Development of a fault diagnosis device for pulsator in milking machines

In this thesis, a low-cost and standalone system was developed to detect real-time faults in electronic and pneumatic pulsators used in milking machines by utilizing the ADXL345 accelerometer sensor and embedded artificial intelligence (Edge AI) technology. At the initial stage, vibration data generated during pulsation cycles were collected using the ADXL345 sensor. These signals were sampled at 100 Hz via the ESP32-S3 microcontroller and transferred to the Python environment for processing. Time and frequency domain features such as RMS, peak amplitude, and zero-crossings were extracted from the raw data, and an XGBoost-based anomaly detection model was trained. The model was then converted into TensorFlow Lite format and deployed on the ESP32-S3 device. The embedded system analyzes pulsation cycles in real time, distinguishing between normal (60 pulses/min) and abnormal (e.g., 55 or 65 pulses/min) operations. The developed solution demonstrated high accuracy and responsiveness during testing, and its wireless, self-contained design proved suitable for use in milking parlors. This work offers an innovative and practical approach for early fault detection in livestock machinery and provides a flexible foundation for similar applications.