Mechanical Design and Simulation

Motionwell Automation provides mechanical design services in Singapore as the front end of machines we then build, with an in-house design team of eight working in SolidWorks from a design centre at 20 Woodlands Link, in the same industrial estate as the assembly floor and the machining shop that cuts the parts. Machine design simulation means two different activities at two different points here: offline robot simulation that verifies reach, clearance and cycle time before anything is fabricated, which is how up to 50 layer patterns on our delivered carton palletizing cell were programmed and their pallet stability and cycle time checked in ABB RobotStudio; and the force and moving-mass analysis that sizes a frame and its drives before steel is ordered. The company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3. The honest headline: almost everything that decides whether a machine holds tolerance, gets serviced and changes over is settled on a screen in the opening weeks, while it is still free to change, and the parts that get analysed hardest are frequently not the parts that decide the outcome.

Where we stand, said plainly before you read further. Design here is drawn by people who have to make it work at factory acceptance testing in the same building, so it is biased towards what our own shop can machine, weld and inspect. That is an advantage on manufacturability and a constraint if your machine needs an exotic process, and a section below sets out the cases where buying design from us is the wrong move.

This page covers what the design stage fixes permanently, why deflection rather than strength sizes a frame, how moving mass sets the drive, where analysis earns its cost and where it does not, what offline simulation proves, designing for the shop and for the person who maintains the machine, tolerance stack-up against the part, and the review gates. The structural arithmetic for long spans is on our gantry and truss robot page, the heavy end of the moving-mass problem is on the heavy duty robot track page, and the full project sequence around this work is in our guide to special purpose machine design. If you have a part drawing and a cycle time, skip ahead and talk to an engineer.

What Gets Decided at Design Stage and Cannot Be Changed Later?

Four things, and none of them is a component choice: frame stiffness, access, the envelope the machine occupies, and the changeover concept.

Two of those are worth a sentence more. Changeover is easily assumed to be a decision that can wait, and it is not: a recipe change is a setting the machine moves under software control, while a change part is geometry that has to be made, stored, identified and fitted, and which of the two you get is decided by whether the axis was motorised on the drawing. The reasoning and the delivered numbers are on our filling line changeover page.

The envelope decision includes kinematics, which is the least reversible choice on the page. Our 5-axis CNC shot peening machine for turbine blades runs X, Y, Z, B and C simultaneously because holding 100 to 200 per cent coverage on a curved blade means presenting the nozzle to the surface at a controlled angle rather than merely reaching the surface. Three axes would have reached every point on that blade and produced the wrong result at most of them. Axis count is a process decision made before any structure exists, and a machine cannot be given another degree of freedom later without becoming a different machine.

Why Does Deflection Rather Than Strength Usually Size a Machine Frame?

Because a frame that is strong enough can still be far too flexible to hold tolerance, and only one of those two properties gets checked by intuition. Strength answers whether the structure breaks. Stiffness answers how far it moves before it stops. A positioning machine rarely goes anywhere near its yield stress, and it fails its acceptance test at a fraction of a millimetre.

The trap is the response that follows. Faced with a droop figure, the reflex is a stronger material, and it does almost nothing, because elastic modulus barely varies across carbon and alloy steels while strength varies a great deal. What buys stiffness is geometry: section depth, the span between supports, and how directly the load path gets back to the floor. Span in particular punishes optimism, and the arithmetic behind that is set out on the gantry and truss robot page rather than repeated here.

Deflection has a second cost that is easier to miss because it appears in the cycle time rather than in the accuracy figure. A structure that bends also rings, so a compliant frame does not simply place the tool in the wrong spot, it takes longer to stop moving before the gripper can open. That penalty is paid on every cycle for the life of the machine. It gets worse with unsupported length, which is why a Z stroke specified generously in case it is useful later is not free, and why overhang between the tool and the guide rails matters as much as the stroke does. The vertical-axis version of that argument is on our servo lift and precision positioning page.

One figure is worth writing into a specification instead of a general stiffness claim: placement measured at the worst position in the envelope with the tool loaded. On a beam that position is near mid-span rather than at a column. It is a number a buyer can verify with an indicator at factory acceptance testing, which no amount of structural description is.

