PAPER, FOREST PRODUCTS & PACKAGING
Mill & Converting Intelligence
Energy-Optimal Papermaking
Zero-Defect Converting
Unified Mill & Converting Intelligence Platform that stabilizes paper machine KPIs with physics-informed ML, eliminates converting defects with high-speed CV, and embeds openLCA-grade footprint accounting into daily decisions.
GenAI + CV
RL + OR
LCA-Driven
Packaging Design
Zero-Defect
Converting
Mill Control Center
Energy
-7.8%
specific steam
Quality
-33%
moisture σ
Yield
-14%
trim-loss
Throughput
+8%
OEE
Paper Machine Control
Stabilized
Converting AOI
Monitoring
LCA Engine
Active
Three-Front War
Papermaking and packaging plants face constant operational challenges
01
High Energy Intensity
Variable energy intensity across fiber prep, paper machine, and recovery
Impact: Steam and electricity waste
Small moisture/basis-weight drifts cascade into significant energy waste and off-spec reels
02
Quality Escapes & Yield Loss
Registration errors, glue/warp defects, poor roll mapping in converting
Impact: Converting inefficiency
Corrugators, slitters, sheeters, die-cutters suffer from optimization challenges
03
Sustainability & Compliance Pressure
FSC/PEFC chain-of-custody, LCA of formats and substrates
Impact: Regulatory compliance
Buyers demand verifiable footprint claims and recycled content verification
04
Process Variability
Moisture and basis-weight drifts affect paper quality
Impact: Off-spec production
Unstable paper machine KPIs lead to quality issues and material waste
05
Trim Loss & Changeovers
Inefficient cutting patterns and scheduling
Impact: Material waste
Poor optimization of corrugated boards and cutting/packing operations
06
Audit & Documentation
Complex chain-of-custody and environmental reporting
Impact: Compliance burden
Manual processes for FSC documentation and LCA calculations
Tensorblue's Answer
A unified Mill & Converting Intelligence Platform that (i) stabilizes paper machine KPIs with physics-informed ML + MPC/RL, (ii) eliminates converting defects with high-speed CV and OR-driven scheduling/cutting, and (iii) embeds openLCA-grade footprint accounting and FSC chain-of-custody into daily decisions.
Real, Scaled Anchors We Build On
Proven foundations for mill and converting intelligence
01
Research Foundation
Energy & Decarbonization Science
Concrete measures around dewatering, recovery, and heat integration provide baselines and targets for control policies.
Source: ScienceDirect 2025
Validated
02
OR Literature
Corrugated/Converting Optimization
End-to-end reviews and new matheuristics for 2D cutting stock & multi-machine scheduling inform OR kernels.
Source: MDPI 2023-2025
Validated
03
Compliance Standard
FSC Chain of Custody
Authoritative, in-force requirements for procurement, segregation/mass-balance, and labeling; operationalized in data model.
Source: FSC-STD-40-004 v3-1
Validated
04
LCA Platform
openLCA Ecosystem
Current LCIA method packages (TRACI 2.2; IPCC 2021 updates) and EF 3.1 database enable auditable LCA.
Source: openlca.org
Validated
Tensorblue Integration
These proven foundations provide scalable, validated baselinesfor Tensorblue's mill and converting intelligence deployments across paper, forest products, and packaging operations.
