CAM Pipeline Overview

Home / Engineering / System Design / CAM Pipeline

The CAM pipeline is Anvil's core system — it transforms a customer's CAD file into a validated, executable manufacturing plan. The pipeline runs against a standardized cell configuration (DN Solutions DVF 5000, Sinumerik One, 120-tool HSK-A63 library, 6061-T651 aluminum) so that every decision is made against known, fixed parameters.

Pipeline Flow

flowchart LR
    CAD["CAD Upload"] --> DI["Data Ingestion\n& Pre-Processing"]
    DI --> FR["Feature\nRecognition"]
    FR --> AICAM["AutoCAM\nPipeline"]
    AICAM --> SIM["Machine\nSimulation"]
    SIM --> EXEC["Machine\nExecution"]
    EXEC --> OPS["Operations &\nWorkflows"]
    EXEC -.->|"feedback"| AICAM
    EXEC -.->|"calibration"| SIM

Sub-Components

#	Component	Description	Detail
1	Data Ingestion & Pre-Processing	Convert input B-Rep formats into a standardized representation; extract PMI and 2D drawing information into 3D annotations	Entry point
2	Feature Recognition	Identify machinable features from part geometry using ML models (AAGNet, BRepMFR, BRepFormer, etc.)	ML-driven
3	AutoCAM Pipeline	Orientation planning, fixture planning, toolpath generation, stock removal simulation, and candidate plan generation — the most complex and highest-risk component	Core system
4	Machine Simulation & Validation	Full kinematic and controller simulation of top candidates; determines cycle times, validates NC programs, and ranks plans	Verification gate
5	Machine Integration & Data Capture	Transfer validated plans to the Sinumerik One controller, manage execution, and stream telemetry back for model training	Closes the loop
6	Operations & Workflow Integration	Job scheduling, machine queueing, and real-world workflow orchestration (soft-jaw 3D printing, stock preparation, shipping)	ERP interface

Key Design Principles

Standardized cell, constrained problem. The pipeline targets a single machine platform, single controller, single tool library, single material. This makes the learning problem tractable and allows software improvements to propagate fleet-wide.

Candidate generation and ranking. AutoCAM generates multiple candidate manufacturing plans. Each is validated through lightweight stock removal simulation (fast, inner loop) and then full machine simulation (thorough, outer loop). The best plan is selected based on cycle time, collision safety, and geometric accuracy.

Feedback loop. Machine telemetry from real cuts flows back to calibrate both the AutoCAM models and the simulation models. Cycle time predictions are compared against actuals. Spindle loads validate cutting physics. Inspection data validates geometric accuracy.

Support Tools

These are not part of the production pipeline but are critical for development:

• HOOPS Exchange — Handle diverse input B-Rep formats before feeding into the pipeline
• Annotation tool — Annotate training data for feature recognition models
• Virtual controller (Siemens VNCK) — Develop and test Sinumerik One interfaces before physical machine availability; also serves as a CI/testing environment