Technical deep dive
From electrical signal
to validated diagnostic
SAM4 turns raw current and voltage waveforms into confirmed fault diagnostics — continuously, for your entire fleet. This page walks through the full pipeline: how signals are captured, how AI models detect anomalies, how engineers validate every alert, and how results reach your team.
This pipeline runs across 10,000+ assets today, delivering a >90% detection rate with a false positive rate below 5%.
Step 1
Signal acquisition at the MCC
Every AC motor draws current and voltage in patterns that reflect the mechanical condition of the full drivetrain. SAM4 captures those patterns at the motor control cabinet — not on the asset.
Hardware
Split-core current transformers (CTs) and voltage taps (VTs) clamp around the motor supply cables inside the MCC. No process shutdown. No mechanical modification. Installation takes under 30 minutes per asset.
Sampling
The SAM4 data acquisition unit samples current and voltage at high frequency — capturing the harmonic content needed to detect mechanical faults reflected in the electrical signature. Sampling parameters are tuned per motor type and application.
Edge processing
Raw waveforms are pre-processed on the edge device before transmission. This includes windowing, basic quality checks, and compression. Only diagnostic-relevant data is sent to the cloud — reducing bandwidth by an order of magnitude while retaining full spectral information.
Connectivity
Data transmits over standard industrial protocols (4G, Ethernet, Wi-Fi) to Samotics' cloud infrastructure. The edge device handles intermittent connectivity gracefully — buffering locally and syncing when reconnected. No IT infrastructure changes required.
Why the MCC, not the asset?
Conventional vibration sensors need physical mounting on the machine — which means physical access, permits, wiring runs, and per-asset installation costs. For submerged pumps, sealed blowers, ATEX zones, or remote sites, that's often impossible. The motor control cabinet is always accessible, always in a safe environment, and serves every motor on the feeder. One installation point. Full drivetrain visibility.
Step 2
Signal processing and feature extraction
Raw electrical waveforms contain thousands of data points per second. The signal processing layer transforms this into a structured set of diagnostic features — the fingerprint of each asset's condition.
Spectral decomposition
Time-domain signals are transformed into the frequency domain using FFT and advanced spectral techniques. This reveals harmonic patterns invisible in the raw waveform — sideband frequencies around the supply frequency that correspond to specific mechanical fault signatures (bearing defect frequencies, gear mesh patterns, rotor bar faults).
Feature extraction
From each processed signal, SAM4 extracts hundreds of diagnostic features: spectral amplitudes at fault-specific frequencies, envelope statistics, current signature symmetry, power factor trends, and load-normalised indicators. These features form a high-dimensional fingerprint that characterises the asset's condition at each measurement point.
Operating state classification
Motors operate across different loads, speeds, and duty cycles. SAM4 classifies each measurement's operating state before comparing it to baseline — ensuring that a change in load isn't misread as a developing fault. This is critical for assets with variable frequency drives (VFDs) or intermittent duty.
Baseline learning
During the initial monitoring period, SAM4 builds a statistical baseline for each asset across its observed operating states. Future measurements are compared against this asset-specific baseline — not generic thresholds. This means every motor is evaluated relative to its own normal behaviour, reducing false positives and catching subtle degradation early.
Step 3
AI-driven anomaly detection
The detection engine compares each asset's current condition against its learned baseline and a library of known fault patterns — trained on data from over 10,000 industrial assets across 80+ customers.
Anomaly scoring
Each feature vector is scored against the asset's baseline using statistical distance metrics. Features that deviate beyond learned thresholds are flagged. The system tracks trend direction and rate of change — distinguishing between a stable anomaly and one that's accelerating toward failure.
Fault classification
When an anomaly is detected, classification models map the pattern of deviating features to known fault types: bearing degradation, misalignment, imbalance, electrical faults, cavitation, and more. The models are trained on labelled fault data from confirmed failures across the installed base — giving them real-world ground truth, not just lab data.
Severity estimation
Beyond detecting *that* a fault exists, the engine estimates *how far* the degradation has progressed. This combines the magnitude of spectral deviation, the rate of trend progression, and fault-type-specific degradation models — giving engineers a severity assessment to prioritise interventions.
