How to read a test bench specification document

If you work in procurement, design, or quality at an Indian PSU — HAL, BEML, NPCIL, BARC, one of the DRDO labs, Indian Railways, the Ordnance factories — you probably see between 20 and 50 test bench specification documents every year. You sign off on a few of them. The ones you sign off on go on to either work reliably for 15 years or they become the reason you are in a post-incident review at 11 PM on a Friday.

The gap between those two outcomes is almost never in the glossy first five pages. It is in the numbers that have units, the clauses buried on page 28, and a handful of things the supplier conveniently forgot to mention. What follows is a working engineer's reading list.

1. The 12 numbers that actually matter

These are the numbers that, if wrong or missing, mean the rig is the wrong rig. Everything else on the spec sheet is commentary.

1. Maximum working pressure, with its accuracy class

Every spec sheet leads with a pressure number. "1,800 bar." "15,000 psi." "700 bar." Useless in isolation. The number you actually want is pressure and the accuracy band at which the rig can hold that pressure — for example, 1,800 bar ± 0.25% of full scale. A rig that can reach 1,800 bar but only hold it to ±20 bar is not a 1,800-bar proof bench. It is a 1,800-bar party trick. For aerospace and defence proof testing, you usually need 0.25 percent FS or better. Anything quoted as "high accuracy" without a number is theatre.

2. Flow rate at working pressure (not peak)

Flow rate is only meaningful when it is paired with the pressure at which the flow is delivered. "400 lpm pump" tells you almost nothing. A pump that delivers 400 lpm at 50 bar is a gentle water feature. A pump that delivers 400 lpm at 350 bar is a 186 kW electrical installation, a 250 kVA transformer, and a water-cooling loop. Ask for the Q-P curve or at minimum the flow rate at the specified test pressure. If the supplier cannot produce it, they do not understand their own rig.

3. Temperature range of the test fluid

Every hydraulic fluid has a viscosity-temperature curve. If the test fluid runs outside a narrow band — typically 30°C to 55°C for mineral oil — your measurements drift, your seals swell or harden, and your dynamic response changes. The spec sheet should state the stable operating temperature range the rig controls to, the heating/cooling capacity (kW), and whether temperature is logged per cycle. "Ambient" is not an answer.

4. Dynamic response time, not "fast response"

Every supplier writes "fast response" on every rig. It means nothing. What you want is a step-change response time measured from a demand step to a settled output within the rig's accuracy band — for example, "0 to 1,200 bar in 0.8 s, settled to ±0.5% FS within 1.2 s." For impulse and endurance rigs this is the number that determines whether your test is valid or decorative.

5. Data logging rate, per channel and simultaneously

"1 kHz data logging" is two numbers, not one. Ask for the logging rate per channel, and critically, whether all channels log simultaneously or are multiplexed through a single ADC. A 16-channel rig logging at "1 kHz" via a multiplexer gives you 62.5 Hz per channel with a rolling timestamp offset. That is not the same rig the supplier described.

6. Number of simultaneous measurement channels

Physical I/O points, terminated at real hardware. Not "supported channels" (a software number) and not "addressable tags" (also a software number). The question is: how many pressure transducers, flow meters, thermocouples, load cells, and LVDTs can the rig read at the same instant? Count the terminal block, not the HMI dropdown.

7. Control architecture — fully stated

This is a line item that most spec sheets reduce to "PLC-based with SCADA." Insufficient. You need all four parts:

• PLC family and CPU — Siemens S7-1500, Allen-Bradley ControlLogix, Beckhoff CX, B&R X20, Mitsubishi MELSEC. The family determines spares ecosystem, programmer availability, and how long you are tied to one integrator.
• SCADA software — WinCC, FactoryTalk, Ignition, Wonderware, or an in-house framework. Each has different licensing, runtime costs, and IT-security posture.
• Source ownership — is the PLC source code delivered to you, in writing, at FAT? Or is it locked under a service contract that makes the supplier the only people who can modify it?
• Data export format — CSV, TDMS, MATLAB, OPC UA. If the only export is a printable PDF report, the rig is a dead end for any R&D workflow.