Why Does Moving Mass Have to Be Analysed Before Steel Is Committed?

Because moving mass and inertia size the drive, and the drive sizes almost everything downstream of it. Holding a load and moving one are different calculations: a static hold needs force, while a move needs torque to accelerate inertia, and it is inertia rather than weight that the motor fights. Motor sizing is a torque calculation from inertia, friction, acceleration and duty. Get the mass wrong at concept and every number after it is wrong too.

The order of consequences is what makes this urgent rather than merely important. Mass sets the force, the force sets the drive and gearing, those set the guide size, the guide and the reaction loads set the frame section, and the frame section sets the footprint and the guard line. Nothing in that sequence can be resized on its own. The heavy-mass version of that chain, where the carriage is mostly moving itself rather than the payload, is worked through on our heavy duty robot track page.

There is a failure mode specific to skipping this. An axis sized only against its move time meets that time on paper and then cannot stop cleanly, and the usual response is a larger motor, which reaches the same ceiling sooner rather than moving it, because the limit on any long axis is generally what it costs to stop the moving mass without exciting the structure rather than the torque available at the drive. Nothing in software fixes an axis that was sized against the wrong mass.

Two inputs are routinely missing from the mass figure when it arrives. The tooling is one: rated payload is measured at the tool flange, so whatever the gripper weighs comes out of the budget left for the part, and a load held away from the flange limits the arm by moment rather than by mass. The other is the heaviest variant nobody mentioned, because part weights drift and someone adds a heavier version next year. The tooling side of that arithmetic is on our end of arm tooling page.

Where Does Finite Element Analysis Earn Its Place, and Where Is It Theatre?

On the load path that decides accuracy, or that carries a load over somebody. Everywhere else it produces a coloured picture that decides nothing, and both the analysis and the meeting about it cost money.

How to read a result matters more than whether one exists. A stress plot proves nothing on its own, so ask three questions of any analysis in a quotation. Which position and load case was modelled, since a machine is not worst at the same place it is convenient to model. What was assumed rigid, because an assumption that the base is infinitely stiff moves the answer more than the mesh does. And what single measurable number came out of it, since an analysis whose prediction cannot be checked with an indicator on the finished machine has not been validated by anyone, including the person who ran it.

State the boundary honestly. Analysis is the wrong instrument for a good deal of what actually goes wrong on a machine, because friction, part-to-part variation, feeder behaviour, chip and coolant contamination and thermal drift across a shift are not in the model. Those get proved on your real parts, at commissioning and at factory acceptance testing, which is a different kind of evidence and usually the more valuable one.

What Does Offline Simulation Prove Before the Cell Exists?

Reach, clearance and cycle structure, and it earns its place because a pose that cannot be reached is a mechanical problem rather than a software one. Discovering it in a model costs a redraw; discovering it after the pedestal is grouted costs a layout.

Three things belong inside the simulation that frequently sit outside it. The dress pack is the first: a wrist axis rotating through a large angle winds its cabling, and the travel limit that follows gets discovered during programming unless somebody models it. The extremes of travel are the second, because on a track the pose that fails is usually at one end of the stroke where nobody checked. The third is the worst layer or the worst position rather than a representative one, which is why pallet stability and cycle time on our delivered palletizing cell were verified in ABB RobotStudio across the stored patterns before deployment rather than on the floor.

What it does not prove is worth being equally clear about. A simulation inherits whatever assumptions you fed it, so a cycle time from a model carries the same optimism as a datasheet figure measured gate to gate with a dummy load: the vision acquisition, the settling before a vacuum releases, the gripper actuation and the PLC handshake are what usually decide whether a station makes rate, and those are timed on real parts. On a robot riding a linear axis, the simulation typically covers the arm properly while the transit between stations remains a spreadsheet estimate. And no kinematic model has an opinion on whether your part sits in the nest the same way every cycle, which is a common way a cell fails in production. How that gets designed against is on our robot integration services page.

What Changes When a Machine Is Drawn for the Shop That Will Build It?

