Wireless A/C compressor vibration diagnostics using machine learning-based signal analysis Z-Freq 2D with refrigerant and oil as faults

• Main Author: Yusri, Muhammad Yuszairie
• Format: Thesis
• Language: English; English
• Published: 2024
• Subjects: T Technology; TJ Mechanical engineering and machinery
• Online Access: http://eprints.utem.edu.my/id/eprint/29378/

Advanced diagnostic monitoring and fault detection in vehicle A/C systems are critical forthe automotive and A/C industries to accurately identify system anomalies and enableearlydetection of mechanical failures, particularly in compressor health and performance. Keycomponents of automotive A/C systems—compressor, condenser, evaporator, thermal expansion valve, and receiver drier—are essential for optimal functionality. Aprimaryfactor contributing to poor compressor performance is insufficient oil lubricant andrefrigerant R134a. This study aims to develop the Z-Freq 2D coefficient as a novel statistical method for detecting faults in vehicle compressors by analyzing vibrationdatainfluenced by refrigerant and lubricant levels, using a wireless diagnostic approachandvalidating findings through machine learning, simulation, and experimental testing. Fault conditions were simulated by varying the speed of the compressor, refrigerant amounts, and lubricant volumes. Vibration data was collected using a PhantomVibrationSensorattached to the compressor of a Myvi 1.5L X vehicle with a registered air conditioningsystem. Data analysis was performed using MATLAB, where the Z-Freq 2Dcoefficient was applied to generate graphical representations and validate results using machinelearning models, specifically Support Vector Machine (SVM) and k-Nearest Neighbors(kNN). The experimental parameters included compressor speeds ranging from750to2000 RPM, refrigerant levels from 280g to 360g, and lubricant volumes from40ml to120ml. Industry-recommended benchmark values were 320–330 g of refrigerant and80–90ml of lubricant. Results indicate that the Z-Freq 2D coefficient, combined withthePhantom Vibration Sensor, effectively identifies compressor faults. The SVMmodel outperformed kNN, achieving 87.1% accuracy and 98.6% sensitivity, compared tokNN’s82.9% accuracy and 88.6% sensitivity. Additionally, an increase in compressor RPMresulted in higher Z-Freq 2D data distribution, correlating with elevated vibrationlevelswhile excluding noise from the vehicle frame. The study also highlights a limitationinthewireless diagnostic method, which depends on stable network connectivity for transmittingdata to cloud-based platforms such as DigivibeMX or EI Analytic. The findingsdemonstrate the reliability of the Z-Freq 2D coefficient as a diagnostic tool for fault detection in automotive compressors. This method, validated through experimentationandmachine learning, offers significant potential for enhancing the accuracy of HVACsystemdiagnostics. The research underscores the importance of maintaining optimal refrigerant and lubricant levels to ensure compressor efficiency and overall systemreliability.