Detekce anomálií vzduchotechnických systémů na mikrokontroléru

Abstract

This diploma thesis deals with anomaly detection in air handling systems on a~microcontroller, with a~focus on an~axial fan driven by a~permanent magnet synchronous motor (PMSM). The~proposed approach is based on internal variables available in the~sensorless motor control structure and examines their suitability for diagnosing operating deviations and fault conditions. In the~practical part, data acquisition was carried out on a~laboratory setup consisting of a~PMSM motor and an~axial fan under several operating conditions representing normal operation as well as selected loaded or faulty states. The~measured data were subsequently analyzed in both the~time and~frequency domains, with emphasis on identifying diagnostically significant changes in the~internal electrical variables of the~drive. Motor Current Signature Analysis (MCSA) was also used for frequency-domain assessment. For anomaly detection, machine learning models were designed and~trained using both the~NXP eIQ Time Series Studio tool and~a~custom solution implemented in Python. The~selected models were then exported to the~C language, integrated into the~control firmware, and~tested directly on the~microcontroller during system operation. The~thesis also includes a~graphical visualization of the~model outputs for tuning and~monitoring the~model behavior during testing. The~results showed that the~internal variables of a~sensorless PMSM drive carry sufficient diagnostic information to distinguish between normal and~anomalous fan operation and~that the~proposed models can be executed directly on a~microcontroller.