End-to-End Architecture
Mill to market intelligence flow
Data Sources
Fiber Sources
Chips
Pulp
Stock Prep
PM Sections
Utilities
Converting Lines
WMS/ERP
FSC Docs
↓
Data Fabric
Time-series ETL, roll genealogy, lab QC, vision streams, FSC CoC records
SCADA/historian
Vision streams
FSC CoC records
↓
AI & Optimization Core
Paper Machine Control
Physics-informed ML + MPC/RL
Vision QA
Reel & web defects; converting AOI
OR Kernels
Cutting stock, trim-loss, scheduling
LCA Engine
openLCA methods + EF3.1 datasets
GenAI Copilots
Specs, changeovers, claims
↓
Apps & Edge
Mill Console
Energy & quality twins
Converting Console
Schedules/AOI
Sustainability Cockpit
LCA, FSC
Operator Mobile
Checklists, roll scan
Module 1: Paper Machine Stabilization
Forming → press → dryer optimization
1
Hybrid Model
Grey-box moisture & basis-weight model with heat/mass balances
Learned residual via sequence models (TCN/Transformer)
2
Economic MPC
Steam box, vacuum, press loading, dryer section steam, reel speed control
Subject to sheet breaks and sheet draws constraints
3
Safe RL
SAC with control barrier functions for supervisory targets
Steam header pressure, specific steam consumption optimization
4
Objective Function
Minimize specific steam per tonne and variance(MD moisture, basis weight)
While meeting specs and formation constraints
Expected Results
5-10%
Steam Reduction
per tonne paper production
Tighter 2σ
Moisture Control
moisture bands at reel
MWh per tonne
Energy Efficiency
reduction in energy consumption
2025 Reviews
Literature Alignment
decarbonization measure classes
Control Parameters
Steam Box Control
Precise steam application for moisture control
Vacuum Systems
Optimized vacuum draw for dewatering
Press Loading
Controlled pressure for water removal
Dryer Section
Multi-zone temperature optimization
Optimization Constraints
Sheet Breaks Prevention
Maintain web stability throughout process
Sheet Draw Control
Optimal tension between sections
Quality Specifications
Meet basis weight and moisture targets
Formation Quality
Maintain fiber distribution and sheet structure
Module 2: Reel Genealogy & Zero-Defect Converting
High-speed CV and root cause analysis
1
Reel Map Creation
Stitch scanner profiles (CD/MD), lab tests, and AOI markings to 2D genealogy
Preserve slit/set history to chase chronic defects across production
2
High-Speed CV
Line cameras + strobe at corrugator and die-cutter
YOLOv8/SAM2 detect warp, delam, print mis-reg, score line breaks, glue skips
3
Domain Adaptation
Learning borrowed from industrial defect SOTAs
Specialized for paper surface anomalies and converting defects
4
Root Cause Mining
Gradient attribution correlates defects with process parameters
Flags upstream PM zones that seed defects in downstream converting
Defect Detection Types
Warp
High
Detection: Line cameras
Delamination
Critical
Detection: SAM2 segmentation
Print Mis-registration
Medium
Detection: YOLOv8
Score Line Breaks
High
Detection: Edge detection
Glue Skips
Medium
Detection: Thermal imaging
Surface Anomalies
Low
Detection: Domain-adapted CV
Process Parameters Monitored
Flute Temperature
Single-facer Pressure
Adhesive Solids
Rollout Position
Speed Settings
Moisture Content
Technical Implementation
2D Genealogy Mapping
Complete production traceability from reel to final product
Real-time AOI Integration
Continuous quality monitoring during production
Gradient Attribution
AI-powered root cause analysis for defect prevention
Upstream Zone Flagging
Identify paper machine sections causing downstream issues
Module 3: OR Kernels for Trim, Cutting, and Corrugator Scheduling
Mathematical optimization for operational efficiency
1
Trim/Cutting Stock
Matheuristic MILP + column generation for 2D guillotine patterns
Minimize trim-loss + changeovers with variable stock optimization
REFERENCE
Carton-box case studies guide parameterization
2
Corrugator Schedule
Sequence SKUs to minimize warm-up/cool-down, splices, flute swaps
Under delivery windows with rolling-horizon re-optimization every 10-15 minutes
REFERENCE
Multi-machine scheduling optimization