Design for manufacture stops being a slogan when the designer has stood at the machine. Our mechanical designers have direct experience operating CNC machines, which removes the designs that look correct on screen and are impractical to cut, and the design centre, the manufacturing shop and the assembly floor are separate units in one industrial estate rather than separate companies in different time zones. The rules that follow are unglamorous: machined components use standard tooling, sheet metal follows standard bend radii, and weldments are laid out so they can be inspected.

The expensive habit is a tolerance the function does not need. Each step tighter moves a part into a slower and dearer process, and the step is a cliff rather than a slope.

Two consequences follow that a buyer can check in a drawing pack. Precision features should be concentrated on a small number of machined parts and the rest of the structure allowed to be ordinary, because that is what lets a sheet metal guard cost what a guard should cost. And every dimension that carries a tolerance should be inspectable with equipment somebody actually owns, which is why weldment access for inspection is a design decision rather than a shop problem. Where components are machined to our drawings by qualified machining partners, that inspectability is the difference between a part being accepted and a part being argued about.

Why Is Design for Maintenance the Item Most Quietly Traded Away?

Because it is the requirement with no acceptance test behind it. Cycle time, accuracy and repeatability are all measured at factory acceptance testing in front of the buyer. Nobody measures how long it takes to change a bearing, so under schedule pressure the service clearance is the space that gets used for something else.

It is preventable at concept, and cheaply, because 3D concept modelling in SolidWorks validates spatial constraints, operator access and maintenance clearances before procurement begins. What that review needs is four questions asked out loud against the model.

1. Where does a person stand to service each item, and is that place inside the guard? If it is, every service visit is a lockout, and that changes how often the service actually happens.
2. What has to come off first? A wear item behind a cable tray behind a guard panel is a two-hour job that looks like a ten-minute one.
3. Can the machine be driven to a service position, and is that position in the program? On single-beam gantry builds this is settled at design stage by deciding whether the Z head can be driven to one end of the row; double-beam builds carry a maintenance walkway and railings for the same reason.
4. After a fault, how does the part get out? Somebody has to extract a workpiece from a closed station with a gripper still holding it, and if that recovery was never designed the operator invents one, usually by defeating something.

The trades here are real and should be made with open eyes rather than by accident. Mounting a collaborative arm overhead frees the entire floor for conveyors and operator access and costs easy maintenance access; that is a trade, not a free upgrade. A part that jams, such as a cap track, needs to be cleared and changed without tools, because it will jam and an operator will be standing there when it does. And the maintenance schedule itself is a design output: levelling and foundation checks belong in it on any long axis, alongside the mechanical drawings, electrical schematics, PLC program documentation, operator manuals and spare parts lists that ship with the machine.

Why Must Tolerance Stack-Up Be Run Against the Part Rather Than the Machine?

Because the machine’s repeatability is one term in a sum, and rarely the largest one. Our linear gantry series holds ±0.1 mm on the compact class, ±0.2 mm on the steel single beam and ±0.3 mm on the double beam; our low-profile seventh-axis track repeat-positions to ±0.05 mm; the delivered SCARA panel assembly cell works at ±0.01 mm repeat positioning. Each of those is a machine returning to a taught point. None of them is a statement about where your part ends up.

The transfer row is where the arithmetic gets brutal and where it is easiest to skip. A mobile robot parks to within roughly plus or minus 50 mm while the arm on it has to reach into a rack slot to a fraction of a millimetre, and the three mechanisms that close a gap of that size are a physical docking feature, hand-eye vision that measures the actual offset before the arm commits, and a compliant gripper for the residue. Leave the vision correction out and the cell passes commissioning and starts dropping parts in month three, when the drift row goes to work.

So the stack has to be run from the feature on your part that must be located, backwards through the gripper, the nest, the axis and the frame, at the position and temperature where the machine is worst. That needs the drawing of your worst part rather than the nominal one, which is why a design review that only ever sees a nominal model has not tested the design at all.

What Do the Design Review Gates Actually Check?

Each gate is a decision point rather than a handover, and what matters about each is what stops being free once it closes.

