Detecting Device Mishandling
Detecting Device Mishandling is an experimental software solution we successfully implemented. Our client operates a maintenance program where damaged returns are replaced. Their devices are handled by thousands of workers in various environments, and improper handling can lead to expensive returns. It was designed as a concept study to explore whether device data could be leveraged to identify patterns of device usage, including potential mishandling.
We successfully differentiated between accidental and intentional damage by analyzing in-built accelerometer data and applying Machine Learning algorithms to detect distinct handling signatures that could indicate misuse or mishandling of the devices.
- Built a custom Android agent to collect accelerometer data from deployed devices
- Trained ML models to classify events such as accidental drops, improper handling (slams, throws), and normal usage
- Achieved 90% accuracy in identifying misuse
- Tuned models to filter out false positives
The system, if productized, can, with high accuracy, differentiate between legitimate damage and user-induced damage. Such a solution can assist in reducing warranty misuse and suggest ruggedization improvements.