3
Packing/Bin-packing
2D cutting and packing for sheets/boxes
Recent 2D C&P advances to lift warehouse throughput and trailer fill
REFERENCE
Advanced warehouse optimization
Optimization Targets
-10-20%
Trim Loss Reduction
Material waste minimization
-15-25%
Changeover Time
Setup time optimization
+5-12%
Schedule Efficiency
OEE improvement
95%+
Delivery Compliance
On-time delivery performance
Mathematical Methods
MILP (Mixed Integer Linear Programming)
Optimal solution for cutting stock problems
Column Generation
Efficient handling of large-scale optimization
Matheuristics
Combination of exact and heuristic methods
Rolling Horizon
Dynamic re-optimization for real-time scheduling
Operational Benefits
Material Efficiency
Minimize waste through optimal cutting patterns
Energy Optimization
Reduce warm-up/cool-down cycles
Setup Minimization
Optimize changeover sequences
Delivery Performance
Meet customer delivery windows
Module 4: Sustainability - LCA + Chain-of-Custody
Operationalized environmental and compliance management
1
openLCA Integration
Compute format/structure alternatives with TRACI/IPCC/EF impact methods
EF 3.1 database integration ensures current factors for accurate LCA calculations
2
Custody Model
Encode FSC-STD-40-004 v3-1 mass-balance/segregation logic and labeling conditions
Audit trails attach to orders/rolls for complete chain-of-custody tracking
3
Designer Feedback
Real-time sustainability impact assessment for design decisions
Switch to 22 ECT B-flute + 30% rOCC; EF-GWP ↓ 11%, trim ↑ 0.7% (acceptable)
4
Compliance Automation
Automated FSC chain-of-custody documentation and reporting
Near-zero major non-conformances through process-encoded FSC rules
LCA Impact Methods
TRACI 2.2
EPA Tool for Reduction and Assessment
US EPA
IPCC 2021
Global Warming Potential factors
Climate
EF 3.1
Environmental Footprint database
EU Commission
ReCiPe 2016
Life cycle impact assessment
International
FSC Chain-of-Custody Requirements
Mass Balance Tracking
Segregation Controls
Labeling Verification
Audit Trail Documentation
Supplier Certification
Product Claims Validation
LCA Engine Features
Real-time Impact Calculation
Instant LCA for design alternatives
Database Integration
EF 3.1 and TRACI 2.2 current factors
Design Optimization
Sustainability-guided material selection
Audit Documentation
Complete traceability for compliance
FSC Compliance Features
Mass Balance Tracking
Complete material flow documentation
Segregation Controls
FSC vs non-FSC material separation
Labeling Verification
Automated label compliance checking
Audit Readiness
Near-zero major non-conformances
Expected Environmental Impact
-5-12%
EF-GWP Reduction
Per unit through substrate/format optimization
Near-zero
FSC Audit Findings
Major non-conformances through process automation
Real-time
LCA Integration
Design decisions with environmental impact
Module 5: GenAI Copilots
Grounded assistance with no hallucinations
1
Spec Copilot
"Draft spec for food-contact mailer 350×250×100 mm; cite internal SOP and FDA/ink limits"
Response
Pulls from controlled corpus; inserts reel genealogy tolerances and compliance requirements
Key Features
Controlled corpus access
SOP integration
FDA compliance
Genealogy tolerances
2
Changeover Copilot
"Proposes sequence & glue/temperature setpoints for next 4 jobs"
Response
Explains energy and quality trade-offs with citations to scheduler runs and historical data
Key Features
Sequence optimization
Setpoint recommendations
Trade-off analysis
Historical citations
3
Claims Copilot
"Auto-builds 8D reports with complete evidence trail"
Response
Backed by roll evidence, AOI frames, and FSC/CoC lot IDs with audit compliance
Key Features
8D report generation
Evidence compilation
AOI frame integration
FSC lot tracking
Grounded Knowledge Base
Internal SOPs (Standard Operating Procedures)
FDA regulations and ink limits
FSC chain-of-custody requirements
Historical production data
Equipment specifications
Quality standards and tolerances
Energy optimization guidelines
Compliance documentation
Grounded AI Features
