CRUXNEXUS · CANS INDUSTRIAL SOFTWARE PLATFORM
Industrial Digital Twin, Operational Intelligence & AI Process Optimisation
CruxNEXUS connects plant-floor data, engineering context, 3D digital twins, operational analytics and AI-assisted optimisation in one industrial software platform.
It is designed to work with the existing PLC, SCADA, BMS, instruments, historians and enterprise systems — adding intelligence without replacing the control infrastructure that already protects and operates the plant.
ONE PLATFORM · FROM FIELD DATA TO ENGINEERING ACTION
Built Around the Existing Plant
CruxNEXUS does not require a greenfield automation architecture. It can sit above existing plant systems and progressively add visibility, context, analytics, AI and authorised optimisation.
Plant Systems
PLC, BMS, SCADA, instruments, meters, gateways and historians.
Data & Asset Layer
Point integrity, units, historian, asset hierarchy, process relationships and engineering limits.
Operational Model
2D/3D plant context, equipment state, KPI overlays, alarms and spatial diagnostics.
Analytics & AI
OEE, energy, anomalies, forecasting, root cause, optimisation and explanations.
Human / Closed Loop
Recommendations, approval, controlled setpoint write-back, audit and verification.
CORE PLATFORM CAPABILITIES
Industrial Software That Connects Operations, Engineering and AI
3D Digital Twin
Browser-based plant visualisation with reusable equipment models, live equipment states, display boards, alarms, drill-down popups, layer control and kiosk views.
OEE & Production Intelligence
Machine states, counts, cycle information, downtime analysis and Availability / Performance / Quality calculations linked to production context.
Energy & HVAC Optimisation
Plant KPIs, baselines, expected-energy models, forecasting, optimisation, savings verification and controlled supervisory interaction with existing BMS/PLC systems.
AI / ML & Forecasting
Plant-trained analytical models, time-series forecasting, anomaly detection, model comparison and task-specific deployment using the model that best fits the operating problem.
Engineering AI & LLM
RAG, engineering Q&A, SOP/manual retrieval, report generation and explanation of validated findings. The LLM is an operator/engineering interface — not the plant control algorithm.
Open Industrial Integration
Designed for integration with PLC/SCADA/BMS, OPC, Modbus, BACnet, APIs, SQL/time-series data, edge devices and enterprise systems according to site architecture.
AI MODELS HAVE DIFFERENT JOBS
Forecast, Optimise, Explain — Without Blurring the Roles
ML Forecasting · TFT + N-BEATS
Temporal Fusion Transformer and N-BEATS can run in parallel for time-series forecasting. They are machine-learning forecasting models, not the optimisation engine.
AI Optimisation · Bayesian Optimisation
Bayesian optimisation searches the feasible operating space for improved setpoints or equipment combinations while respecting approved constraints.
Explainability · SHAP
Site-calibrated SHAP explanations identify the variables that contributed most strongly to a forecast or model result so engineers can interrogate the recommendation.
Engineering Interface · LLM / RAG
Local or approved cloud LLMs can support engineering Q&A, document retrieval, report generation and explanation. They remain outside the deterministic plant control loop.
POTENTIAL AI / OPTIMISATION MODEL LIBRARY
Use the Right Model for the Industrial Problem
CruxNEXUS can host different model families and optimisation engines according to the use case. The technologies below are potential deployment options, selected and validated per project; they are not all enabled by default and should not be interpreted as one universal AI stack.
VISION INSPECTION
Detection, Segmentation & Anomaly Models
YOLO / RT-DETR for fast object and defect detection; DINOv2 / ViT / ConvNeXt for classification and robust visual feature extraction; SAM 2 / SegFormer / U-Net for pixel-level segmentation; PatchCore / PaDiM / FastFlow for low-sample anomaly detection; PaddleOCR / TrOCR for markings and labels; and CLIP / SigLIP or approved vision-language models for few-shot semantic checks and complex QA interpretation.
Typical use: defect detection, missing components, dimensional/visual checks, surface anomalies, OCR and automated PLC reject decisions.
AUTO-SCHEDULING
Constraint & Multi-Objective Scheduling
CP-SAT / MILP are the primary constraint-optimisation engines for production, manpower, maintenance and resource schedules. Genetic / NSGA-II methods can handle difficult multi-objective trade-offs. Reinforcement learning may be considered only where a validated simulator exists and adaptive dispatching adds value.
Typical use: machine sequencing, shift allocation, job dispatch, maintenance windows, AGV/AMR assignments and production recovery.
ERP & BUSINESS PREDICTION
Forecasting & Risk Models
XGBoost / LightGBM / CatBoost are practical for lead-time, late-order, quality, purchasing and stock-out risk prediction. TFT, N-BEATS, DeepAR or time-series foundation models can support demand, inventory, energy, cash-flow or throughput forecasting where sufficient history exists.
Typical use: demand planning, purchasing, inventory optimisation, delivery-risk alerts, supplier performance and production/ERP forecasting.
