Motionwell Automation builds battery module automation in Singapore for lines where the workpiece arrives energised and cannot be switched off. The delivered reference is an EV battery dismantling line, the work that took us into this industry in 2023: two ABB IRB 6700-200/2.60 six-axis robots at 200 kg payload and 2.6 m reach performing fastener removal, busbar disconnection and module extraction inside fenced stations, every safety function engineered to ISO 13849 Performance Level d, insulated end effectors working on a pack live at 400 V and above, and modules of 30 to 80 kg lifted by vacuum-assisted mechanical grippers with load cells that confirm module weight before transfer. Machines are designed, assembled and tested at our Woodlands Link facility, and the company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3.
Where we stand, said plainly before you read further. We do not manufacture robot arms, cameras, drives or cells. We design everything between the arm and your pack: the tooling, the fixturing, the sequence, the safeguarding, the control system and the recovery logic. We do not do battery chemistry, so electrolyte processing, leaching and cathode recovery belong to somebody else, and we are not a notified body and do not issue CE certificates.
This page covers what a live workpiece does to the safety case, the sequence that has to be proven before anyone approaches, what a voltage rating on tooling actually commits you to, what module mass does to handling and to fixtures that must hold packs which are not identical, why undoing a joint is harder than making it, how vision copes with variation, what has to be recorded and at which level, and how a second-life end state changes the machine compared with a recycling one. The line itself is documented in the EV battery dismantling case study and this page does not repeat it. If you have a pack format and a throughput target, skip ahead and talk to an engineer.
Why Does a Live Pack Change Every Safety Assumption?
Because the ordinary way of making an automation cell safe is unavailable. Normally you isolate the energy and then work. A pack arrives with its energy stored inside it and there is no switch that takes it away, so every safeguard has to be designed for a cell that is energised the whole time somebody is near it.
That inverts which object in the room is dangerous. On almost every other cell we build, the arm is the hazard and the safeguarding is drawn around its working envelope. Here the arm is the least dangerous object in the cell. It follows that a force limit on the arm buys nothing: limiting contact energy limits what the robot can do to a person, and says nothing about what the workpiece can do to a person, or about what the robot can do to the workpiece, where an unexpected move could puncture a cell and start thermal runaway. Collaborative rules were deliberately not used to justify an open cell on this line, and the general form of that decision is worked through in our comparison of collaborative and industrial robot cells.
| Hazard in the cell | Why limiting the robot does not address it | What actually controls it |
|---|---|---|
| Stored electrical energy at 400 V and above | Present whether or not anything moves, and present with the power off | Insulated tooling, verified isolation before contact, locked access until voltage is confirmed low |
| Thermal runaway from a punctured or crushed cell | A slower arm still punctures a cell if it goes to the wrong place | Safety-rated axis range and standstill, condition screening before handling, gas detection and extraction |
| Electrolyte vapour and off-gassing | Not a mechanical contact hazard at all | Local exhaust ventilation, continuous gas detection, negative pressure relative to the building |
| Module mass of 30 to 80 kg lifted at extension | Force limiting caps contact force, not gravity | Arm sized for payload at reach, mechanical guarding, load verified before transfer |
| Electrostatic discharge near exposed terminals | Unrelated to robot speed | Dissipative belt material and bonded grounding through the conveyor structure |
| A pack that arrives already damaged | The robot cannot tell by touching it | Identification and visual condition screening before a gripper is allowed to close |
One collaborative arm does appear on this line, at a vision station inside the larger automated sequence. That is the useful distinction to carry away: collaborative operation is decided per task and per workpiece, not per line. Looking at a pack and handling a live one are different jobs with different answers.
The standards route is the same as any machine. An ISO 12100 risk assessment sets what has to be reduced, the robot and the integrated system fall under ISO 10218-1:2025 and ISO 10218-2, and each safety function carries a required performance level calculated and validated under ISO 13849-1:2023. Two things about this line came out of that assessment rather than out of a catalogue: every safety function was engineered to Performance Level d with dual-channel monitoring, and the fenced robot stations were separated from the manual loading area at the head of the line, so the people who load incoming packs are never inside the disassembly space. A design still documented against the 2015 edition of ISO 13849-1 will need its performance level calculations restated when the machine is next assessed. The wider compliance scope is on our machine safety and CE marking page.
What Has to Be Proven Before a Person Can Approach the Pack?