8. Safety envelope and containment rating

For any rig operating above 500 bar, ask for the containment rating in bar with a cited design code — ASME BPVC Section VIII, API 520, PD 5500, or an equivalent. "Heavy duty enclosure" is not an answer. Neither is "tested at factory." You want the hydrostatic qualification pressure, the burst safety factor, and the vent sizing calculation. On a 1,500-bar bench, a containment failure produces fragments at supersonic velocity. This is not a paragraph to skim.

9. Calibration chain traceability

Every primary transducer on the rig — pressure, flow, temperature, force, displacement — should carry a calibration certificate traceable to NABL (or equivalent ILAC-MRA signatory) primary standards. The certificate must name the reference instrument, the date, the uncertainty, and the laboratory accreditation number. If the supplier says "calibrated in-house" without naming a primary standard chain, the certificate is not useful in any audit that matters.

10. Repeatability across cycles

Accuracy tells you how close the rig gets to the target once. Repeatability tells you whether 1,000 identical cycles produce 1,000 identical results. Ask for the standard deviation of measured pressure across 1,000 test cycles at rated pressure. Any supplier serious about endurance or production testing will have this number. Suppliers who have never measured it will change the subject.

11. Cycle time, door-to-door

The cycle time that matters is not the pressurise-and-hold time. It is the full cycle: part load, clamp, pressurise, dwell, depressurise, unclamp, part unload, and any automated purge or reset steps. A rig that pressurises in 8 seconds but takes 4 minutes to load and unload has a 4-minute cycle time. Your production planner needs that number, not the pressurisation number.

12. Utilities: electrical, water, compressed air

The rig does not exist in a vacuum. Ask for:

• Connected electrical load (kW) and the feed you need at your site (415 V 3Φ 50 Hz is standard; verify).
• Cooling water flow rate, inlet temperature, and return pressure.
• Compressed air consumption at SCFM and line pressure.
• Hydraulic fluid volume on first fill and the grade required.

Missing these numbers from the spec sheet is the single most common reason a rig delivered on time sits in the shed for six months while civil and electrical works catch up.

2. The 40+ numbers that are noise

Not everything on a spec sheet is load-bearing. A significant share of what you read is decoration — numbers that look precise, sound technical, and mean nothing in isolation. Here is a working list of what to quietly ignore, or at least deflate.

Display and HMI theatre

• "Resolution 0.01 bar" — cosmetic. If the transducer accuracy is ±1 bar, the last two decimal places are noise. Resolution without accuracy is theatre.
• "21-inch capacitive touch-screen HMI" — so? Screen size is not an engineering metric. Ask for the SCADA tag list and the screen-transition response time.
• "User-friendly GUI" — every supplier writes this. None of them can prove it. Ask to operate the rig yourself for 15 minutes during FAT.
• "Multi-language support" — it is Windows. It supports multi-language. This is not a feature.
• "Password-protected access levels" — mandatory in 2026, not a differentiator.

Software number salad

• "2,000 channels supported" — a software limit. Your rig has 12 physical I/O points. The 2,000 is marketing.
• "Unlimited test recipes" — disk space is cheap. "Unlimited" means nothing.
• "Cloud-ready architecture" — a phrase. Ask what ports, what protocols, what IT security model.
• "AI-powered analytics" — a moving average with a dashboard. Unless the supplier can name the model, the training data, and the false-positive rate, it is a buzzword.
• "Industry 4.0 compliant" — Industry 4.0 is not a standard. There is nothing to be compliant with.

Meaningless generics

• "PLC-based" — every rig built after 1995 is PLC-based. Which PLC?
• "Closed-loop control" — as opposed to open-loop control, which nobody builds. Which controller? Which tuning method?
• "Servo-hydraulic" — ask for the servo valve make, response time, and spool hysteresis. Otherwise it is a category, not a specification.
• "Robust construction" — no engineering content.
• "Heavy-duty frame" — what wall thickness? What steel grade?
• "Precision instrumentation" — same rule: what class, traceable to what standard?

Bloated certification lists

• "CE, ISO, RoHS, REACH, UKCA" — three of these apply to the rig and two to the paint. Ask which directive under CE and which harmonised standard.
• "Compliant with international standards" — name the standards.
• "Certified manufacturer" — certified by whom, to do what?

