Auditing a Chinese Robot Supplier: Documents and Gates
At a glance: A supplier audit that is not written into the purchase order is a favour, not a control. The ten documents to request as issued files, the six production gates worth an inspector's time, the sampling numbers that decide how many units to open, and the three contract clauses that decide whether a defect becomes a warranty repair or an argument.
The Problem With Auditing a Supplier You Have Never Met
A European buyer placing a first fleet order with a Chinese robot manufacturer is usually asked to accept one of two positions: fly out for a two-day factory visit, or buy on the strength of a catalogue and a certificate PDF. Neither is a real audit. A two-day visit shows a showroom and a production line running at demonstration pace. A certificate PDF proves that a management system was certified at some point, not that the unit on your purchase order was built inside it.
The workable middle is a document audit plus a set of discretionary gates you specify in the purchase order and let a third-party inspector execute. That combination is cheaper than a flight, it is repeatable across every subsequent order, and it produces records you can hand to your own quality team. This guide lays out the document set to request, the six production gates worth paying an inspector to stand at, the arithmetic behind the sampling plan, and the three contract clauses that decide whether a defect becomes a warranty claim or an argument.

The Document Set: What to Request Before Any Money Moves
Request this list in one email and ask for the documents as issued files, not screenshots. Screenshots cannot be checked for validity dates, scope statements or the certificate holder's legal entity name. Five of the most common audit findings are entity mismatches: the certificate is held by a trading company while the invoice comes from a factory, or the scope statement covers "electronic assembly" and never mentions robotics.
| Document | What it actually proves | The check that catches problems |
|---|---|---|
| Business licence | Legal entity name, registered address, scope of business, registered capital | Entity name must match the invoice issuer and the certificate holder exactly. A trading company invoice against a factory certificate is a different supplier than the one you audited |
| ISO 9001 certificate | That a quality management system was audited by an accredited body | Issuing body must appear on IAF CertSearch or the CNAS register. Check scope wording and expiry, not the logo |
| Factory audit report (if any) | Whether a third party has already physically inspected the site | Any report older than 12 months should be treated as a starting point, not evidence. Ask which findings were closed and when |
| UN 38.3 test summary | That the lithium cells and the assembled pack passed the transport test series | Must be a test summary, not a certificate of conformity. Check the cell manufacturer and pack model, and that the revision is current |
| IEC 62133-2 report | Cell and pack safety under the portable-lithium standard | Report must name the pack model you are buying, not a sister product |
| CE documentation package | Which directives were applied and who signed the declaration | Declaration of Conformity must name the exact model and the manufacturer's authorised representative in the EU. A generic declaration covering "robotic equipment" is not usable in a customs or market-surveillance conversation |
| EN ISO 3691-4 test evidence | Driverless industrial truck safety functions | Ask for the specific clauses tested and the results, not a claim of compliance |
| Battery cell datasheet and cycle data | Chemistry, capacity, charge and discharge limits, cycle life at depth of discharge | Datasheet must trace to the cell maker. A pack datasheet with no cell reference is a red flag |
| Spare parts price list and lead times | What a year-three repair actually costs | Request it as a priced annexe to the contract, valid for the warranty term, with a stated escalation cap |
| Sub-supplier list for critical components | Concentration risk on lidar, drive units, batteries, controllers | Look for a single-source component with a single factory. This is the failure mode that produces a six-month stoppage |
If the candidate supplies its own battery packs rather than buying them, add the pack-level design record to that list. The battery pack specification and sourcing guide covers the specific questions to put to a pack builder, and the spare parts and consumables planning guide explains how to convert the spare-parts annexe into a stock commitment that survives year three.
Two documents in that list do most of the work. The business licence catches trading-company substitution, which is the single most common way buyers end up with a supplier that cannot support the product. The UN 38.3 test summary catches battery substitution, which is the single most common way buyers end up with a fleet that cannot legally be shipped or that fails early. If a supplier hesitates on either, that hesitation is the finding.
Six Gates Worth an Inspector's Time
A full factory audit follows a checklist of a hundred items and costs several thousand dollars. For a fleet order of five to fifty units, six gates capture most of the risk. Each gate is a point in the process where a physical check produces a yes or no answer, and each can be written into the purchase order as an inspection right.
