Mondelez Icon

Deep Learning AI Model

Built from first principles of Engineering and Mathematics (see ellipse diagram), a cutting-edge Neural network is designed for Mondelez International India manufacturing plant to precisely control Weight of chocolate getting deposited in moulds in Closed feedback loop. Architecture below:

Vectors & Matrices First, Second, and Third order Partial Differential Equations Stochastic Probability Distributions Fluid Mechanics Heat Transfer Thermodynamics Chemical Reaction Engineering Electrical & Control Systems GPU Computation
10 sensors 6 sensors 33 sensors 6 parameters
42 sensors 4 sensors 12 sensors 11 sensors 9 parameters
84 parameters
Every 15 mins Multi-Output Deep Neural Network Model Paste Mixer (mixes milk powder, cocoa butter, sugar) Refiner (reduces mix particle size by crushing) Conch (develops chocolate flavors by Mallard Chemical reactions) Quality lab data Auto-correction of Depositor settings Chocolate Storage tanks Chocolate Service tank Temperer (Nucleation & growth of Form-5 Crystals) Depositor (deposits liquid chocolate on plastic moulds) Quality lab data Cooling Tunnel (only chocolate bars sample collection at exit) Quality lab data

NETWORK ARCHITECTURE
(In General)

Zone 1 – Quality Lab Data
Zone 1 – OT Network
(Sensors / PLC / DCS)
Zone 2 – Edge Server (Industrial DAL- Data Abstraction Layer)
Zone 3 – Data Center (GPU computation for DL AI Model)

Double Encryption

  • mTLS (HTTPS) for network security
  • Secured API communication only
  • AES payload encryption

Mutual Authentication

  • Trust certificate identity verification
  • Only authorized Gateway connects
  • Prevents unauthorized injection

Zone 2: Industrial DAL

Transforms raw sensor signals into statistical parameters such as Reynolds Number, Prandtl Number, Nusselt Number, Agitator Power Number, Overall Heat Transfer Coefficient, Log Mean Temperature Difference (LMTD), Residence Time Distribution (RTD), Specific Enthalpy, Compressibility Factor, Coefficient of Variation, Skewness, Covariance Matrix, Transient Response Gradients, etc.

These derived variables capture underlying transport phenomena, thermodynamic behaviour, process variability, and dynamic system responses, thereby providing stable and normalized inputs to Deep learning AI model which are pre-requisites for Neural Network calculations. Edge Server is supplied and managed by NeuralFactoryAI.

Zone 3: Data Center Computation

This zone hosts a GPU-enabled AI execution environment where Deep learning neural network processes inputs received from Industrial DAL. AI engine operates within a secured runtime, performing high-speed inference to generate outputs. It is designed for low-latency execution and continuous adaptation to process dynamics. AI execution environment is managed by NeuralFactoryAI.

From Data to Intelligence

Deep-Learning Neural Network AI systems build to transform Mondelez Manufacturing line from Traditional human controlled operation to a Real-time AI controlled operation.

For Mondelez manufacturing line, the AI model is developed from first principles of engineering and applied mathematics, incorporating core process engineering, statistical learning techniques, and advanced neural network architectures. The solution involves mathematical modeling of key process equipment, including kinematic behavior of material flow, mechanical dynamics, and inherent process non-linearities.

The complete solution includes seamless integration with your existing PLC/SCADA infrastructure along with manual quality lab data, enabling continuous closed-loop feedback automated operation.