Controlled Corpus
No external knowledge, only verified internal sources
Citation Tracking
Every response backed by specific documentation
Compliance Integration
Built-in regulatory and certification requirements
Historical Context
Leverages production history for recommendations
Operational Benefits
Specification Accuracy
Precise technical specifications with compliance
Changeover Efficiency
Optimized sequences with energy/quality trade-offs
Audit Readiness
Automated documentation for compliance audits
Knowledge Retention
Institutional knowledge preservation and access
Web & Mobile
Operator-grade UX for mill and converting operations
1
Mill Console
moisture/basis-weight heatmaps, specific-steam dashboards, predicted break risk
Moisture heatmaps
Specific steam dashboards
Break risk prediction
Energy optimization
Example
""Why high steam?" explainer points to press felt saturation or vacuum draw change"
2
Converting Console
corrugator Gantt with warm-up profiles; AOI defect feed with root-cause attribution
Corrugator Gantt charts
Warm-up profiles
AOI defect feed
Root-cause attribution
Example
"Trim-loss tracker with real-time optimization recommendations"
3
Sustainability Cockpit
LCA of design alternatives, recycled content, EF/TRACI footprints; CoC audit status
LCA design alternatives
Recycled content tracking
EF/TRACI footprints
CoC audit status
Example
"Mass-balance ledger with automated compliance reporting"
4
Operator Mobile
roll QR scan → genealogy & risk; camera assist for warp measurement; guided FSC paperwork
Roll QR scanning
Genealogy & risk access
Camera warp measurement
Guided FSC paperwork
Example
"Photo to structured record capture for chain-of-custody"
Key Platform Features
Real-time Monitoring
Live data from paper machines and converting lines
Mobile-First Design
Optimized for field operations and mobile devices
Compliance Integration
Built-in FSC and environmental reporting
User Experience Highlights
Mill Operations
• Intuitive dashboards for paper machine control
• Predictive analytics for break prevention
• Energy optimization recommendations
• Real-time quality monitoring
Converting Operations
• Visual scheduling with Gantt charts
• AOI defect detection and analysis
• Trim-loss optimization tracking
• Mobile roll scanning and genealogy
Data, Infrastructure & Deployment
Industrial-grade infrastructure for paper and packaging operations
| Layer | Stack | Notes |
|---|---|---|
Ingestion | OPC-UA/Modbus → Kafka; scanner/AOI frames to S3 | Real-time data collection |
Storage | Delta Lake + TimescaleDB; reel genealogy graph (Neo4j) | Time-series and graph data |
Feature store | Feast (per-reel/per-zone features; corrugator temps; glue solids) | ML feature management |
Training | PyTorch Lightning/Ray (control & CV); OR-Tools + commercial MILP | Model development |
Serving | Triton (CV/control); FastAPI planners; edge inference at corrugator | Production deployment |
LCA | openLCA engine with EF 3.1/TRACI/IPCC packages | Environmental assessment |
Governance | Model registry (MLflow), CFR-style audit logs, FSC ledger | Compliance and audit |
Security | RBAC; supplier partitions for CoC; on-prem/VPC option | Access control and data protection |
Real-time Data Processing
OPC-UA/Modbus integration with Kafka streaming for continuous data flow from paper machines, converting lines, and quality control systems.
Graph Database Integration
Neo4j for reel genealogy tracking and TimescaleDB for time-series data storage, enabling complete traceability from raw materials to final products.
Compliance & Security
CFR-style audit logs, FSC ledger management, and RBAC with supplier partitions for complete chain-of-custody documentation and security.
Deployment Flexibility
Edge Deployment
Triton inference servers and FastAPI planners deployed at corrugator locations for real-time optimization and minimal latency processing.
Cloud & On-Premise
Flexible deployment options supporting both cloud-based analytics and on-premise/VPC installations for sensitive data and regulatory compliance.
