Skip to main content

Predictive Maintenance and Vibration Analysis

Predictive maintenance (PdM) leverages the Always-On capabilities of the Syntiant® NDP to transform industrial monitoring from scheduled inspections to continuous, intelligent oversight.

By processing high-frequency sensor data locally, the TML120 can identify the earliest signs of mechanical wear—often weeks before a human operator could detect them. This proactive approach significantly reduces unplanned downtime and extends the operational life of expensive machinery.

  • Acoustic Fault Detection: The NDP is trained to recognize specific spectral signatures associated with mechanical failure, such as the high-frequency "metal-on-metal" squeal of a dry bearing or the rhythmic, low-frequency thumping of a misaligned shaft.

  • Vibration Anomaly Identification: By interfacing directly with 3-axis accelerometers, the NDP monitors harmonic patterns in real-time. To quantify the intensity of these vibrations and identify dangerous deviations from the baseline, we calculate the Vibration RMS.

Formula: Vibration RMS (vrmsv_{\text{rms}})

vrms=1n∑i=1nai2v_{\text{rms}} = \sqrt{\frac{1}{n}\sum_{i=1}^{n} a_i^2}

(where a is the acceleration sample at time i.)

Formula: Slip Frequency (fsf_s)

fs=fsync−factualf_s = f_{\text{sync}} - f_{\text{actual}}

(where fsync is the synchronous speed of the magnetic field and factual is the measured rotor speed.)

  • Anomaly Detection at the Edge: Instead of continuously streaming gigabytes of raw vibration data to a cloud gateway—which would be cost-prohibitive and power-intensive—the device remains in a low-power "Sleep" state. It only triggers a high-power alert when the processed features exceed the learned Confidence Threshold (T) established during validation.
⚠️ The "Baseline" Calibration

"When deploying in a factory, use the first 48 hours of operation to 'Calibrate the Baseline.' Every machine has a unique acoustic and vibration fingerprint. By capturing a machine-specific baseline, you can lower the Confidence Threshold (T) to detect even subtler anomalies without increasing the False Acceptance Rate (FAR)."