Automating tool for fault analysis of washing machines : towards predictive maintenance of home appliances

The number of disposed washing machines nowadays increases at an alarming rate due to their repair complexity. Therefore, the need for efficient and automated fault analysis and predictive maintenance systems becomes crucial. For now, only a handful of projects are trying to resolve this matter but most of them are using complex or expensive systems. Those setups are not efficient for neither the collection of large datasets nor the deployment. This is where this master thesis provides a physical tool to facilitate such a process. Using accessible and low-cost hardware, this work aims to give a tool to allow further projects to implement machine learning for failure analysis and predictive maintenance. The device developed is based on a microcontroller and an additional microcomputer acting as a server to store the data captured by two inertial measurement units. Each unit samples acceleration and gyroscopic data below 250Hz from the top and the side of the outer shell of a washing machine. The steps in building and developing the device are presented to allow replication and improvement. The collected and stored data from the device, allowed a first manual detection of multiple failures and anomalies on a washing machine only using few processing steps such as the Fast Fourier Transform. Further guidelines on an improved version and a complete deployment of such a device via the citizen science framework are also presented.