KPIs & Target Bands
Defensible improvements across all operational areas
| Area | KPI | Typical Improvement | Reference |
|---|---|---|---|
Energy | Specific steam (t steam / t paper) | −5–10% via deeper dewatering/steam control | Aligned to 2025 reviews |
Quality | Reel moisture σ; converting defect PPM | −25–40% / −30–60% (AOI + genealogy) | Computer vision and traceability |
Yield | Trim-loss / changeover minutes | −10–20% / −15–25% (OR kernels) | Mathematical optimization |
Throughput | Corrugator OEE | +5–12% (sequence + warm-up optimization) | Scheduling and energy optimization |
Sustainability | EF-GWP per unit | −5–12% (substrate/format swaps via LCA) | Environmental footprint optimization |
Compliance | CoC audit findings | Near-zero major NCs (process-encoded FSC rules) | Automated compliance management |
-10%
Average Energy Reduction
Specific steam per tonne paper
+12%
Throughput Improvement
Corrugator OEE optimization
Near-zero
Compliance Issues
FSC audit non-conformances
Measurable Impact
These improvements are defensible and measurablethrough established industry benchmarks, scientific literature validation, and real-world deployment resultsacross paper mills and converting operations.
Case Snapshots (Anonymized)
Patterned on real deployments across paper and packaging operations
01
Integrated Mill (Packaging Grades)
Economic-MPC on steam/vacuum/pressing stabilization
Specific Steam
↓ 7.8%
Moisture σ
↓ 33%
Sheet Breaks
−18%
Literature Alignment
2025 Reviews
Economic-MPC on steam/vacuum/pressing stabilized MD moisture; specific steam ↓ 7.8%; moisture σ at reel ↓ 33%; sheet breaks −18%. Gains match levers highlighted in 2025 energy-optimization reviews.
02
Regional Corrugator Network
Matheuristic cutting + rolling-horizon schedule
Trim-loss
−14%
Changeovers
−19%
OEE
+8%
AOI Detection
Early warp
Matheuristic cutting + rolling-horizon schedule: trim-loss −14%, changeovers −19%, OEE +8%; AOI caught warp early (glue solids drift). Literature-consistent improvements for carton plants.
03
Brand Packaging Redesign
openLCA run (TRACI/IPCC; EF3.1 datasets) on 4 alternatives
GWP Reduction
−9.6%
Cost Impact
Neutral
Material Selection
rOCC-heavy B-flute
Documentation
Audit-ready
openLCA run (TRACI/IPCC; EF3.1 datasets) on 4 alternatives: selected rOCC-heavy B-flute; GWP −9.6% with neutral cost; FSC mass-balance ledger produced audit-ready documentation.
Implementation Pattern
All cases demonstrate physics-informed controlto cut steam & breaks, OR-backed schedulesto kill trim, AOI to prevent escapes, and auditable LCA + FSC to win sustainable packaging bids with facts—not slogans.
Risks & Mitigations
Risk
RL/controller instability
Mitigation
Guard with MPC constraints & rate limits; shadow mode before activation
Risk
CV domain shift (new papers/inks)
Mitigation
Few-shot domain adaptation; active learning loop from AOI rejects
Risk
FSC/CoC data gaps
Mitigation
Enforce ledger completeness at goods-receipt; exception workflow tied to labeling permissions
Risk
LCA data licensing
Mitigation
Use client-licensed Nexus datasets; cache only references; reproduce results on demand
Closed-loop operating system for mills and converters
Tensorblue delivers a closed-loop operating systemfor mills and converters: physics-informed control to cut steam & breaks, OR-backed schedules to kill trim, AOI to prevent escapes, and auditable LCA + FSCto win sustainable packaging bids with facts—not slogans.
Key Differentiators
Physics-Informed Control
Cut steam & breaks with ML + MPC/RL
OR-Backed Schedules
Kill trim with mathematical optimization
AOI Prevention
Prevent escapes with high-speed CV
Auditable LCA + FSC
Win sustainable bids with facts
Result: unified intelligence platform — combining physics-informed control, mathematical optimization, computer vision, and environmental compliance for measurable operational and sustainability impact