An exposed fastener on an EV battery case looks like a small thing — until one is missing.
A single battery case carries hundreds of exposed fasteners: rivet nuts, weld nuts, lock bolts. They are all installed by hand. Miss one, or leave one not fully tightened, and it is one more rework on the line at best — and an installation failure at the customer at worst. This is exactly the problem a major NEV manufacturer's battery case project (the E311 project) faced, and the problem this article reconstructs.
What follows is a delivery record of an exposed-fastener 3D vision inspection machine that Zhongbai Robot supplied the motion system for. The end customer name is withheld as is standard practice; all figures come from the project's own documentation.
1. The problem is not "detection is hard" — it is that manual inspection cannot keep up
The E311 battery case carries 251 exposed fasteners (E311-500 variant): 95 M5 rivet nuts on the front cover, 9 in the electrical compartment, 12 blind rivets, 26 M8 weld nuts, 10 rivet nuts on the vehicle skid plate, 73 M6 and 26 M8 lock bolts on the bottom plate, and more.
Since start of production, there had been 35 quality incidents of missing or under-tightened fasteners, 5 of which escaped to end customers and caused installation complaints. To contain this, three manual inspection stages had already been added — 7 production workers, 4 GP12 inspectors, and 4 customer incoming inspectors — each spending 2–3 minutes per part flipping and marking.
The core issue: 251 fasteners, all installed by hand. At that volume, a missed fastener in manual inspection is a probability problem, not an attitude problem. Three inspection stages are manpower, not a solution.
2. What makes the job hard: large size, two sides, millimetre-level deviation
The difficulty is not one dimension — it is three stacked together.
- Large part: E311-500 / E311-400 cases measure about 1472 × 1689 mm, and the compatible 541 variant about 1990 × 1500 mm;
- Many features on two sides: the front carries M5 rivet nuts, M8 weld nuts, sleeves and blind rivets; the back carries M6 bolts, M6/M8 rivet nuts and the central lug lock nuts — the part must be flipped and inspected on both sides;
- Small deviation: two sleeves on the crossbeam can be out of position by millimetre-level amounts, which must be caught.
Large part means the motion stage needs a 2.5-metre-class stroke. Millimetre deviation means the stage's repeatability must be tight, or the camera's measurement reference itself drifts.
3. Architecture: two independent gantries, front and back inspected separately
The solution uses two independent inspection machines, one per side, in the flow: load → locate → scan code → front inspection → flip → locate → back inspection → unload. The flip and conveyor reuse the existing line; Zhongbai delivered the inspection sections (load, locate, scan, inspect), with electrical and partial rework coordinated between both parties.
Each machine is built around a gantry robot carrying a 3D line-scan camera, checking each fastener for presence, seating, hole presence and rivet presence. The frame is column-plus-gantry, the electrical and vision cabinets are sheet metal, enclosed by aluminium-profile mesh guarding with front and rear safety light curtains.
4. The motion scheme: rack-and-pinion X/Y, ball-screw Z, rotary R
This is where non-standard integration ability shows most. Large part plus millimetre-level detection demands "long stroke, moves well, stops accurately" — all three at once.
- X/Y on rack-and-pinion modules: 2500 mm travel in X/Y, built from HGR25 linear rails, M2 rack and a gearbox. Rack-and-pinion holds stiffness and repeatability over long stroke without the accumulated error a long timing belt would carry;
- Z on a ball-screw module: 600 mm Z travel, moving the camera up and down for field-of-view compatibility, with the screw's self-locking helping the axis hold position;
- R rotary axis: changes the camera angle, working with Z to reach features at different positions;
- Repeatability ±0.05 mm, end-of-arm payload ≤10 kg, with limit switches, drag chains and feet included.
The two machines run 8 axes total (X/Y/Z/R × 2), all under one control system.
5. Vision scheme: 3D line scan + code reading + in-house algorithm
The inspection uses Hikvision 3D laser line-scan cameras (with lens, LED structured light and controller); a code-reading camera is fitted on the front-inspection machine only. The 3D scan runs at about 100 mm/s, reduced as needed by motion stability to hold the overall cycle.
The software and algorithms are developed in-house, with detection models for the E311-500, E311-400 and 541 variants. Inspection data is stored automatically for at least one month with periodic backup, and scan results can be handed off to MES.
6. Control scheme: Inovance PLC + 8 servo axes + EtherCAT
Control uses an Inovance H5U PLC with SV630N 400W servos (8 axes across X/Y/Z/R) and ISMH motors, a 10-inch HMI, over EtherCAT. High and low voltage run in separate trunking, with the main electrical cabinet and the vision cabinet kept independent.
7. Key parameters and acceptance criteria
| Item | Value / requirement |
|---|---|
| Gantry travel | X/Y 2500 mm, Z 600 mm |
| Repeatability | ±0.05 mm |
| End-of-arm payload | ≤10 kg |
| 3D scan speed | approx. 100 mm/s |
| Transmission | X/Y rack-and-pinion, Z ball screw, R rotary |
| Cycle | 15 JPH |
| Availability | ≥90% |
Acceptance follows the final signed technical agreement, with five core metrics: fastener presence recognition 100%; correct seating ≥99.97%; code reading 100%; cycle 15 JPH; availability ≥90%.
8. Why the scheme is built this way
Each selection maps to a specific engineering constraint:
- A gantry instead of a fixed camera matrix: a 1.5-metre-square part with features spread across two sides makes a fixed camera matrix expensive and hard to maintain; one 2.5-metre gantry covers an entire face, and a new product variant means a new algorithm, not a new mechanism;
- Rack-and-pinion instead of a belt on X/Y: over 2500 mm, rack-and-pinion gives more reliable stiffness, position retention and long-distance consistency, avoiding the tension decay and accumulated error of a long belt;
- Ball screw on Z plus a rotary R: Z must stop accurately and hold, which suits a screw's self-locking; R handles camera angle and reduces dependence on part orientation;
- 3D line scan instead of 2D area scan: fasteners are a "present or not" and "seated or not" judgement, and millimetre-level height deviation (a sleeve sitting proud of the crossbeam) is only reliably caught in 3D.
9. Zhongbai's role in this machine
Zhongbai Robot delivered the machine's mechanical body and motion system: the gantry, the rack-and-pinion X/Y, the ball-screw Z module, the rotary R axis, and the matching limit switches, drag chains and feet. Vision, electrical and control were integrated by the project as a whole. The machine ships in 8–10 weeks with a 12-month warranty from SAT.
This is Zhongbai's consistent positioning: not the whole-line general contractor, but the partner that makes "long-stroke, high-accuracy, non-standard motion" solid, so the integrator and end customer can land the whole machine.
If you are building a large-part inspection, pick-and-place or transfer machine and need a long-stroke, high-accuracy non-standard motion stage, contact us with your stroke, payload, accuracy and cycle time. See the rack-and-pinion module range and fully enclosed ball-screw modules.
For full specifications and selection support, download the Zhongbai product catalogue (PDF).