Title Blockspage one of findability

The boring square is busier than it looks

Australian engineering drawings live under two overlapping standards: AS 1100 (the Australian standard for technical drawing) and ISO 7200 (the international standard for title-block data fields). AS 1100.101 — approved on 25 August 1992 and published on 16 November 1992 — mandates ten essential components on every compliant title block: company name, drawing title, drawing number, revision number and date, scale, author, date created, checked-by, sheet number, and a reference to the drawing standard itself. ISO 7200:2004, published 15 February 2004 to replace its 1984 first edition, trims that list to eight strictly mandatory fields and adds eleven more as optional.

Most Australian mining and heavy-industry organisations follow AS 1100 in principle. In practice, vendor and contractor templates blend the two, add half a dozen client-specific fields, and silently mutate over the lifetime of a long-running asset. A title block on a 2003 substation drawing from one contractor does not look like a title block on a 2019 conveyor drawing from another, even when they describe equipment fifty metres apart on the same site.

That blend is not a problem in itself. AS 1100 was designed to be a floor, not a ceiling. The problem is what happens when an organisation tries to find drawings ten or fifteen years later, and discovers that the data their search depends on lives in pixels, not fields. And it is compounded by a quieter fact: the standards themselves stopped moving long ago.

The standards stopped moving decades ago

The reason title blocks drift is not only that vendors improvise. It is that the documents defining them were written for a drawing office, not a search index, and then largely left alone. The lineage is older than most of the assets it governs.

Read the table as one sentence. The rules that decide how a drawing is identified — AS 1100.101 and ISO 7200 — were fixed in 1992 and 2004 respectively. AS 1100.101 has not had a technical revision in over thirty years; it was merely reconfirmed in 2014. ISO 7200 has not moved since 2004. Both predate the modern EDMS, predate large-scale scanning programmes, and predate machine retrieval entirely.

Meanwhile the standard that governs how this data is actually used in operations — ISO 55001, asset management — was modernised in 2024. The expectations sitting on top of the title block moved on by thirty years. The definition of the title block did not. That gap is not a vendor problem; it is a structural one, and every organisation inherits it.

None of this makes the old standards wrong. AS 1100.101 and ISO 7200 still describe a sound floor. They were simply never asked to be machine-readable, never asked to absorb forty years of contractor variation, and never updated when retrieval — not draughting — became the job that mattered.

What is actually in a real title block

A worked example. An underground electrical drawing from an Australian coal operation, the kind of drawing a frontline electrician might pull up before isolating a piece of equipment, typically carries something like this in its title-block region.

Three observations from a single drawing.

This is what the title block actually contains. It is not a polite list of fields. It is the entire history of a piece of plant, compressed into roughly 1% of the drawing's surface area.

The 88% problem

A DrawingHub customer, an Electrical Engineering Manager at a large Australian operation, described the state of their drawing library after a recent EDMS migration like this:

That ratio, 88% generic and 12% useful, is not unusual. It is the rule, not the exception, for organisations that migrated paper or scanned PDFs into an EDMS in the 2000s and 2010s, when "metadata entry" meant a contractor typing the drawing number into a single field and ticking "done".

The library was uploaded. The library was technically searchable. The library was not findable.

The distinction matters. An EDMS report will tell you the percentage of drawings uploaded. It will not tell you the percentage of drawings discoverable by the right person, looking for the right thing, under time pressure. That second number is what costs you in safety, in downtime, in audit findings, and in the slow erosion of trust that ends with engineers maintaining their own folders on their own laptops because the official system "doesn't work".

The title block is where that gap lives.

Why title blocks defeat naive OCR

If reading title blocks were straightforward, this would not be an interesting problem. It is not straightforward. Six routine realities in Australian mining drawing libraries break naive OCR at scale, even modern and well-tuned OCR.

Template variation

A single asset's drawing library typically contains 30 to 100 distinct title-block templates, accumulated over decades of contractors, vendors, and CAD migrations such as Bentley, AutoCAD, and hand-stamped legacy prints. The fields are in different positions, with different labels, on every template family.

Multi-line titles

AS 1100 permits a title to wrap across multiple lines. Most templates use a hierarchical structure of system, scope, and sheet. OCR sees four strings. Without knowing the hierarchy, the four-line title becomes searchable as a single 150-character mash.