DECISION INTELLIGENCE
Explainable & Causal Decision Support
Bayesian networks, causal forests / uplift models, tree ensembles and probabilistic risk models can separate likely drivers from simple correlation and quantify operational risk. Monte Carlo or discrete-event simulation can test scenarios before action, while Pareto / multi-criteria optimisation ranks alternatives when cost, energy, quality, throughput, service level and risk conflict. Contextual bandits may be evaluated for low-risk adaptive recommendations with measurable feedback.
Typical use: root-cause support, capex prioritisation, operational trade-offs, scenario ranking and management decision support.
CONDITION & ANOMALY
Predictive Maintenance Models
Isolation Forest, autoencoders, 1D CNNs and transformer-based sequence models can complement engineered vibration, electrical and process features. Classification models can then combine condition evidence with operating context and maintenance history.
Typical use: motor health, vibration anomalies, bearing degradation, process drift, abnormal energy signatures and early-fault detection.
PROCESS OPTIMISATION
Optimisation, MPC & Safe Adaptive Control
Bayesian optimisation is suitable where evaluations are expensive and constraints matter. Model Predictive Control can coordinate multivariable dynamic processes where a validated model exists. Safe reinforcement learning is a later-stage option for selected applications, not the default plant-control method.
Typical use: HVAC, energy, utilities, recipe tuning, throughput/quality optimisation and supervisory closed-loop control.
MES & PRODUCTION FLOW
Process, WIP & Quality Intelligence
Process-mining models, sequence transformers / TCNs and XGBoost / LightGBM can predict WIP completion, cycle-time deviation, bottlenecks, scrap or rework risk and abnormal routing. Graph-based models can represent dependencies between machines, operations, lots and materials when the process network is important.
Typical use: WIP ETA, bottleneck prediction, dynamic dispatch support, yield-risk alerts, traceability exceptions and production recovery.
WMS & INTRALOGISTICS
Inventory, Slotting & Dispatch Optimisation
LightGBM / CatBoost, TFT and probabilistic forecasting can predict demand, replenishment and stock-out risk. CP-SAT / MILP / vehicle-routing solvers are appropriate for slotting, wave planning, picking routes, dock allocation and AGV/AMR dispatch. Graph models or reinforcement learning may be evaluated for highly dynamic routing problems.
Typical use: inventory positioning, replenishment, slotting, picking-route optimisation, dock scheduling and autonomous material movement.
CMMS & ASSET RELIABILITY
Failure Risk, RUL & Maintenance Intelligence
Weibull / Cox survival models, Random Survival Forests, gradient boosting, TCN / LSTM / transformer sequence models can estimate failure probability or remaining useful life. Embeddings and LLM/RAG can classify work orders, retrieve maintenance knowledge and summarise recurring failure patterns.
Typical use: predictive maintenance, PM optimisation, work-order triage, failure-risk ranking, spare-parts planning and maintenance knowledge retrieval.
Engineering rule: optimisation solvers such as CP-SAT, MILP and MPC are not machine-learning models, but they belong in the same CruxNEXUS decision stack because many industrial scheduling and control problems are solved more reliably by explicit constraints than by a neural network. LLMs and vision-language models remain advisory/interface components unless a specific deterministic action path has been engineered and validated.
3D DIGITAL TWIN DEMONSTRATIONS
See the Plant, Not Just the Dashboard
CruxNEXUS 3D views combine equipment relationships, live state, process flow, alarms, KPI boards and AI context in an engineering-oriented spatial view.
The public demo gallery is designed to accept the current CruxNEXUS WebM reference clips directly and, once published to the CANS YouTube channel, can be switched to the automatic YouTube feed without changing the page layout.
CruxNEXUS 3D Demo Gallery
Selected CruxNEXUS 3D digital-twin demonstrations will be published here as part of the CANS engineering video library.
CONTROLLED DEPLOYMENT
From Visibility to Autonomous Optimisation
01 · OBSERVE
Baseline
Validate data, understand operating modes and establish plant behaviour without changing control.
02 · RECOMMEND
Advisory AI
Generate recommendations with expected impact, evidence and explanation for engineering review.
03 · SUPERVISED
Approved Action
Apply selected recommendations with operator approval, engineering limits and audit logging.
04 · AUTONOMOUS
Closed Loop
Authorise appropriate variables for automatic optimisation only after validation, with interlocks and fallback retained.
ONE PLATFORM · MULTIPLE INDUSTRIAL APPLICATIONS
Apply the Same Data and Digital-Twin Foundation Across the Plant
Manufacturing OEE · MES and production intelligence · ERP forecasting · WMS and intralogistics optimisation · CMMS and predictive maintenance · auto-scheduling and dispatch · AI vision inspection · HVAC and chilled-water optimisation · water and utilities · ports and infrastructure · energy intelligence · remote operations.
Start With the Operational Problem — Then Configure the Platform Around It
CruxNEXUS is configured around the actual plant architecture, available instrumentation, data quality, operating constraints and business objectives. The software layer can begin with visibility and analytics, then expand into digital twins, AI and controlled optimisation as the site is ready.