A sequence of gates, each producing evidence, with the next step refusing to start until the previous one returns a green status. The order matters more than any individual device in it.
| Gate | What is verified | What stays locked until it passes |
|---|---|---|
| Incoming identification | Pack identity read by barcode, and a visual condition check for swelling or deformation | Entry to the first disassembly station |
| Electrical characterisation | Automated voltage and insulation resistance check on the arriving pack | Any mechanical operation on the casing |
| Pre-disconnection verification | Insulation resistance measured with a Fluke 1587 integrated into the cell controller, in front of every busbar disconnection | The disconnection itself |
| Module separation | Load confirmed against the expected module weight before transfer | Release of the module to the next station |
| Human access | Robot at safe standstill, and area voltage confirmed below 60 V DC by the isolation monitoring relay | Schmersal AZM300 solenoid-locking door interlocks on every access door |
The last row is the one that separates this cell from an ordinary robot cell, and the word doing the work is “and”. Mechanical standstill is not enough. A stationary robot beside an energised pack is a cell nobody should be standing in, so the interlock waits on two independent conditions and the door stays locked until both are satisfied.
The emergency stop is designed on the same logic. It implements Safe Torque Off as the primary stopping method, removing power from the robot servo drives within 10 ms, so the arm stops under mechanical friction and gravity rather than decelerating under motor control. A category 1 stop is kinder to the mechanics and leaves the arm where the program put it, and it was rejected here because it keeps the drives energised for the duration of the deceleration. Above a punctured cell risk, the design accepts a coasting arm over a powered one. The price is paid at recovery: after any stop the robot needs a reference run and an operator has to acknowledge the safety circuit before production resumes, so a nuisance trip costs real minutes and the fault logic has to be worth trusting.
Two of the safeguards on this line protect against the process rather than against the machine, and they run whether or not anything is moving. Local exhaust ventilation hoods at each disassembly station extract through HEPA filtration rated at 2,000 m3/h, Draeger Polytron 8000 sensors monitor continuously for hydrogen fluoride and volatile organic compounds, and the line is held at slight negative pressure relative to the surrounding facility so that anything released travels towards the extraction rather than into the building.
What Does “Rated” Actually Mean for Insulated Tooling?
It means a claim about a whole assembly, held to an interval, with somebody’s name against it. It does not mean a material with a number on its data sheet, and the difference is where insulated tooling projects go wrong.
On this line every robot end effector and gripping tool is rated for operation in high-voltage DC environments up to 1,000 V, built in three layers that fail differently. Nomex aramid paper wrapping is the primary insulation on metallic gripper fingers and tool mounting plates, chosen because it holds its dielectric strength when it has been contaminated with electrolyte rather than only when it is clean. GPO-3 fibreglass laminate mounting brackets sit between the robot flange and the tool body at 45 kV/mm, so the insulation is also part of the load path rather than a sleeve added to it. Delrin liners cover the fixture surfaces that touch module terminals, which addresses a different failure: not a person completing a circuit, but the fixture itself bridging positive to negative across a module it is only supposed to be holding.
The part that makes the rating mean anything is the verification interval. Insulation resistance is measured weekly with a 1,000 V DC megohmmeter and any tool reading below 100 megohm comes out of service. Insulation on a working cell degrades by contamination and abrasion rather than by failing outright, so a tool that passed at commissioning tells you nothing about the same tool in month nine, and a tool that has been re-tipped or repaired has to be re-tested rather than assumed good.
The consequence for whoever owns the machine is worth stating plainly, because it is usually missed at handover. The gripper is a safety function. It belongs in the safety function register alongside the interlocks and the scanners, not in the spares list, and the technician who changes a finger has modified a safety function and needs a procedure that says so. How tooling like this gets designed and built is on our end of arm tooling page.
What Does Module Weight Change About the Handling?
It sets the arm class before any other requirement gets a vote, and it does so through reach rather than through mass. Modules run 30 to 80 kg and have to be lifted at extension inside a pack, so the useful number is payload at reach with the tooling attached. Rated payload is measured at the tool flange, which means the insulation, the nutrunner and the gripper structure all come out of the same budget the module needs. That is how a 30 to 80 kg module ends up on a 200 kg arm.
The gripper naming on this line is the design in miniature: vacuum-assisted mechanical grippers, in that order. The mechanical element carries the load and the vacuum assists, which is the right way round on a coated, vented, sometimes contaminated casing where a vacuum cup has no guarantee of a clean sealing surface.