Crossed-out revisions

When a drawing moves from one revision to the next, older revision letters are often struck through. The current revision might be "D" but an OCR pass returns "ABCD". A pipeline that does not understand strikethroughs will index every drawing in your archive at the wrong revision.

Status stamps

"AS BUILT", "FOR CONSTRUCTION", "DRAFT", "ISSUED FOR REVIEW". These are often applied as stamps over an existing title block, partially obscuring the fields beneath. They are also the single most important indicator of whether a drawing should be trusted for operational decisions.

Sheet relationships

"Sheet 1 of 11" is straightforward. "Sheet 1 to 11" (a single sheet describing the relationship between an 11-sheet set) is a different statement entirely. Many legacy templates use the two interchangeably.

Rotation and scan quality

Mining site copiers from the 2010s scanned A1 prints rotated, skewed, and partially cropped. A typical legacy library has 5 to 15% of drawings rotated by 90 or 180 degrees, and another 10% with embedded bitmap regions that destroy text recognition.

Each of these problems is solvable individually. Most products handle two or three. The compounding effect, a rotated 2008 scan with a four-line title, three crossed-out revisions, an "AS BUILT" stamp over the revision block, and three competing ID systems, is what reliably breaks pipelines built for cleaner inputs.

How DrawingHub reads a title block

We treat title-block extraction as four problems, solved in order, not one.

1. 01Locate the title block, not the text

   A title block has a structure. Even when its contents are unreadable, its frame is detectable. We have trained our own technology to locate the right areas of a title block to target for extraction, so we work from the parts of the drawing that actually carry the metadata instead of guessing across the whole page.

2. 02Treat the template as a fingerprint, not as text

   Across a large drawing library, the same template repeats hundreds or thousands of times. We have developed a methodology that gives each title block a unique digital signature, so drawings that share a template are grouped together automatically. That grouping is what lets our human-in-the-loop process do bulk metadata extraction at scale, instead of treating every drawing as a one-off.

3. 03Reach consensus on critical fields

   For the key fields that decide whether a drawing is the right one (drawing number, revision, title, sheet number), we have developed an “agreement” process that uses many AI (large language) models to help DrawingHub converge on the data being extracted. Extracted metadata is only stored when a quorum of models agree on an answer. In an environment where you don’t have “ground truth” (the fully verified correct answer), this method enables DrawingHub to converge on an answer it can stand behind.

4. 04Read more than the title block

   The title block is the floor. It is the contract, the source of truth, and the part of the drawing we will not infer, guess, or quietly improve — if a value is missing we say so, and if it is inconsistent across sheets we surface it. Everything else worth extracting sits on top of that foundation: the discipline implied by symbology, the equipment hierarchy implied by the content, the relationship between sheet 1 and sheet 11 of an 11-sheet set. These are real signals, and operations teams need them — but none of them are safe to act on until the title block underneath is clean, complete, and reliable. Once that foundation is solid you can read the rest of the drawing and connect it outward: to the asset it describes, the project it belongs to, and the systems downstream that need to know about it. An inferred equipment hierarchy attached to a drawing with the wrong revision number is worse than no hierarchy at all — so the order matters.

This is the operational meaning of a sentence on our About page: every engineering drawing system can be significantly improved with better metadata, not just what is written in the title block, but what can be reliably inferred from the drawings and the context around them. The title block is page one. The rest of the drawing is page two.

Why findability starts at the title block

When an organisation approaches DrawingHub, the first artefact we produce is not an extraction. It is a findability report. The report opens with a single number: the Findability Score.

The Findability Score is the percentage of drawings whose title-block metadata is good enough that a frontline engineer can locate the drawing by searching for what it actually is. Not a per-field coverage table. Not a maturity index. One headline number, expressed against a defined sample of your library.

The interesting half of a findability report is not the findable bucket. It is the unfindable one. The report classifies failure into four patterns. Each pattern is a title-block problem masquerading as a search problem.

Single-word titles

A title block that contains only "INDEX", "COVER", or "SCHEMATIC". The drawing exists. The metadata exists. The metadata does not differentiate the drawing from a thousand others. In our sample report, single-word titles make up roughly 82% of the unfindable bucket.

Placeholder text

"TBA", "TBD", "XXXX", or a copy-paste of the previous drawing's title. The title-block field is filled, the validation rule passed at upload, the data is meaningless.

Administrative or meta documents

Cover sheets, drawing registers, transmittal forms uploaded into the drawing system during a migration. They have title blocks, but those title blocks belong to a different document class entirely.

