INTELLIGENT MOTION SENSOR PLATFORM FOR MEASURING TOOLS

This thesis presents the design, development, and experimental validation of an embedded anomaly detection system for vibration and shock monitoring in industrial tools. The project was developed during a curricular internship at Hilti AG, in response to real-world issues observed in the field. Two key challenges motivated this work: first, end users of Hilti’s rotating lasers and positioning layout tools reported unreliable measurements and tool failures caused by undetected shocks or vibrations during operation. Second, the repair centers faced difficulties in determining whether returned tools had been damaged due to misuse (e.g., during transport or storage), making it challenging to fairly apply warranty policies. To address these issues, a comprehensive set of use cases was defined to represent possible misuse and failure scenarios across the tool’s lifecycle—from job site operation to rental returns. Based on these, a detailed list of functional requirements was derived, including motion detection, shock classification, user feedback, and diagnostic data logging. A prototype system was built using the STM32F401RE microcontroller and the ISM330ISN intelligent sensor, which includes an on-board Intelligent Sensor Processing Unit (ISPU). A machine learning model was trained using real-world sensor data to classify anomalies and adapt to acceptable behavior patterns through on-device learning. The firmware was implemented in C/C++ using a real-time operating system (RTOS) and organized as a finite state machine. For testing, the hardware was integrated into a 3D-printed, custom-designed enclosure featuring an OLED display, joystick interface, and status indicators. Real-world testing was conducted in Hilti’s Anchors Department, where the system was exposed to typical industrial vibrations and shocks. The prototype successfully detected and classified anomalies in line with the defined use cases, distinguishing between harmful and harmless events and reacting promptly to unexpected conditions. These results validate the use of embedded machine learning for dynamic anomaly detection in industrial tools. A practical integration approach was also proposed: a compact, low-cost PCB module designed to retrofit existing tools. To support warranty assessments, a high-g accelerometer was added for detecting extreme impacts, along with a backup battery to enable logging even when the main battery is disconnected. Overall, the system fulfills the majority of the defined requirements, offering a valid and scalable solution to improve tool reliability, support repair diagnostics,and enhance customer satisfaction.