Data-Driven Evaluation of Optimal IMU Placement on Ice Hockey Helmets

Contact sports such as ice hockey expose athletes to a considerable risk of concussions, emphasizing the importance of reliable tools for head impact monitoring. To address this challenge, a smart helmet was developed, embedding inertial measurement units (IMUs) at three distinct locations on the outer shell. Experimental tests were carried out in a controlled laboratory environment, where a pendulum impactor was used to deliver standardized impacts to a dummy's head according to a predefined protocol. An additional IMU, placed inside the headform, served as the reference system. Each sensor recorded both linear accelerations and angular velocities along the three spatial axes, and the measurements collected from the helmet-mounted IMUs were systematically compared against the reference IMU. The objective of this study is to determine which sensor location provides signals most consistent with the ground truth, thereby identifying the most reliable placement for wearable sensors intended to capture head impact kinematics. Data processing involved segmentation and synchronization of the signals to isolate single impacts, followed by feature extraction in the time, frequency, and time–frequency domains. Features were heuristically selected to provide meaningful descriptors of the signals and to address two research questions: which helmet-mounted IMU exhibits patterns most similar to the ground truth, and whether impact direction influences measurement accuracy. A statistical analysis was conducted to compare features across sensors, while a supervised machine learning framework was employed to assess the degree of decoupling between the helmet and the headform, labeling impacts as either low or high decoupling. The results demonstrate that sensor position has a measurable effect on signal reliability. Moreover, the machine learning analysis provides additional support by quantifying the susceptibility of each sensor to decoupling, thereby offering complementary insights into overall measurement reliability.