本文目前僅提供英文版本。
Picture four people at work on the same Tuesday morning.
A mechanical estimator has a bid due Friday. The set is 140 sheets with two addenda, and she needs every rooftop unit, VAV box, diffuser, damper and duct run counted and checked against the schedules before she can price anything.
A document controller at an EPC has just received a reissue of 60 P&IDs. Some changes are clouded, some are not, and the line list, the instrument index and the procurement packages all depend on knowing exactly what moved.
An LDAR coordinator at a gas processing plant has an audit in six weeks. Her inventory was built years ago, and since then the plant has been through a turnaround and a dozen MOCs.
A process safety engineer at a chemical plant is preparing a PHA revalidation. Before the team meets, he needs current P&IDs, the changes since the last study and the design data for every node.
Four industries, four regulatory regimes, four different deliverables. The first step of all four jobs is the same: read a set of drawings closely enough to trust a list built from them.
The same document, four ways
Engineering drawings differ by trade, but they are built from the same parts.
- Symbols on a legend. P&IDs typically draw on ISA 5.1 for instrumentation and on an owner's own symbol set for equipment and valves. Mechanical plans follow the engineer's conventions, usually laid out on a legend sheet. In both cases the legend on the drawing, not a textbook, is the authority.
- Tags with a grammar. A line number like 6"-P-1204-B2A packs size, service, sequence and piping class into one string. An instrument tag carries a function and a loop number. An HVAC tag like VAV-3-14 carries a floor and a sequence. Read the grammar and a tag becomes data; misread it and the error travels.
- Tables that carry the attributes. Equipment schedules, line lists, instrument indexes and air device schedules hold the capacities, sizes and models that the drawings only point to.
- Connections across sheets. Pipes and ducts run off one sheet and onto another through off-page connectors and match lines. Most of what matters, from what a relief valve protects to how much duct a floor needs, depends on following them.
- Revisions. Every set changes. The history of what changed, when and why is often worth as much as the current state.
So the reading problem has one shape everywhere: find the symbols in the drawing's own legend, resolve the tags, read the tables, follow the connections, and keep track of revisions. A tool that does this well for P&IDs has done most of the hard work for isometrics and mechanical plans too.
Where the industries really differ
The differences are real, but they live downstream of the reading.
| What the list is for | Who is accountable | What an error costs | |
|---|---|---|---|
| HVAC and mechanical | Pricing a bid, checking submittals | The estimator and project manager | Margin, or a lost bid |
| Construction and EPC | Building, changing and handing over | Project and document control | Rework, claims, a weak handover |
| Energy operations | Compliance and integrity programs | Operations and EHS | Findings, emissions, downtime |
| Chemicals and process | Process safety information | Process safety engineering | Safety gaps |
An estimate is a commercial commitment. An LDAR inventory is a regulatory record. Process safety information under OSHA's PSM standard has to be kept current by rule. Each of these sets its own tolerance for error and its own rules for sign-off. They belong in the workflows on top of the reading, where they can be configured per customer, not in four separate readers.
Why one layer, not four tools
The usual answer to four problems is four tools: a take-off product for the estimator, a document control system for the EPC, an LDAR database for the coordinator, a PHA package for the safety engineer. Each one has to read drawings, and each one reads them again.
Three things make a single layer the better design.
Verification is the expensive part. Machine reading is fast. Confirming it is not, because it takes an engineer's time. When a tag is verified once, every workflow that touches it can rely on that verification. Four tools means verifying the same tag four times, or not at all.
The same asset passes through all four industries. A process unit is designed by an EPC, built by contractors including mechanical and electrical trades, handed over to an owner, and then operated, maintained and changed for decades. Its drawings move with it. Every handoff where the next party has to re-read the documents is a place where knowledge is lost. A layer that carries verified data across those handoffs keeps it.
Point tools recreate the problem at every boundary. The take-off tool's counts don't reach the submittal review. The document control system's revision history doesn't reach the MOC. Each boundary is a spreadsheet export and a person reconciling two lists by hand.
What a layer has to get right
A shared layer is only worth having if the people accountable for the work can trust it. That sets a high bar.
- Read in the customer's legend. Symbol sets and tag conventions vary by owner, by engineer and by decade. The layer learns them from the documents instead of forcing a standard.
- Score everything and route doubt to people. Every item carries a confidence score. Anything below the threshold goes to an engineer with the sheet region open beside it, and the correction is recorded.
- Keep a graph, not a pile of extractions. Tags, lines, connections and revisions, linked, so that questions like "what does this change touch" are queries rather than searches.
- Attach provenance to every fact. Each value points back to the sheet, revision and region it came from, and to the person who verified it.
- Separate what must stay separate. Projects, companies and roles each see only what they should, and customer data trains nothing but that customer's own work.
We wrote about the verification step in why we verify every tag and about the graph in reading a P&ID like an engineer.
What changes for the four people
With the layer in place, each of the four still owns her or his job. What changes is where their hours go.
- The estimator starts from a complete take-off where every count opens the region of the sheet behind it. She checks the flagged items and the schedule mismatches, then spends the rest of the week on pricing and risk. See HVAC take-off.
- The document controller reviews a structured diff of the reissue, including the changes nobody clouded, and the register updates with each signed-off change. See revision control.
- The LDAR coordinator gets a reconciliation between the drawings and the inventory, with evidence for each discrepancy, and spends her six weeks on dispositions instead of discovery. See LDAR coverage.
- The process safety engineer edits a study pack with the changes since the last PHA, each linked to its MOC, and proposed nodes marked on the drawings. See PHA and HAZOP preparation.
In every case the person signs. The layer does the reading and the bookkeeping; the judgment stays where the accountability is.
Why this matters to owners of many companies
The shared-layer argument gets stronger as companies are combined. A private equity platform in industrial services might own a mechanical contractor, an electrical contractor, a fire protection company and a turnaround services business. Four trades, four sets of drawings, four ways of working.
Bought as point tools, that portfolio ends up with a dozen products and no common data. Built on one layer, each company keeps its own data, permissions and habits, while the workflows that work in one company can be rolled out to the next. We wrote about this in the missing layer in the industrial roll-up, and it is the reason our vision runs from one company to a portfolio.
The documents are the common ground
Industries organize themselves by what they do: heating and cooling buildings, building plants, moving gas, making chemicals. Underneath, they share a medium. Every one of them describes its physical work in drawings, schedules and specifications, and every one of them is limited by how fast people can read those documents.
That is why we build one layer rather than four products. The reading is the same problem. Solving it once, and verifying it once, is what lets everything on top of it be different.
