Currux Vision

A Self- Learning AI Anomaly Detection and Predictive Analytics System

Currux Vision One of The Best AI & Machine Learning Products of 2018

Vision nominated as AI & Machine Learning Product of the Year by Product Hunt

Some of the other nominees included:
  • Amazon Scout
  • Google News
  • Google AI Assistant (Duplex)
"Currux Vision Corp Partners with Costar Technologies and Rastrac Net, Inc., to Implement Cutting Edge AI Systems." 2019





Our Goal Is to Enable the Autonomous Machine. Anomaly Prediction Is The First Step
Currux Vision Corp built a proprietary and highly adaptable anomaly detection system.The Problems Currux Vision Addresses:

  • Multiple industrial sensors generate large volumes of noisy data
  • Less than 1% of all sensor data is used or analyzed;
  • Breakdowns are expensive;
  • Current solutions require data preprocessing and labeling, cloud connection and most are not real time;
  • Current solutions can be very expensive;
  • SCADA systems have little to no predictive capabilities;
  • Industrial machines can have 20 – 100 + sensors per machine;
  • Large volumes of sensor data are impossible for humans to interpret, especially in real time.
How Is Currux Vision Different?
Easy to use, automated, comprehensive, local
  • No cloud required; works locally as well as in cloud;
  • Uses data from existing sensors; no new hardware required;
  • Includes an automated data labeling, pre-processing and augmentation tool;
  • Custom trained AI models for each system / machine;
  • Integrates with most SCADA systems to receive sensor data;
  • Works in real time;
  • Alerts about potential anomalies before they happen;
  • Cost effective, minimal CPU hardware requirements.

Currux Vision Comparison

Companies
Microsoft Azure
Numenta
Currux Vision
Access
Cloud
Cloud
Local machine and Cloud
Methods
Spectral Residual Analysis + Convolutional Neural Network (CNN)
Hierarchical Temporal Memory
5 Discrete AI/ML Frameworks
Live stream of data
Speed
Very slow (uses windowing method on every step)
Fast
Fast
Accuracy
Depends a lot on the chosen parameters
Good
Very Good
Data Export
.csv
.csv (or any other formats)
Universality to input data
Predictive Capabilities
Predicting Catastrophic Failure:

  • Chemical company experienced two potentially catastrophic equipment failures

  • Existing anomaly detection / SCADA tools did not help

  • The goal was to create a custom AI/ML model to accurately predict future equipment breakdowns

  • Historic data from 21 sensors was processed and used to train the AI model

  • System became able predict future failures 39-44 hours in advance

  • 99.782 % accuracy achieved as tested