Process Framework Technologies Workshop Results Get in Touch
An Original Framework by Avadh Nagaralawala

Intelligent Rare Earth Process Systems

A novel digital twin framework — conceived, designed and developed by Avadh Nagaralawala — for rare earth extraction processes. Combining real-time sensing, predictive simulation, and intelligent control optimization through cyber-physical integration.

Physical Workshop Extraction · Control · Sensing Virtual Workshop 3D Model · Animation · Sync Digital Twin Data Platform Real-time · Management · Optimized Service System Control Opt · Process Sim · Inspection Drive Data Sim Data DT System Architecture RARE EARTH EXTRACTION PROCESS
<5% Max Relative Error
10 min Component Detection (from 1h)
4 Key DT Technologies
60 Extraction Stages Modelled
01 — Extraction Process

Rare Earth Production
Process Overview

The rare earth extraction process (REEP) is characterized by nonlinear behavior, long time delays, and strong coupling of process variables across thousands of cascade sub-processes.

01

Dissolution Circuit

Raw material powder is mixed with acid and water, then neutralized with alkali to a precise pH value. The solution is precipitated and filtered to produce the next-stage raw material.

02

Extraction Circuit

Feed liquid is separated across multi-stage extraction tanks using solvent P507. Organic and aqueous phases flow counter-currently — the core separation step that increases rare earth purity.

03

Scrubbing Section

Washing liquid HCl is introduced to further purify the separated components. Hard-extracted product YB exits in the aqueous phase; easy-extracted product YA exits in the organic phase.

04

Precipitation & Dehydration

Separated products are obtained by adding a precipitating agent with stirring. Dehydration completes the production cycle, delivering purified rare earth compounds for industrial application.

EXTRACTION PROCESS FLOW Dissolution Acid + Alkali Extraction n stages Scrubbing m stages Precipitation + Dehydration organic phase (counter-current) Y_A product Y_B product CASCADE EXTRACTION TANKS S1 S2 S3 Sn W1 W2 Wm COMPONENT DISTRIBUTION Ce Pr Nd
02 — DT Architecture

Digital Twin Framework

The DT architecture integrates four interconnected layers — physical workshop, virtual workshop, service system, and dynamic database — forming a closed-loop cyber-physical system.

Physical Workshop

Controls motors, dosing pumps and solenoid valves. Collects real-time data including solution component content, tank level, temperature and flow via SCADA and DCS networks.

Digital Twin Data Platform

Central data hub that synchronizes real-time, management, operational and optimized data between the physical and virtual systems. Drives iterative model updates.

Service System

Provides control optimization, process simulation, and virtual inspection modules. Reads real-time data to compute optimal control strategies via case-based reasoning.

Virtual Workshop

3D digital replica built with 3Ds Max and Unity. Achieves virtual-real interaction, data synchronization and consistent equipment behavior for remote inspection and fault warning.

03 — Key Technologies

Four Enabling Technologies

The system's intelligence rests on four tightly integrated technologies enabling real-time sensing, rapid prediction, control optimization and efficient inspection.

Soft Measurement of Component Content

Process engineers traditionally assess rare earth component content by visually observing the color of extracted solutions. This system formalizes that insight into a robust computational model.

The approach uses Grey Edge algorithm-based illumination compensation, followed by HSI and RGB feature extraction. A Weighted Least Squares Support Vector Machine (WLSSVM) then maps color characteristics to component content — ideal for small sample industrial environments.

Grey Edge algorithm corrects for hostile lighting conditions via Genetic Algorithm optimization
HSI and RGB color space features used as soft measurement model inputs
WLSSVM selected for fast learning and suitability with limited training samples
CC-WLSSVM variant achieves lowest relative error across all experimental cases
SOFT MEASUREMENT PIPELINE Solution Image Capture Grey Edge Compensation HSI + RGB Feature Extract WLSSVM Model Genetic Algorithm Parameter Optimization COLOR FEATURE CHANNELS H S I + R G B Component Content y y = f(wh·Ch, ws·Cs, wi·Ci, wR·CR...)

Process Simulation with Mechanism Compensation

A dynamic process simulation model overcomes the limitations of static mechanism methods by incorporating an improved Particle Swarm Optimization (PSO) algorithm to iteratively refine compensation coefficients.

The model is built on mass and element balance equations for each extraction stage, with separation coefficients β between rare earth components. A compensation factor K is introduced to account for insufficient extraction and is continuously updated with production data.

26 extraction + 34 scrubbing stages modelled (60 total)
Improved PSO with functional inertia weight and constriction factor
Compensation coefficient K updated iteratively with real production data
MEANRE below 0.43% for Ce, Pr, Nd after calibration
PSO OPTIMIZATION LOOP Compensation Coefficient K Init Model Fitness Eval Update Velocity Global Best Historical Data → Optimal K → Accurate Simulation

Case-Based Reasoning Control Strategy

Rather than relying solely on manual operator experience, the control system uses Case-Based Reasoning (CBR) to derive optimal extractant and detergent flow rate presets from a structured historical case library.

