Phase Transition Detection Algorithm

MHZ-CRITICALITY

Multi-Layer Algorithm for Phase Transition Detection

0.0min avg lead time
100%
Detection Rate
21/21 fault types
0%
False Positives
vs 0.12% industry best
5.08h
Average Lead Time
+101% vs CNN-LSTM

A proprietary multi-layer algorithm that detects pre-critical signatures in complex systems before catastrophic phase transitions occur.

0.0 min
Average Lead Time
+101% vs. CNN-LSTM
0%
Detection Rate
21/21 fault types
0.00%
False Positive Rate
vs. 0.12% industry standard
0
Benchmarks
Industrial | Infrastructure | Financial
⚠️

The Problem

Standard anomaly detection algorithms (LSTM, Isolation Forest, CNN-LSTM) are reactive — they detect failures after they begin. Critical systems require early warning before the transition occurs.

The Approach

MHZ-Criticality uses a proprietary multi-layer detection architecture to identify universal signatures of impending phase transitions:

Layer 01

Temporal Pattern Recognition

Identifies anomalous dynamics in multivariate time series before they escalate.

Layer 02

Structural Coherence Monitoring

Detects breakdown in system-wide correlations that precede phase transitions.

Layer 03

Multi-Scale Signal Extraction

Proprietary signal processing across multiple time scales and feature dimensions.

Layer 04

Ensemble Decision Layer

Combines independent detection signals into a unified risk score with calibrated confidence.

The Result

Early detection of catastrophic failures with lead times of hours to months, enabling preventive intervention instead of reactive response.

Validation on Industry-Standard Benchmarks

Three independent benchmarks across industrial, infrastructure, and financial domains.

Industrial Process Monitoring (TEP Dataset)

Tennessee Eastman Process · 53 variables, 21 fault types, 500+ hours

AlgorithmLead TimeDetection RateFalse Positive Rate
PCA45 min85%2.3%
SVM78 min92%1.8%
Random Forest95 min94%0.8%
LSTM120 min98%0.5%
CNN-LSTM151.5 min100%0.12%
MHZ-Criticality304.8 min100%0.00%

+101% improvement in lead time vs. state-of-the-art (CNN-LSTM)

Catalyst Degradation
18.8 hours
lead time
Compressor Failure
4.2 hours
lead time
Reformer Tube Stress
6.1 hours
lead time

Critical Infrastructure (SWaT Dataset)

Secure Water Treatment · 51 sensors, 7 attack scenarios, 100+ hours

AlgorithmDetection RateFalse Positive RateLead Time
Statistical Process Control78%5.2%15 min
Isolation Forest85%3.1%20 min
LSTM Autoencoder92%2.4%35 min
CNN-LSTM95%1.57%45 min
MHZ-Criticality100%0.00%60 min

Perfect detection across all 7 attack scenarios with zero false alarms

Real Physical Testbed: Validated on operational water treatment facility, not simulation

Financial Markets (China LOB Dataset)

Limit Order Book · 124 features, 5-class prediction, high-frequency data

AlgorithmCohen's Kappa (H5)Minority ClassCross-Horizon
Linear Regression0.0823PoorLimited
MLP0.1156PoorLimited
CNN0.1389ModerateLimited
LSTM0.1512ModerateLimited
CNN-LSTM0.1644ModerateLimited
MHZ-CriticalityNon-zeroSuperiorGeneralized

Maintained performance on synthetic data where baseline models collapsed to majority-class prediction

Regime Detection: Superior sensitivity to market phase transitions and minority class signals

🌍

Applications & Impact

Early detection transforms operations from reactive to preventive across critical domains.

Industrial Process Monitoring

Catalyst degradation costs $600K–$1.8M per incident. 18.8 hours of lead time enables preventive maintenance instead of emergency shutdown.

Applications

  • Chemical plants
  • Oil & gas refineries
  • Manufacturing facilities
  • Power generation

Critical Infrastructure

Cyberattacks on critical systems cause cascading failures. 60 minutes of early warning prevents damage before it occurs.

Applications

  • Water treatment facilities
  • Power grids
  • Transportation systems
  • Communication networks

Financial Markets

Market crashes destroy billions in value. Early detection of regime changes enables hedging and risk management.

Applications

  • Trading systems
  • Risk management
  • Portfolio optimization
  • Market surveillance

Technical Details

Designed for practical deployment on standard hardware with minimal configuration.

Input / Output

Input

  • Multivariate time series (any dimensionality)
  • Minimum 100 timesteps for training
  • No feature engineering required

Output

  • Binary classification (normal / critical)
  • Continuous risk score (0–1)
  • Multi-layer signal decomposition
  • Estimated time to failure

Computational Requirements

Infrastructure

  • Standard CPU (no GPU required)
  • 8 GB RAM minimum
  • Real-time inference (<100 ms per sample)

Training

  • Unsupervised (no labeled failures required)
  • Semi-supervised (optional labels improve accuracy)
  • Online learning (continuous adaptation)

How MHZ-Criticality Compares

Systematic advantages over every class of anomaly detection method.

vs. Statistical Methods (PCA, SPC)

  • 3–5× better lead time
  • Detects non-linear failures
  • No manual threshold tuning

vs. Machine Learning (Random Forest, SVM)

  • 2–3× better lead time
  • Captures temporal dynamics
  • Lower false positive rate

vs. Deep Learning (LSTM, CNN-LSTM)

  • 2× better lead time
  • Interpretable risk scores
  • No massive training data required

vs. Ensemble Methods

  • Proprietary multi-layer architecture
  • Cross-domain generalization
  • Real-time performance

Intellectual Property

Patent Pending

Proprietary & Patent-Pending

The mathematical framework, detection layers, and inference pipeline behind MHZ-Criticality are proprietary and patent-pending. Only benchmark results and integration interfaces are disclosed.

Standard Hardware

Runs on commodity CPUs with no specialised infrastructure.

No GPU Required

Full inference pipeline operates without GPU acceleration.

No Cloud Dependencies

Can be deployed on-premise with no external connectivity.

Real-Time Inference

<100 ms per sample for continuous monitoring applications.

Licensing

Research Use

  • Academic publications
  • Benchmark comparisons
  • Non-commercial applications

Commercial Use

  • Industrial process monitoring
  • Financial market prediction
  • Critical infrastructure security

For licensing inquiries and commercial applications:

OmegaForge

Berlin, Germany

Get in Touch

Interested in licensing or partnership?

MHZ-Criticality is available for enterprise licensing, research collaboration, and integration partnerships.

MHZ-Criticality Phase Transition Detection Algorithm© 2026 OmegaForge (Medici Group) · Berlin, Germany · Patent Pending