The Sentinel AI Platform: Engineering Intelligence, Not Just Alerts
Most structural monitoring systems tell you when a threshold has been crossed. Sentinel AI tells you why — and what happens next.
Sentinel AI is a multi-layer analytical engine built on 15 years of structural health monitoring research at the University of Illinois Urbana-Champaign. It converts continuous raw sensor data into engineering-grade condition intelligence: modal analysis, fatigue accumulation tracking, anomaly classification, and a continuously updated Structural Condition Index — all automated, all in real time, and all without requiring a structural engineer to manually review raw waveforms.
The result is a platform that closes the gap between inspection cycles, quantifies what periodic inspection cannot see, and converts physical structural behavior into the kind of actionable intelligence that asset managers, engineers, and operations leaders can act on immediately.
Below is a technical description of each functional layer of the Sentinel AI stack.
1. Signal Preprocessing, Feature Extraction, and Event Classification
Raw time-series data from the Xnode sensor suite is ingested continuously for event-triggered capture. Before any analytical output is generated, Sentinel’s on-device edge processing module applies:
- Normalization and detrending to remove sensor drift, mounted orientation variance, temperature-induced drift and offset, and environmental baseline variation, creating consistent datasets suitable for structural analysis
- Feature extraction identifies the relevant parameters from the normalized time-series data (peak acceleration/displacement, peak frequencies, spectral energy, etc.) for structural analysis and ML-based event classification.
- Event classification via ML to separate genuine structural response signals from routine ambient noise, eliminating false positives caused by wind, traffic, or process-induced vibration that does not materially impact structural condition
This preprocessing layer is what allows Sentinel to produce actionable outputs rather than raw waveforms that require manual engineering interpretation. By excluding structurally irrelevant datasets at the source, it also dramatically reduces the volume of retained sensor data, lowering both bandwidth and data storage costs.
2. Operational Modal Analysis (OMA)
Sentinel performs Operational Modal Analysis on the preprocessed time-series data. OMA identifies the natural frequencies, mode shapes, and damping ratios of each monitored structure using only ambient excitation — no controlled input loading is required. Key technical methods include:
- Power Spectral Density (PSD) analysis to identify dominant frequency content and track modal frequency shifts over time
- Empirical Mode Decomposition (EMD) to decompose nonlinear, non-stationary structural vibration signals into intrinsic mode functions, enabling separation of structural response components across frequency ranges
- Automated baseline establishment during initial deployment, creating a per-asset modal fingerprint against which subsequent data is continuously compared
A statistically significant shift in a structure’s natural frequencies is a direct indicator of stiffness loss — the earliest measurable precursor to structural degradation. Sentinel detects these shifts automatically and without requiring a structural engineer to manually review and analyze raw waveform data.
3. Fatigue Accumulation Modeling
For structures subject to cyclic loading — pipe racks, flare stacks, crane runway girders, offshore jacket members, conveyor frames — Sentinel performs empirical fatigue accumulation tracking using measured dynamic response rather than assumed load histories:
- S-N curve methodology applied to real-time sensor data, consistent with AISC Design Guide 7, DNV fatigue standards, and API structural fatigue guidance
- Rainflow counting algorithms applied to time-series sensor data to extract stress cycle amplitudes and compute damage accumulation per cycle
- Sensor-fed digital twin updated continuously, replacing point-in-time FEA models built on assumed load inputs with an empirically grounded remaining fatigue life estimate
The output is a continuously updated fatigue accumulation curve per asset — not a periodic engineering estimate, but a live empirical record of actual structural loading.
4. Anomaly Detection and Event Classification
Sentinel’s anomaly detection layer, which takes a holistic view at the structure using a combination of deployed sensors and structural models, runs in parallel with modal analysis and fatigue tracking:
- ML-based event classification distinguishes between vessel impacts, seismic ground motion, wind loading events, pressure relief excursions, and routine operational vibration
- Event-triggered data capture automatically activates sampling during anomalous events, capturing the full structural dynamic response without requiring manual intervention
- Post-event modal comparison compares post-event natural frequencies to the pre-event baseline, enabling automated detection of stiffness changes that indicate structural damage — within hours of the event, before any inspector enters the field
- Role-based alerting delivers timely, actionable notifications of detected anomalies to relevant personnel.
5. The Structural Condition Index (SCI)
The SCI is Sentinel AI’s headline output and StructureIQ’s core IP. It is a continuous, normalized 0-100 score per asset, computed from the integrated outputs of modal analysis, fatigue accumulation, anomaly classification, and trend analysis:
- Raw sensor time-series is normalized, detrended, and classified by ML to strip noise from signal
- The SCI reflects the asset’s current condition state across three tiers: Healthy / Watch / At Risk
- Paired fragility curves express the probability of structural damage at given future loading levels, enabling forward-looking risk assessment rather than only backward-looking condition reporting
- The SCI is sector-agnostic — directly comparable across buildings, bridges, offshore platforms, and industrial structures — enabling portfolio-level condition management and providing a quantitative substitute for qualitative inspection-based judgment
6. Model-Based Damage Localization
For assets with structural models, Sentinel operates a living digital twin continuously tuned by real sensor measurements: