure, and discharge pressure, making it highly scalable. Results indicate an average R 2 of 0.931, an average RMSE of 0.514, and an MAPE of 2.59% on the testing set. Finally, a simple stroke adjustment method based on the ANN model is proposed, enabling stroke control and natural frequency calculation in linear compressors. This method enhances the efficiency of refrigeration systems and adapts to diverse operational scenarios.