A fuzzy inference compensation model then dynamically adjusts flow rates in real time based on soft measurement feedback. This two-layer approach — preset via CBR, refined via fuzzy control — dramatically improves process stability and product quality.

7 initial production parameters as condition descriptors (CBR case features)
KNN nearest-neighbor retrieval finds closest historical operating case
Fuzzy control compensates dynamically for real-time process disturbances
Flow pre-set layer + PID flow loop control layer in cascade architecture
CBR + FUZZY CONTROL ARCHITECTURE Case Library Historical Cases X_k = (x1,...,x7) CBR Engine KNN Retrieval min distance(X) Flow Preset Vs⁰ (Extractant) Vw⁰ (Detergent) PID Flow Loop Control Extractant · Material · Acid Flow Control Rare Earth Extraction Y₁(K) · Y₂(K) Fuzzy Inference Compensation ΔVs ΔVw Soft Measure Y₁ feedback
04 — Virtual Inspection

Virtual Workshop &
3D Digital Inspection

A fully interactive 3D virtual replica of the rare earth production floor enables operators to inspect equipment state, monitor process animations, and receive automated fault warnings — remotely and in real time.

01

Process Demonstration

Fluid animations in Unity demonstrate the change and distribution of material liquid across all production stages, providing an intuitive understanding of extraction dynamics.

02

Real-Time Data Association

C# database scripts synchronize live sensor data from the DT data platform to the virtual model, ensuring the virtual workshop reflects actual physical conditions at all times.

03

Free Equipment Inspection

Operators navigate the 3D environment freely, accessing status information for each motor, pump and extraction tank without physical presence on the production floor.

04

Automated Fault Warning

The system scans motor states using timestamps and device IDs, detecting overrun conditions and issuing predictive fault warnings to prevent costly production shutdowns.

Virtual Workshop — Motor Status
MOTOR STATUS NORMAL NORMAL WARNING NORMAL 3Ds Max + Unity · Real-time Sync
05 — Validation

Model Validation & Results

The framework was validated across 100-sample soft measurement experiments and a 60-stage process simulation. All component prediction errors remain well within the accepted 5% threshold.

Process Simulation Accuracy — Mean Relative Error

Ce
0.425%
Pr
0.418%
Nd
0.175%
Max
5.0% limit

Relative Error After Calibration

Component MEANRE (%) MAXRE (%) RMSE (%) Status
Ce 0.4254 4.4344 1.38×10⁻⁴ ✓ Pass
Pr 0.4176 3.6566 2.25×10⁻⁴ ✓ Pass
Nd 0.1753 3.2692 1.44×10⁻⁴ ✓ Pass

Operational Efficiency Improvements

Inspection time: 30 min
5 min
83% reduction in inspection time
Detection time: 1 hour
10 min
83% reduction in component detection
06 — Conclusions & Future Work

Research Conclusions

Original Research & Innovation

Conceived, Designed & Developed by Avadh Nagaralawala

This digital twin framework represents an entirely novel contribution to intelligent process engineering. Every component — from the color-based soft measurement model to the improved PSO simulation engine and CBR-driven control strategy — was independently conceived and engineered by Avadh Nagaralawala to solve real industrial challenges in rare earth extraction.

This novel framework demonstrates a fully integrated digital twin system for rare earth extraction processes, addressing the critical challenges of nonlinear behavior, large time delays, and strong coupling between process variables that have historically made REEP difficult to automate.

The four-technology framework — soft measurement via color-based WLSSVM, mechanism-compensation process simulation with improved PSO, case-based reasoning control strategy, and a Unity-based virtual workshop — delivers a cohesive cyber-physical solution validated against real industrial data.

The validated model achieves mean relative errors below 0.5% for all three rare earth components (Ce, Pr, Nd), with all maximum errors comfortably under the 5% industrial threshold. The operational efficiencies achieved represent transformational improvements in production oversight.

Enhance soft measurement accuracy and computation speed under varying container transparency
Further improve process simulation algorithm for higher-stage extraction systems
Migrate the DT service interface to a web-based platform with server-side prediction
Increase the integration degree of the current DT system across all production modules
Improve usability of the virtual workshop for non-specialist operators
07 — Contact

Get in Touch

This framework represents a novel contribution to intelligent process engineering — applying digital twin architecture to one of the world's most strategically important industrial processes. I welcome collaboration, technical discussion, and inquiries from researchers, engineers, and industry professionals.

Areas of Expertise

Research & Innovation Profile

Digital Twin Systems Rare Earth Processing Process Simulation Soft Measurement Intelligent Control Machine Learning Cyber-Physical Systems PSO Optimization Case-Based Reasoning Virtual Reality Industry 4.0 Smart Manufacturing

Open to research collaborations, technical consultancy, and speaking engagements related to digital twin technology, intelligent manufacturing systems, and rare earth process engineering.