The load cells are the more interesting part, and they are not a lifting check. They confirm module weight against expected values before transfer, which makes weight a state check on an assembly nobody can see inside. A module that reads heavy is usually still fastened to something. A module that reads light is not the module the recipe expected. Either way the cell has learned something about the pack before the module is in the air, and that is the cheapest diagnostic available on a workpiece that cannot be opened for inspection.
Transport carries its own constraints. The line runs a Bosch Rexroth TS 5 heavy-duty belt conveyor rated for 500 kg per pallet position, with Interroll zero-pressure accumulation zones for transfer between stations and a three-position accumulation buffer between disassembly and sorting so the two ends do not have to share a cycle time. The belt surfaces are carbon-fibre-loaded PVC with surface resistivity in the dissipative range under IEC 61340-5-1, and grounding straps bond the conveyor frame to the facility ground bus every 3 m, so charge bleeds away continuously instead of accumulating and discharging near an exposed terminal. Where a machine of this kind incorporates lifting equipment, it has to be examined and certified by an approved authorised examiner before use in Singapore, which we handle as part of delivery.
How Do You Fixture a Pack That Is Not Perfectly Consistent?
By accepting up front that the incoming mix is not under anyone’s control. A recycler takes what arrives, so the fixture cannot be designed around one pack the way an assembly fixture is designed around one product.
The delivered answer is a reconfigurable base plate rather than a dedicated nest: a grid of M8 threaded inserts on 50 mm centres, with Destaco pneumatic swing clamps of adjustable stroke and clamping force repositioned against that grid, and 3D-printed Nylon 12 locating nests carrying the module-specific features. The plates are anodised aluminium 6061-T6, chosen for two reasons that both matter here: the strength-to-weight ratio lets a technician reposition a plate by hand, and the material is non-magnetic, which keeps the fixture from interfering with battery management system electronics during disassembly. Changing between prismatic and pouch module formats means repositioning the clamps and updating the clamp-position recipe in the PLC, about 30 minutes with two technicians, with no welding, drilling or permanent modification.
Flexibility is not free and it is worth being explicit about the bill. Dedicated nests for a single format would clamp faster, locate better and need no recipe management at all. What the grid buys is the ability to accept a format that was not on the drawing when the line was built, and what it costs is the changeover time above and a clamp-position recipe that has to be right, because a mislocated module is a mechanical interference on a live pack rather than a scrapped part. The same trade appears on packaging equipment in a lower-consequence form, set out on our filling line changeover page.
Why Is Taking a Pack Apart Harder Than Putting One Together?
Because assembly knows the state of every part before it touches it, and disassembly does not. This is the single assumption that has to be surrendered when a line runs in reverse, and most of the engineering difficulty follows from it.
| What an assembly line assumes | What a disassembly line finds |
|---|---|
| Parts arrive to a drawing and to a tolerance | The pack arrives as it comes back from service, with a history nobody recorded |
| Fasteners are new, clean and correctly torqued | Fasteners may be corroded, painted over, preloaded, or already rounded |
| The joint was designed to be made | The joint was designed to survive vibration and crash loads for a decade, not to be undone |
| The sequence is fixed and repeats | The sequence has to branch on what is found, including the branch where a fastener will not move |
| A failed operation produces a scrapped part | A failed operation is happening on an energised assembly |
| Adhesive and sealant are process steps | Bonded joints have no reverse operation at all |
The instrumentation answers the first four rows. Atlas Copco QST torque-controlled nutrunners mount on the robot flange, with torque and angle logged per fastener against the pack barcode. On an assembly line that record proves a joint was made correctly. On this line the same signal is a detector: a fastener that has not broken loose within the expected angle window is a stop condition rather than a reject, because the alternative to stopping is a rounded head on a live pack and a manual recovery task nobody wanted.
The last two rows are why the delivered line stops where it does. Its scope ends at module separation, and cell-level teardown, electrolyte handling and cathode processing were deliberately left outside it. Automating those means treating breached cells as a normal operating condition rather than as a fault, which is a different machine with a different fire case behind it, and drawing that boundary is usually the most consequential decision made on a project of this shape. How that scoping conversation runs before a design exists is on our custom automation page.
How Does Vision Find a Fastener on a Pack That Varies?
By supplying an offset rather than a template. Teaching a fixed position works when the fixture holds the same part in the same place every cycle, which is exactly the condition a reconfigurable grid gives up. So the camera locates the features on the pack in front of it and the robot corrects to them, instead of the program assuming where they ought to be.