The largest schedule risk in this sequence is not in the machine at all: it is late finalisation of your product design. If the part geometry moves after detail engineering has begun, the tooling changes and the fixture usually changes with it, so freeze the product before that gate rather than after it. The same logic applies to your own requirements document, and how to write one that survives this process is on our automation URS page.

Two compliance points belong at the first gate rather than the last. The risk assessment is a design input, because the safeguarding it demands changes the layout, the access points and the guard line, and each safety function then carries a required performance level under ISO 13849-1:2023 that has to be architected rather than purchased afterwards; the physical scope that follows is on our machine safety and CE marking page. And for a machine destined for the EU, Regulation (EU) 2023/1230 applies from 20 January 2027 with no transitional period, and it is the first EU machinery law to put software integrity, updates and connected functions alongside mechanical safety, which changes what the design pack has to contain for a networked machine. A machine being designed now can be documented against it voluntarily, and that is a decision for the concept gate because it is a documentation architecture rather than a final checkbox.

When Is a Separate Design Package the Wrong Thing to Buy?

Five cases, and we would rather name them than discover them in week three.

A proven standard machine covers the process. Where a catalogue machine handles your product at your rate for less than anything we would design, the useful answer is to say so, and it costs you a conversation rather than a commitment.

The product is not frozen. Detail engineering against a moving geometry buys drawings that get redrawn at your expense. If the part is still changing, the honest first purchase is a shorter concept study that brackets the options, not a full design.

What you actually need is a requirements document. If the rate, the variants, the utilities and the acceptance criteria have not been written down anywhere, that gap is cheaper to close first, and closing it makes every subsequent quotation comparable to every other one.

The problem is upstream of the machine. Where parts arrive in a heap, where the presentation is inconsistent or where the bottleneck is a process dwell, better mechanical design of the wrong station changes nothing measurable.

The analysis is wanted as a document rather than as a decision. If a report is needed to satisfy a third party rather than to choose between two designs, say so at the outset, because it is a different piece of work with a different deliverable and we would rather quote it honestly than dress up an engineering study as one.

Two exclusions while we are being direct. We do not build production welding cells. And we are not a notified body and do not issue CE certificates: we design and build to a specification and support your conformity work, including LVD and CE testing and Ministry of Manpower lifting certification where the machine includes lifting equipment.

Design, fabrication, assembly and testing all happen at Woodlands Link with the design team in the same estate as the shop, which is why a Singapore buyer attends the factory acceptance test rather than flying to it. That matters more on design work than it appears to, because the questions that decide a machine get answered by putting a real part into a real fixture in front of the engineer who drew it. The questions worth asking any builder about their design capability, including us, are set out in our guide to selecting a system integrator.

Frequently Asked Questions

Is a stiffer machine frame just a heavier one?

No. Mass and stiffness are different properties, and adding plate to a section that is the wrong shape buys very little of the second while costing a great deal of the first. Extra moving mass has to be accelerated and stopped on every cycle, so it comes back as drive torque, as heat in the motor and as settling time you wait for before the gripper can open. Grade does not rescue it either, because elastic modulus barely varies across carbon and alloy steels. What buys stiffness is section depth, a shorter unsupported span and another column, none of which weigh much.

Do you run finite element analysis on every machine you design?

No, and a supplier who says yes is describing a slide deck rather than an engineering method. It earns its cost where the structure decides accuracy or carries a load over somebody: the long span with the tool hanging at the far end, the tall column, the lifting structure. Brackets, trays, covers and plates bolted flat to a machined base are settled by a standard section and a hand calculation. Analysis hours are finite, and spending them on a bracket means not spending them on the one span whose droop you will be measured against at acceptance.

How long is the design phase before fabrication can start?

Concept design runs 2 to 4 weeks and detail engineering a further 4 to 6, inside an overall 16 to 24 weeks from concept approval to factory acceptance testing on a standard build, or 24 to 32 weeks where cleanroom compatibility or full validation documentation applies. Which end of that range you land on is decided at the first gate rather than during fabrication, so the preparation worth doing is a firm variant list, a stated sustained rate and an acceptance criterion you are willing to sign. Those shorten concept design rather than lengthen it.