Machine Learning

Neural Flow uses three types of machine learning engines to reveal hidden patterns and correlation within the existing data

Clustering - Detects anomalies in batches

Clustering unsupervised learning algorithm for anomaly detection that works on the principle of ‘isolating anomalies”. The algorithm looks at the spread of the dots, and starts to randomly partition them. If the dots are close to each other, it will take the algorithm longer to separate each dot. If the dots are far from each other, less separation is required, and the algorithm works faster. The figure below demonstrates (in a 2D domain) that isolated anomalies typically require less amount of splits.

Heating stage - LEARNING CURVE
Heating stage - LOSS DISTRIBUTION

Neural Network

Neural Network – enables time series data prediction.The model is a Recurrent Neural Network (RNN) model that enables time series data prediction, based on sensor correlations. The model identifies non-linear correlations between the sensors, and alerts of abnormal sensor behavior. The model tries to learn the data representation of its input, usually, by learning an efficient encoding that uses fewer parameters / memory. In a sense, the model learns the most important features of the data using as few parameters as possible.

 

FIGUR 15 - ANOMALY - DETECTION - LEARNING PHASE ANOMALY PREDICTION

Heating stage - LEARNING CURVE
Heating stage - LOSS DISTRIBUTION
Heating stage - LOSS VS. TIME

Statistical Model

statistical model for signal analysis and forecasting. It calculates the signal behavior based on three parameters:

  • Trend: a general systematic linear or (most often) nonlinear component that changes over time and does not repeat.
  • Seasonality: a general systematic linear or (most often) nonlinear component that changes over time and does repeat.
  • Noise: a non-systematic component that is neither a Trend or a Seasonality within the data.

Statistical Model- SENSORS S1, STATE heating 1(FREQ=1000)

ABOUT US

Founded in 2019, Neural Flow is an Israeli-based 4.0 startup and part of the MGT Liquid & Process Systems Group, an industry leader with over 50 years of experience in manufacturing custom reactors, mixing systems and process solutions.

As process and chemical engineers and mathematicians, our mission is to streamline process management, utilizing information that is usually overlooked in the standard way of doing things. Our software platform is designed to assist engineers in their daily work, and provide tools to improve process results and company profitability.

The Neural Flow system is implemented in the chemical, pharmaceutical, paint & ink, cosmetics and F&B industries, and is suitable for any process involving reactors.

We look forward to demonstrating our system’s capabilities and show you how we can help you achieve better results.