Gate 1: Incoming component verification
Ask for the incoming inspection record for the specific drive units, lidar modules and battery packs allocated to your order, matched by serial number or lot. The question to answer is not whether incoming inspection exists but whether the components in your units are the ones named in the approved bill of materials. Sample five serials and trace each back to a purchase receipt.
Gate 2: Battery pack assembly and testing
This is the highest-value gate on the list. Witness or review the pack-level test record: capacity at a stated C-rate, internal resistance, insulation resistance, and the result of the BMS over-voltage and over-temperature trip test. Pack build quality determines both the shipping legality and the second-year failure rate, and it is the step most often subcontracted to a third party without disclosure.
Gate 3: Chassis and drive assembly
Check drive unit torque settings against the specification, wheel alignment, and the record of the run-in test. A misaligned drive shows up nine months later as uneven tyre wear and a navigation drift the supplier will attribute to your floor.
Gate 4: End-of-line functional test
Review the end-of-line test sheet for your units: lift stroke and repeatability, navigation and localisation under the factory's test course, emergency stop response time, bumper and lidar safety-field trip test, and a full charge-discharge cycle. The E-stop response time is worth an actual measurement. It is cheap to test, it is a direct safety claim under EN ISO 3691-4, and it is the number suppliers are most tempted to estimate rather than measure.
Gate 5: Ageing and burn-in
Ask how long each unit ran before packing and what the pass criteria were. A four-hour burn-in catches assembly defects that a fifteen-minute functional test does not. If a supplier has no burn-in step, that is a finding, not a gap in your checklist.
Gate 6: Packing and shipping configuration
Verify the battery state of charge at packing against the shipping requirement, the UN 38.3 documentation attached to the shipment, the crate drop protection, and the moisture barrier. A unit that arrives with a battery at transport-prohibited state of charge creates a customs problem at your end, not theirs.
| Gate | Record to demand | Failure signature it prevents | Inspection effort |
|---|---|---|---|
| 1. Incoming components | Lot or serial trace against approved BOM | Substituted drive unit or lidar of lower grade | Desk review of 5 serials |
| 2. Battery assembly | Pack capacity, IR, insulation and BMS trip test record | Inability to ship, or early capacity collapse | Witness or record review, half day |
| 3. Chassis assembly | Torque record, alignment measurement | Uneven tyre wear and navigation drift at month nine | Half day at the line |
| 4. End-of-line test | Full test sheet with measured E-stop time | Safety-field trip failing on site, not in the factory | Witness one unit, one hour |
| 5. Ageing | Burn-in duration and pass criteria | Infant-mortality failures in the first month of service | Record review |
| 6. Packing | State of charge, UN 38.3 pack attached, crate spec | Customs hold and transit damage | Photo evidence at loading |

Sampling Arithmetic: How Many Units to Inspect
Buyers usually ask whether they must inspect every unit. For fleet orders the answer is a sampled acceptance plan, and it should be written into the contract as a number rather than left to the inspector's judgement. The arithmetic that matters is the probability of accepting a batch that contains a defect. With a lot of 20 units and an acceptance number of zero, inspecting 5 units accepts a lot with a 10 percent defect rate about 77 percent of the time. Inspecting 10 units drops that to roughly 35 percent. Inspecting all 20 drops it to zero.
The practical settlement for a first fleet order is to inspect 100 percent of units at gates 2 and 4, and sample gates 1, 3 and 5 at five units per lot. Serial defects live at gate 2 and gate 4 because those are the two steps where a wrong component or a mis-set parameter is copied across every unit built that day. Assembly variance lives at gate 3 and gate 5, where a sample tells you whether the process is in control.

Concentration Risk: The Question Nobody Asks
The most expensive supplier failure is not a quality defect. It is a single-source component whose only factory stops producing. Ask three questions and record the answers in the contract annexe: which component in the bill of materials has only one approved supplier, where is that supplier's factory, and what is the qualified alternative.
In commercial service robots the usual single-source items are the lidar module, the drive unit and the battery cells. A mid-size Chinese robot maker typically has one qualified lidar supplier, one or two drive-unit suppliers, and a cell supplier relationship that changes with price. If the answer to the alternative question is "we would requalify", the requalification time is your downtime. Ask for the last time they actually requalified an alternative and how long it took. An answer in weeks is a managed risk. An answer with no date is an unmanaged one.
