From rotary piercing mill to real-time operational intelligence

Who CP Industries is

Christy Park Industries — operating globally as CP Industries — is a heavy-industrial manufacturer headquartered in McKeesport, Pennsylvania. Founded in 1897 as a seamless tube mill, they are among the world's leading producers of large-diameter, seamless steel high-pressure containment vessels and gas cylinders.

Their pressure vessels operate in some of the most demanding environments on Earth:

• Aerospace & Defense — ground-support storage at NASA launch sites; high-pressure flasks for the U.S. Navy.
• Energy & Transportation — DOT-approved cylinders for tube trailers hauling compressed industrial gases across North America.
• Industrial & Chemical — large-scale storage for CNG, hydrogen, and petrochemical processing.

Every vessel CP Industries ships must survive ASME Section VIII, DOT 3AA/3T, and programme-specific NASA or Navy qualification. The margin for error is zero.

---

The operational challenges

1. Factory-floor telemetry with no unified home

CP Industries' McKeesport plant runs massive rotary piercing mills, multi-pass drawing benches, induction furnaces, and hydraulic test stands. Each station emits continuous sub-millisecond streams of vibration, thermal, and pressure readings. Today that data lives in isolated PLCs and SCADA historians. There is no single layer that can correlate a furnace temperature drift with the downstream wall-thickness variance it will cause three process steps later.

2. Compliance documentation is a paper mountain

Every cylinder produced generates:

• Material test reports (MTRs) for each heat of steel
• Ultrasonic and magnetic particle inspection logs
• Heat-treatment and normalising certificates
• Hydrostatic test records
• Serialised certificates of conformance for DOT, ASME, NASA, and Navy

These documents arrive in PDF, XLSX, XML, and proprietary mill formats. Matching them to the correct serial-numbered vessel — and confirming all required documents are present before a shipment is released — is largely a manual, error-prone process that creates bottlenecks on every government contract.

3. Production routing is unpredictable

Historical employee feedback from the McKeesport plant documents operational friction: heavy steel products are routed between manufacturing stations without a coherent, data-driven sequence. A billet that should flow Piercing → Drawing → Annealing → Straightening → NDT → Hydro-Test instead bounces based on machine availability signals that are never aggregated into a single routing decision layer. The result is excess material handling, extended cycle times, and utilisation imbalance across the plant floor.

---

What Purple8 does — concretely

Purple8 is a single-binary, in-process database that combines graph storage, vector search, full-text retrieval, document intelligence, and a workflow engine in one deployable unit. No Kafka. No separate vector DB. No external compliance microservice. One docker run.

Real-time telemetry and predictive maintenance

Purple8 Hyper Graph ingests high-frequency time- series data from mill sensors directly into the graph. The bounded memory management keeps peak memory usage hardware-limited regardless of how many data points accumulate — the same hardware that handles 2 million readings handles 20 million without re-provisioning.

Once in the graph, Purple8's Journey Engine models each cylinder's production lifecycle as a stateful workflow:

[Billet Cast] → [Rotary Piercing] → [Mandrel Drawing] → [Annealing]
→ [Straightening] → [UT Inspection] → [Hydro Test] → [Certified Ship-Ready]

Every stage carries SLAPolicy thresholds. If furnace temperature telemetry deviates beyond a configurable band during Annealing, the Journey Engine flags the cylinder instance, triggers a Human-In-The-Loop (HITL) review, and emits a graph edge SLA_BREACHED — immutably, before the part advances. Plant engineers see a real-time alert, not a post-mortem failure report.

The AEC structural module (aec/structural.py) and MEP flow analysis (aec/mep.py) — originally built for building information modelling — map directly onto mill equipment topology: critical machine relationships (which piercing mill feeds which draw bench), max-flow capacity constraints across production lines, and bridge-detection to identify single points of failure that would cascade a line stoppage.