Other and uncategorised

The long tail. Often where the most interesting fixable patterns hide once the obvious three buckets are stripped out.

Each of these failures surfaces in the title block. A single-word title is a title-block field that carries no signal. A placeholder is a title-block field that lies. An administrative document is a title block that belongs to a different document class. Catching the failure means looking at one specific region, on one specific shape, on one specific page. That is why every findability report starts where it starts.

A practical audit you can run yourself

A drawing manager who suspects their library has a findability problem can produce a rough version of the first page of a findability report in an afternoon. Pick 100 drawings at random from the EDMS, and answer five questions.

1. 01On how many of the 100 can you, by eye, locate the title block in under five seconds?
2. 02On how many can you read the drawing number, revision, and title without ambiguity?
3. 03How many distinct title-block templates are in your sample?
4. 04For ten randomly chosen drawings, pull the EDMS metadata. How many match the drawing's stamped values exactly?
5. 05For three drawings, attempt the search a frontline engineer would actually perform: by equipment tag, by area, by discipline. Do you find the right drawing first?

These questions take an hour. They are not a substitute for a real findability report and they are not statistically rigorous. They will give you, fast, a defensible answer to the question your operations team is already asking. Is this a search problem, or is this a metadata problem? If it is a metadata problem, the title block is page one.

A title-block reference built for machines, not for 1992

If the standards that define the title block were frozen before machine retrieval existed, the answer is not to wait for a committee to reconfirm them again in the 2030s. It is to build the layer the standards never had: a reference that starts from what AS 1100.101 and ISO 7200 already specify, then absorbs the forty years of vendor and contractor drift on top of them as first-class reality rather than as exceptions.

That is what DrawingHub maintains, and it is the position behind every findability report. We call it the DrawingHub Title-Block Reference. It is deliberately not a ratified standard, and we are careful not to dress it up as one. It is an open, living, machine-readable reference — a proposal, published in the open, designed so that an AI system reading an engineering drawing has something current to reference instead of a PDF last reconfirmed in 2014.

Four principles separate it from the documents it builds on.

It starts from the standards, it does not replace them

AS 1100.101 and ISO 7200 are the floor, not the enemy. The reference encodes their mandatory fields exactly, then maps the real-world identifiers — vendor numbers, EPCM job numbers, legacy migration traces — onto that base instead of discarding them.

It treats forty years of drift as data, not noise

Every contractor variant, multi-line title convention, struck-through revision pattern and status stamp is catalogued as a known shape. Variation is the subject of the reference, not an inconvenience it ignores.

It is written for machines first

The reference exists in a form an AI system can consume and cite directly — field names, allowed forms, and confidence rules — not as forty pages of prose a model has to interpret. It is the artefact our consensus extraction reads against.

It moves when the evidence moves

A standard reconfirmed once in two decades cannot track how drawings are actually labelled today. This reference updates as new template families and conventions appear in the libraries we read — continuously, in the open, with the change history visible.

This is the new way the title block has been waiting for: not a rejection of AS 1100.101 and ISO 7200, but the living, citable layer on top of them that four decades of practice — and machine retrieval — always needed. We publish it openly so it can become the thing an AI references when it has to read a drawing it has never seen before.

Where to go from here

The deeper the operations dependency on a drawing library, the higher the cost of metadata that almost works. Safety cases reference the wrong revision. HAZOP studies pause while someone walks the floor to confirm what is actually installed. Shutdown plans use the As Built that turned out not to be the As Built. None of these are search problems. All of them are title-block problems.

DrawingHub's findability report exists to make that distinction visible, and to give an organisation a baseline number it can improve against, drawing by drawing, template by template, revision by revision.

The boring square in the corner is doing more work than anyone gives it credit for. It deserves to be read properly.

Further reading

• Standard

  AS 1100.101-1992: Australian Standard for Technical Drawing — published 16 November 1992, reconfirmed 2014 (Standards Australia)↗

• Standard

  ISO 7200:2004: Title-block data fields — second edition, February 2004, replacing the 1984 first edition (ISO)↗

• Standard

  ISO 55001:2024: Asset management system requirements — second edition, July 2024, replacing the 2014 first edition (ISO)↗

• DrawingHub

  Drawing Management Principles

• DrawingHub

  How DrawingHub Works

• DrawingHub

  FAQ: What does a Findability Report actually tell us?