Classification of washing machines based on vibration sensor signals: towards predictive maintenance of household appliances

Integrating machine learning into predictive maintenance could be a milestone towards better predicting future machine failures. This is particularly true for household appliances such as washing machines, which suffer from complex and costly failure causes (dampers, bearings, electronics, etc.) that lead to these machines being scrapped. In this project, the results are based on a non-intrusive sensing module capable of collecting vibration data at a sampling rate of between 100 Hz and 120 Hz. The device is based on the combination of a microcontroller (FireBeetle-ESP32) with two accelerometer sensors (ADXL345 and MPU6050). This master’s thesis proposes strategies for influencing the acquisition sampling rate, for calibrating such sensors and for processing the classification using machine learning. The problem studied is the classification of two washing machine models, using simple classifiers such as Naive Bayes, Decision Tree, K-Nearest Neighbors and Support Vector Machine. Although working with a small dataset of cycle records, the approach described in this project provides encouraging results for future development.