Depot-level spares are the second half of the same question. A spare-parts annexe that lists modules with prices but no stock commitment is a price list, not a supply plan. The commitment worth having is a stated holding quantity for the three highest-failure modules, held at the factory or at a European depot, with a replenishment lead time in days.
Three Clauses That Decide Whether a Defect Is Covered
Warranty disputes on imported robot fleets rarely turn on whether a part failed. They turn on whether the supplier can attribute the failure to something the buyer did. Three clauses determine that, and all three are negotiable before the first order.
The consumables and wear-parts schedule
Every warranty excludes wear parts. The problem is that the definition is often written after the failure. Insist on a schedule listing each wear part, its expected service life in hours or cycles, and its price, annexed to the contract. A tyre excluded as a wear part is reasonable. A drive unit excluded because the supplier classifies it as a wear part is a warranty that does not cover the drivetrain.
The operating-envelope clause
Warranties are voided by operation outside a stated envelope: ambient temperature range, floor gradient, aisle gradient, water exposure, charge-cycle count per day, payload. Ask for the envelope as a table and compare it to your building. Most disputes that end badly are not the supplier refusing a valid claim; they are the buyer having operated at 38 degrees in a plant room the datasheet rated to 35.
The attribution and inspection clause
When a unit fails in your building, who determines cause, and where does the unit go? A clause that requires return to the factory before any determination means a six-week round trip and a shipping bill, and it gives the supplier sole custody of the evidence. The clause worth negotiating is diagnostic access first: remote log download, a joint inspection at your site or at the European depot, and a written determination with a stated response time. If the unit must return, the return is at the supplier's cost when the failure is theirs, and the clause should say so explicitly rather than leaving cost to be argued.
| Clause | What the supplier will propose | What to negotiate instead | Why it matters at year three |
|---|---|---|---|
| Wear-parts definition | A general exclusion for "consumable and wear items" | An annexed schedule naming each part, its service life and its price | Prevents the drivetrain being reclassified as a wear item after a failure |
| Operating envelope | A datasheet range in marketing language | A contract table of temperature, gradient, water and cycle limits, referenced to your site conditions | Removes the most common ground for a declined claim |
| Attribution | Return to factory for determination at buyer's cost | Remote log access plus joint inspection at site or depot, written determination on a clock | Turns a six-week argument into a two-week repair |
| Spare parts pricing | A price list valid at signature | Priced annexe valid for the warranty term with an annual escalation cap | Stops year-three parts pricing from being set unilaterally |
| Software and firmware | Firmware updates at supplier's discretion | Access to security updates for the service life, with a stated end-of-support date | Prevents a fleet becoming unsupportable before it is worn out |
What a Remote Audit Can and Cannot Establish
A document audit plus six gates with photo and record evidence is a genuine substitute for a first visit in most cases. It establishes that the entity is real, the components are traceable, the test evidence exists, and the contract allocates risk. What it cannot establish is culture: whether the second shift runs the same process as the first, whether the line is clean, whether the people building your unit look like they have done it a thousand times.
That gap is best closed later rather than first. Place a pilot order sized to the sampling plan, run the gates, commission the units on your own site, and treat the first year of service data as the real audit. Then spend the flight on a supplier you already have evidence about, at the point where you are negotiating a second order and the visit has commercial leverage. The vendor evaluation framework sets out how to score the result of that first year, and the warranty claims and RMA guide covers what happens if the evidence turns out to be weaker than the paperwork suggested.


Putting the Audit Into the Purchase Order
An audit that is not in the purchase order is a favour. The mechanical way to make it binding is a quality annexe with four parts: the document set as a numbered list with submission deadlines, the six gates as inspection rights with a notice period, the sampling plan as explicit numbers, and the three clauses above as negotiated text. Then a single sentence that gives the annexe teeth: goods that fail a gate are rejected at the supplier's cost, and the rejection right survives final payment.
That sentence is the difference between a supplier who sends you a tidy PDF pack and a supplier who knows a unit can be sent back. It costs nothing to ask for and it changes how the order is built. For buyers comparing several candidates, run the same annexe against each one and score the responses side by side; the tender evaluation scoring guide covers the weighting, and the battery certification and transport guide covers the shipping documents in more detail. AOMAN builds to this annexe structure by default, and the full product range ships with the battery pack test record and the Declaration of Conformity per unit rather than per model family.