Numbers with no units

• "High flow", "ultra-high pressure", "low noise", "minimal footprint", "fast cycle" — all of these should have a number and a unit. If they don't, they are not specifications.
• "Up to X" — the most dangerous phrase on a spec sheet. "Up to 1,800 bar" means it reached 1,800 bar once, on a good day, with no instrument attached. Ask for the continuous rating.

Taken individually, any one of these is a harmless marketing choice. Taken together, they are a technique. A supplier who fills 70 percent of a spec sheet with items from this second list and 30 percent with items from the first list is hoping you will count total lines rather than total information.

3. The fine print that matters more than the specs

The technical specification is typically pages 3 through 20. The commercial terms are typically pages 22 through the end. In most failed deliveries we have seen, the technical spec was broadly correct and the commercial terms were not. Read this section twice.

The inclusion / exclusion list

Every quotation has an exclusion list. It is usually the shortest section and the most expensive one. Look for:

• Civil foundation works — included or not?
• Electrical cabling from the incoming MCC to the rig — who scopes it?
• Cooling water supply line — from where to where?
• Compressed air piping — same question.
• Hydraulic fluid first fill — quantity and grade, and whose cost?
• Spare transducers and seals for the first year — included or separate PO?
• Calibration standards (master gauge, reference transducer) — included or separate?
• Packing, forwarding, freight, insurance — Ex-works, FOR-destination, or CIF?
• Taxes, duties, GST — in the headline price or added later?

A headline price that is 15 percent lower than the nearest competitor usually reconciles exactly to the exclusion list. Tally it before you benchmark.

Force majeure wording

Force majeure clauses are often asymmetric. The supplier gets broad relief for any event beyond their reasonable control; the buyer gets narrowly defined relief. Read both sides of the clause. They should be symmetrical. If the clause protects the supplier from pandemic, shipping delay, component shortage, and "any event of similar nature", it should protect you the same way for your civil works, your site readiness, and your inspection delays.

Spare parts lead time

Ask for a written spare-parts lead time matrix — what component, what lead time, what price. Critical spares (servo valves, pressure transducers, main pumps) should be ex-stock or 4 weeks maximum from an Indian warehouse. If the lead time is "8 to 16 weeks from factory", the rig has an MTBF problem waiting to happen.

PLC code ownership

Non-negotiable for defence and DRDO customers and quietly non-negotiable for everyone else. You want the PLC program, the HMI project, the SCADA tag database, and the commented ladder or structured text delivered at FAT, in writing, with the right to modify. If the supplier says "we deliver an executable, not the source", the rig becomes unmaintainable the moment their business model changes.

Training hours included

Training is usually an afterthought and usually under-scoped. A realistic baseline for a PLC-SCADA test bench is 5 days of operator training and 3 days of maintenance training, on-site, for up to 6 participants, with a written syllabus and signed attendance. Any less than that and the rig will be misused in its first month.

FAT and SAT scope

Factory Acceptance Test and Site Acceptance Test should each have a written protocol agreed in advance, a line-by-line test matrix, and a defined duration. "FAT: 2 days" is not a scope. The protocol should cover pressure accuracy, flow accuracy, repeatability, alarm functions, emergency stop, data export, and a dry run of every test recipe you will use in production.

Warranty exclusions and start date

Warranty is where suppliers play the longest game. Two traps:

• Start date. Warranty should start at SAT completion, not at dispatch from the supplier's factory. The difference is typically 6 to 14 weeks, and it is a deliberate choice in every contract we read.
• Exclusions. "Consumables", "wear parts", "mis-use", and "operator error" are all defensible in principle and often abusable in practice. Ask for the exclusion list to be closed — i.e. anything not on the list is covered.

4. A real-world red-flag checklist

Eight things that should make you slow down, ask questions, and probably walk away.

5. The spec-sheet game, decomposed

Here is a snippet of the kind of thing you will read in a quotation. Every phrase is drawn from real spec sheets, anonymised.