Vision does two separate jobs on this line and they fail in different ways. The first is identity and condition at the head of the line, where the barcode read and a visual condition check share the same camera and lighting approach: a swollen or deformed casing has to be caught before a gripper touches it, not after, which makes that station a safety interlock wearing a camera. The second is feature location during disassembly, where fastener heads, connector bodies and busbar terminations have to be found on a casing whose position and revision both vary.
What decides whether the second job works is presentation and lighting rather than sensor resolution, since a defect or a feature the light does not reveal is invisible at any resolution. Battery casings make that harder than average: dark anodised and painted surfaces sit next to specular busbars and fastener heads in the same field of view, and fasteners frequently sit recessed below a surface that shadows them. Those are lighting geometry problems, solved at the station rather than in software, and the approach we take to them is set out on our machine vision inspection page.
State the limit clearly, because on a battery line the temptation to over-read a vision result is real. A camera cannot see state of charge, an internal short, or the condition of a cell through its casing. The electrical checks do that work. Vision confirms geometry and identity, and a vision pass must never be wired into the sequence as though it were an electrical clearance.
What Has to Be Recorded, and at Which Level?
At pack, module and cell level, with the awkward requirement that the record has to survive the pack ceasing to exist.
Traceability on an assembly or packaging line builds parents from children: units into cases, cases onto pallets, each link written as the parent is formed. Disassembly runs that chain backwards and hits a problem the forward direction never has. The parent identity is the only identity that arrives at the door. Every child leaves the line as an independent item, and if a module leaves without a record binding it to the pack it came out of, its history is gone. For material recovery that may not matter much. For anything sold on, that history is most of what the buyer is paying for.
The delivered line binds records to the pack barcode, including torque and angle for every fastener, and sorts its output into four recycling streams: cathode material, anode material, copper busbars and aluminium casing. The principle transfers directly from the traceability work we do on other lines, where a measurement is worthless unless it is bound to an identity at the moment it was taken, and where retrofitting that link later is painful enough to be worth designing in from the start. The read side of it is on our code reading and traceability page.
The battery-specific difficulty is that a module’s own identity is often not machine-readable, or not there at all. It may be a label under a bracket, a marking never intended for a camera, or nothing. Where no readable module identity exists, the line has to create one at the point of separation, and that is a specification decision rather than a commissioning decision, because adding a marking station afterwards is a rebuild of the sorting end rather than an addition to it.
How Do Second Life and Recycling Change the Machine?
They are different end states, and the difference lands on the machine rather than on the process sheet. A line built for one is not a line for the other with a label changed.
| Requirement | Material recovery | Second life |
|---|---|---|
| What leaves the line | Separated material streams | An intact, working, graded module |
| Damage on the way out | A nuisance, since the module is going to be broken down anyway | A loss of the whole value of the item |
| Electrical test depth | Enough to make handling safe | Enough to grade capacity, resistance and state of health per module |
| Identity and record | Bound to the pack for process and compliance | Bound to the module, and durable enough for a downstream buyer to accept |
| Fixture and gripper priority | Speed and separation quality | Non-marking, non-deforming handling of a saleable item |
| Sorting logic | By material | By grade, and by whatever the receiving application will accept |
| What limits throughput | Disassembly and separation | Test time per module, which no faster arm shortens |
Our delivered work sits at the material recovery end, and the honest statement is that we have not delivered a second-life grading and test line. What transfers between the two is the whole front half: identification, electrical verification, safe opening, fastener removal and module extraction are the same problem up to the moment the module is off the pack. What does not transfer is the back half, and the back half of a second-life line is largely test rather than disassembly, which is automated test equipment work with a different specification, a different cycle time and a different data model behind it. Anyone scoping a second-life line as a dismantling project with testing added at the end has the proportions the wrong way round.
When Is a Robot Cell the Wrong Answer for Battery Work?
Four situations where we would tell you so rather than quote.
The volume does not exist yet. Where a handful of packs a week arrive across many formats, a trained team with insulated hand tools and a proper permit-to-work system will beat a machine on cost and on flexibility, and the reconfigurable fixturing that makes a machine flexible is exactly the part that costs money and changeover time. Automation earns its place when a format repeats often enough for a recipe to be worth having.
The formats change faster than a recipe can follow. A clamp grid absorbs variation within a family. It does not absorb a genuinely new pack architecture every month, and pretending otherwise produces a machine that is in changeover more often than it is in production.
The joining process is the project. Pack assembly rather than disassembly usually means welded busbar connections, and we do not build production welding cells. If your line is defined by the weld, the process belongs with a supplier who owns it.