How the models improve
Every validated alert feeds back into the training pipeline. When a Samotics engineer confirms a fault type and the customer validates the finding during maintenance, that data pair strengthens the models. With 80+ customers generating continuous feedback, the detection engine compounds in accuracy over time — an advantage that grows with scale.
Step 4
Every alert reviewed by a reliability engineer
AI flags the anomaly. A human confirms it. Every alert that reaches your team has been reviewed by a Samotics reliability engineer — no unfiltered machine output, no alert fatigue.
AI flags anomaly
The detection engine identifies a statistically significant deviation and generates a preliminary classification with confidence score.
Engineer reviews
A Samotics reliability engineer examines the spectral data, reviews the asset's operating context, and confirms or refines the fault classification. They assess severity and write a specific maintenance recommendation.
Critical escalation
Time-critical alerts — rapidly progressing faults, risk of imminent failure — are escalated immediately via direct notification. Your team gets a phone call, not just a dashboard update.
Alert delivered
Validated diagnostics are published to the SAM4 dashboard with full supporting data: spectral plots, trend history, fault classification, severity, and recommended action. Every diagnostic is transparent — your team can see the evidence behind every conclusion.
Why human-in-the-loop matters
Fully automated systems generate alert fatigue. Teams stop trusting them. SAM4's hybrid approach means every alert carries an engineer's judgement — which is why customers act on SAM4 alerts. The human review also catches edge cases that pure ML would miss: unusual operating conditions, maintenance activities that mimic faults, or asset configurations the models haven't seen before.
Step 5
Alert delivery and system integration
Validated diagnostics reach your team through the SAM4 dashboard and push directly into your existing maintenance systems — fitting into the workflow your team already uses.
SAM4 dashboard
Fleet-level overview with drill-down to individual assets. Each diagnostic includes: fault classification, severity rating, trend history, spectral evidence, and recommended action. Designed for maintenance planners, not data scientists.
CMMS integration
SAM4 pushes validated alerts directly into your CMMS (SAP PM, Maximo, Ultimo, and others) as work orders or notifications. No manual re-entry. Alerts carry the fault type, severity, asset identifier, and recommended action — ready for your planning process.
API access
For teams building custom dashboards or integrating into existing data platforms, SAM4 provides a REST API. Pull asset health data, alert history, and trend metrics into your own systems. Full documentation available during onboarding.
Alert taxonomy
SAM4 alerts use a structured severity scale: Watch (early-stage deviation, monitor), Warning (confirmed fault, plan intervention), and Critical (imminent failure risk, act now). Each level has a defined SLA for engineering review and customer notification — so your team knows exactly what each alert means and how quickly to respond.
Reporting
Monthly fleet health reports summarise asset condition, open diagnostics, resolved findings, and fleet-level trends. Quarterly business reviews map SAM4 findings to maintenance outcomes — confirmed saves, avoided downtime, and energy efficiency gains. All data available for export.
Performance
Detection accuracy across 10,000+ assets
These metrics are measured across confirmed failures in the installed base — not lab conditions. Definitions and methodology are available on request.
A note on metrics
Detection rate and false positive rate are measured differently across the industry — and some vendors report lab results rather than field performance. SAM4's metrics reflect real-world performance across the installed base. We define detection rate as the proportion of confirmed failures (validated by the customer during maintenance) where SAM4 had raised an alert before failure occurred. We publish these metrics because enterprise buyers deserve auditable claims. Request full methodology →
Security & data
Enterprise-grade data handling
Data ownership
Your data remains yours. Samotics processes electrical signatures to deliver diagnostics. Raw data and diagnostic results are available for export at any time. Data retention policies are contractually defined.
Cloud infrastructure
SAM4 runs on enterprise-grade cloud infrastructure with encryption at rest and in transit, role-based access control, and SOC 2–aligned security practices. Architecture details are available during technical due diligence.
Network requirements
The edge device requires outbound-only connectivity — no inbound ports, no VPN tunnels, no changes to your firewall rules. It works over 4G, Ethernet, or Wi-Fi. Bandwidth requirements are minimal after edge compression.
Go deeper or get started
Whether you're evaluating the technology or ready to see it on your assets.