"State-of-the-art hydraulic test bench with 15,000 psi operational pressure, PLC-based SCADA running on 21-inch touch screen HMI with 1,000 Hz data logging and 250 channels supported. Industry 4.0 compliant, cloud-ready, with AI-powered analytics. Robust, heavy-duty construction. User-friendly interface. Fully automated. CE, ISO, RoHS certified. Suitable for all industries."

Let us decompose, left to right:

• "State-of-the-art" — filler. No engineering content. Delete.
• "15,000 psi operational pressure" — a real number (roughly 1,034 bar). Useful, but missing the accuracy class, the working-pressure flow rate, and the continuous vs peak distinction. Ask for all three.
• "PLC-based SCADA" — means nothing. Which PLC? Which SCADA?
• "21-inch touch screen HMI" — screen size. Not a specification.
• "1,000 Hz data logging" — per channel or aggregate? Simultaneous or multiplexed? Without the answer, this is a headline number, not a capability.
• "250 channels supported" — software ceiling. Ask for the physical I/O count.
• "Industry 4.0 compliant" — Industry 4.0 is not a standard. Delete.
• "Cloud-ready" — what protocol? What IT security? What data ownership model? Without answers, a phrase.
• "AI-powered analytics" — buzzword. Ask the supplier to name the algorithm.
• "Robust, heavy-duty construction" — no units. Ask for frame member dimensions, steel grade, and finite element analysis summary.
• "User-friendly interface" — unverifiable. Operate it yourself at FAT.
• "Fully automated" — automated from where to where? Part load? Clamping? Purge? Reset?
• "CE, ISO, RoHS" — which directive under CE, which ISO number, is RoHS even relevant for a stationary industrial rig?
• "Suitable for all industries" — no test bench is suitable for all industries. This is either a lie or a confession.

Of the 15 phrases in that paragraph, one is a real engineering number, two are numbers that need clarification, and 12 are filler. The supplier has written 65 words to deliver about 3 words of information. That ratio — real-to-filler — is the fastest way we know to rank competing proposals.

6. What a good spec sheet looks like

Good test bench specifications are boring to read. They are usually structured as:

• Numbered requirements. "R-3.2.1: Maximum working pressure shall be 1,800 bar, continuous rated, accuracy 0.25% of full scale, measured per ISO 10012."
• A separate acceptance criteria section. Each requirement has a matching test step that proves it. "AC-3.2.1: Acceptance shall be demonstrated by 10 consecutive cycles at 1,800 bar, held for 30 seconds, deviation recorded at 10 Hz, standard deviation below 4.5 bar."
• Citations to standards. Every numeric requirement references the standard it derives from — ASME BPVC, ISO 10012, MIL-STD-1522, NABL, API 520 — with the section number.
• Minimal adjectives. Words like "excellent", "state-of-the-art", "best-in-class" do not appear. Adjectives that cannot be measured are not useful in a specification.
• An exclusion list as explicit as the inclusion list. What is not included is spelled out, not implied.
• A dated revision history. Specifications change. A document without a revision table is a document nobody owns.

If you receive a specification like the above from a supplier — even a supplier you do not know well — you are working with people who take the rig seriously. If you receive 40 pages of adjectives and a bill of materials with no units, you are working with people who take the quotation seriously.

7. Closing notes from the shop floor

We build these rigs. We also operate them for our own internal testing, and every now and then we are on the buying side of one. The pattern is always the same: the specification is where the contract is decided. Everything after — PO placement, kickoff meetings, FAT, SAT, warranty — is litigation of the specification.

If you take one thing away from this post, let it be this: a specification document is not a brochure and it is not a contract, but it becomes the terms of both. When the rig is on your floor, the only thing anyone will argue about is what the specification said. So read it the way you would read a legal contract, because in three years, in an audit or a post-incident review, that is exactly what it will be.

The 12 numbers at the top of this post are, in our experience, the ones that decide whether the rig works. The exclusion list, PLC code ownership, warranty start date, and change-order mechanism are the four clauses that decide whether it arrives on time and can be maintained for 15 years. Everything else is commentary.

If you are earlier in the process — still deciding which suppliers even belong on the shortlist — our longer pillar guide on how to choose a test bench manufacturer covers supplier evaluation from installed-base checks through factory audit to contract signature. It is the companion piece to this post.