PSI - Issue 75

Mohamed El Yazrhi et al. / Procedia Structural Integrity 75 (2025) 262–275 Mohamed El Yazrhi , Jean-Yves Disson / Structural Integrity Procedia (2025)

265

4

To account for mission-length extrapolation and specified risk levels, AFNOR XP X50-144 introduces two risk based spectra: • XRS : Gumbel extreme-value extrapolation of extreme responses ∗ . For a chosen exceedance probability , the XRS ordinate at ωi is ∗ (α)=μ −β ln(− ln(1 − α)), where are the fitted Gumbel location and scale parameters [9]. • XFS : Central-limit extrapolation of cumulative damage D c . Under block partitioning and assuming approximate normality, the damage at risk level α is ∗ (α)= ̅̅̅ + α σ , With ̅ the mean damage per block, its standard deviation, and the standard normal quantile [9]. These spectra enable both tailored qualification testing — by matching laboratory shaker profiles to field-derived FDS/XRS curves — and real-time health tracking, where on-board comparisons of running damage to ∗ (α) trigger maintenance alerts before critical thresholds are exceeded. 3. Monitoring of the mechanical environment harshness: POC embedded monitoring platform 3.1. System architecture and functions NOMAD is a compact proof-of-concept embedded platform (95 × 60 × 45 mm) for on-site vibration and fatigue monitoring. Its key components are: • The Acquisition of vibration signals using a 3-axis capacitive MEMS accelerometer (Analog Devices ADXL356) mounted at the equipment’s base, offering a 2 500 Hz bandwidth and an ADC sampling rate adjustable up to 33 kHz per channel. • Quad-core ARM Cortex processor with 8 GB RAM • Real-time calculation of dedicated indicators, such as ERS and FDS. • Optional machine-learning models for automatic usage-pattern classification • Web-based interface for remote parameter setup, live visualization, and data download • Expandable I/O with slots for CAN-bus and GPS cards

Figure 1 : POC NOMAD

3.2. Real-time pipeline

NOMAD continuously acquires vibration data and processes it on board in real time. The signal first undergoes basic pre-processing to remove offsets, spikes, or other anomalies. It is then segmented into time-disjoint blocks, typically ranging from 1 to 10 seconds, allowing independent processing of each segment. For every block,