Automated ASME and DOT compliance processing

Purple8 DocIntel (port 8200) parses every document format CP Industries produces — PDF test reports, XLSX material certs, XML mill data — and extracts structured entities (heat number, pressure rating, cylinder serial, inspection result, certifying engineer) with no template configuration required.

Those entities are committed back to the Purple8 Graph against the correct cylinder node via POST /ingest/commit. Compliance status becomes a live graph query:

Is cylinder SN-2847-X ship-ready for Aerospace contract A-7731?
→ Check: MTR present ✓  UT log present ✓  Hydro cert present ✓
  NASA NHB 1700.7 cert present ✗  → HOLD, flag for document re-submission

This replaces a manual clerk review on every government shipment with a sub-second graph traversal. Discrepancies surface before a crate is loaded, not after a Navy QA inspector rejects it at port.

Purple8's RAG pipeline (rag.hybrid_query) lets compliance engineers ask natural-language questions across the full document corpus:

"Show me all cylinders from Heat 44A that shipped to NASA KSC since January 2025 and confirm their hydrostatic test pressures."

The answer is retrieved from graph nodes and original source documents in one round-trip — no SQL, no Elasticsearch cluster, no separate document management system.

Predictive routing optimisation

Purple8's graph algorithms (core/algorithms.py) provide the decision layer for production routing:

• Shortest path (Dijkstra) across the station graph gives the minimum- handling route for a cylinder of a given OD and wall thickness.
• Betweenness centrality identifies which stations are current bottlenecks — the nodes every route passes through — so plant managers can target capacity investment accurately.
• Community detection (Louvain) clusters product families with similar routing profiles, enabling schedule batching that reduces furnace setups and tooling changeovers.
• Link prediction surfaces latent correlations between upstream process parameters (pierce ratio, mandrel speed) and downstream inspection failures, giving process engineers earlier leverage than end-of-line defect analysis currently allows.

The cost-based query planner selects the cheapest strategy for each routing query automatically — a PROPERTY_INDEX_SCAN for serial-number lookup, a TRAVERSAL for station-graph navigation, a HYBRID search for "find similar cylinders with the same failure pattern." No query tuning needed.

---

Architecture: what replaces what

| Today at CP Industries | With Purple8 | |---|---| | Isolated SCADA historians | One Purple8 graph — all telemetry co-located with product records | | Manual MTR / cert matching | DocIntel auto-extraction + graph compliance queries | | Spreadsheet-based routing decisions | Graph algorithms (shortest path, centrality, Louvain) | | Paper/email HITL review for anomalies | Journey Engine HITL gates — tracked, audited, immutable | | Separate compliance DB per contract | Multi-tenant namespacing — one engine, isolated per programme | | No predictive maintenance layer | SLA monitors + anomaly edges fire before failures cascade |

One process. One port. Zero new infrastructure services to operate.

---

Why the single-binary architecture matters for heavy industrial

CP Industries' IT environment is not a cloud-native startup. It is an industrial plant network with air-gapped segments, strict change-control windows, and OT/IT separation requirements. Purple8's single-binary model (pip install or docker run) means:

• One approval, one deployment — not eight services, eight change tickets, eight runbooks.
• On-premises or private cloud — no data leaves the facility. AES-256-GCM envelope encryption is on by default from the first GraphEngine() call.
• Bounded resource footprint — CortexAutoConfig solves the right memory geometry for whatever hardware the plant network provides. The engine runs on a 16 GB industrial server the same way it runs on a 128 GB cloud VM.
• MCP-native — an AI agent (Claude, Copilot, or an internal assistant model) can query compliance status, trigger routing recalculations, and resolve HITL holds entirely through natural language, without writing application code.

---

The bottom line

CP Industries has been manufacturing to zero-defect standards since 1897. The operational intelligence layer around their plant has not kept pace with the precision of the vessels they build. Purple8 closes that gap — not by adding complexity, but by consolidating what already exists into a single, AI-native operational backbone.

One binary. Sub-millisecond ingestion. Instant compliance.
Predictive routing. Full audit trail.
The same machine that handles today's production handles tomorrow's growth.