You want the cell to run unattended. Lights-out is not a realistic target where a fault means somebody eventually approaching an energised assembly. Design for recovery instead: what the cell does when a fastener will not move, who opens the door, and what has to be proven before they do. A line that stops safely but needs an engineer to clear it will be worked around by the third shift within a month.
One boundary beyond the machine is worth naming early. Our scope ends at the cell and its extraction. How incoming packs are stored, at what state of charge, and under what fire strategy is a building question answered with your own fire safety consultant and the authority having jurisdiction, and it usually needs to be settled before the layout is fixed rather than after.
Which Parts of This Would You Actually Need?
Most enquiries turn out to need some of the line rather than all of it. The table below is how we sort them at the first conversation.
| Your situation | Start from | Why |
|---|---|---|
| Packs arrive in a mix nobody controls | Reconfigurable clamp grid and a clamp-position recipe | The fixture cannot be designed around one pack, and the changeover is the price of accepting that |
| One pack format at volume | Dedicated locating nests | Faster clamping, better location and no recipe to manage |
| Output goes to material recovery | Separation quality and stream sorting | Damage in transit costs little, so the design optimises for safe throughput |
| Output goes to second life | Per-module characterisation and a durable module identity | Test time sets the rate, and the record is part of the product |
| Operators open packs by hand today | The safety case, before any robot is selected | The hazard inventory decides the guarding, the tooling and the sequence |
| You already run an assembly line | A branching sequence, not a reversed program | Disassembly cannot assume the state of what it receives |
| A few packs a week across many formats | Trained people with insulated hand tools | The flexibility a machine needs here is the expensive part of it |
| Busbar joining by welding | A welding process supplier | Not our scope |
Lead time runs 16 to 24 weeks from concept approval to factory acceptance testing on a standard build, and design, fabrication, assembly and testing happen at Woodlands Link, which is why a Singapore buyer attends the factory acceptance test instead of flying to it. On a line of this kind that matters more than usual, because the details that decide whether it runs are found on your own packs rather than on a drawing.
Which standard editions apply right now?
The editions below are the ones we design and document against on current projects. We check them on the date shown rather than assuming last year's edition still holds.
| Standard | Current edition | What it means for your machine |
|---|---|---|
| ISO 10218-1 — Robotics, safety requirements, Part 1: industrial robots | ISO 10218-1:2025 | Published February 2025, the third edition and the first substantive revision since 2011. It adds robot classifications with matching functional safety requirements, safety-related cybersecurity requirements, and end-effector guidance. Most of ISO/TS 15066:2016 on collaborative operation moved into Part 2. |
| ISO 13849-1 — Safety of machinery, safety-related parts of control systems | ISO 13849-1:2023 | The 2023 edition is the version referenced by ISO 10218-1:2025 for robot control system safety functions. Designs still documented against the 2015 edition will need their PL calculations restated when the machine is re-assessed. |
Frequently Asked Questions
What do you need from us if the pack manufacturer will not release drawings?
A physical pack and permission to take one apart, which is what the design actually gets built from. Drawings help and we ask for them, but a dismantling line is designed against fastener positions, joint types, harness routing and the way a casing actually comes off, and none of that is reliably readable from a model of the assembled state even where one exists. What we need is at least one representative pack per format, its nominal voltage and its arrival state of charge, and a written account of how your people handle it today. A proving trial on real packs turns an opinion about feasibility into a number, and we would rather run it before quoting than after.
Can an assembly line be reversed to run disassembly?
Rarely as it stands, and the reason is structural rather than commercial. An assembly line moves from known parts to a known product, so its fixtures locate features guaranteed to be there and its sequence never branches. A disassembly line receives one item whose internal state it cannot assume, and has to branch on what it finds, including the case where a fastener will not come out. The station order can sometimes be reused, and the frames, conveyors and control hardware often can. The fixturing, the tooling and most of the control logic get redesigned. Control system modernisation of existing production machines is our largest line of work this year, so partial reuse is worth assessing rather than assuming.
How many packs per hour should a robotic dismantling cell run?
Ask for a different number, because parts per hour measures the wrong thing here. Cycle time was never the selection criterion on the delivered cell and could not have been: a voltage and insulation resistance verification step runs in front of every disconnection and takes as long as it takes, so a faster arm does not compress it. The robots were sized for payload at reach rather than for speed, because modules of 30 to 80 kg have to be lifted at extension. The figures worth putting in a specification are packs per shift against a stated format mix, the changeover time you can live with, and how much of the sequence has to complete with nobody inside the cell.