Every used machine I have ever opened up had the same story: it worked in CAD, and it fought in the flesh. The bearing fits, the sprocket clears the guard by two millimeters on screen, and then the first prototype jams because nobody checked the interference between the sensor bracket and the cable tray. This article is about interference checking the way it actually happens in a working design office, not the way the tool demos show it, and about building a check routine that catches the real problems before the workshop does.
1. The Gap Between the Checkbox and the Real Clash
The interference command in any mainstream CAD package will happily report a list of overlapping bodies, and that list is the least useful output you will get. A raw list of N clashes tells you nothing about which ones matter, which ones are allowed, and which ones the program has no opinion about. The real skill is triage: turning the list into priorities that a busy workshop can act on without a degree in assembly theory.
On my machines the checklist looks like this. First, moving parts against stationary parts, the classic jam source. Second, fasteners and their tool access, because a bolt that fits on screen can still need a socket you cannot get into. Third, cables, hoses, and anything flexible, because the rigid model will not tell you where a wrapped harness actually settles. Fourth, the shrink-fit of inflated tolerances, parts that are fine at nominal and collide at tolerance limits.
Honestly, the majority of clashes that bite me are not geometry errors in the model; they are unknowns the model was never given. A cable route, a welded envelope distortion, a thermal growth band, none of those exist as solid bodies until someone builds them, and the interference list will never see them.
2. The Interference Matrix: Whose Clearance Is It Anyway
Before you run a single command, define who is allowed to touch what. In my assemblies I build a clearance matrix, a table where the rows and columns are the subsystem groups and the cells say whether contact is allowed, what the minimum clearance is, and who owns the responsibility for signing it off. It sounds like overhead, and it is, but it converts the vague question is this okay into a precise question this bracket is 1.5 mm off the pump housing, and the matrix says the minimum is 3 mm, who changes what.
| Group A | Group B | Minimum Clearance | Allowed Contact |
|---|---|---|---|
| Rotating shaft | Stationary housing | 2 mm | No |
| Moving arm | Frame | 5 mm | No |
| Fastener head | Tool access zone | Socket clearance | Yes, with tool |
| Cable harness | Chassis edges | 10 mm | No, add guide |
| Thermal expansion | Adjacent pipe | 6 mm | No |
You will be amazed how many apparent clashes disappear once someone is forced to name the clearance. Half the time the real answer is that nobody ever decided, and the designer guessed, and the checker guessed differently, and the machine was built on the intersection of two guesses.
3. Tolerances Are Where Interference Is Born
A model at nominal size is a fiction. Real parts vary, and the variation is where interference appears between two parts that never touch at exactly nominal. The honest approach is to check three bounding conditions: the worst-case interference condition, where every dimension stacks up to close the gap; the nominal, where the model sits; and the worst-case gap condition, where everything opens up, useful for rattling and loose fits.
I keep two commands wired into my check routine: a full deviation sweep on the assembly with the tolerance table applied, and a targeted worst-case study on the handful of pairs the matrix marks critical. The full sweep finds the surprises, and the targeted study gives the numbers that you can defend to the workshop foreman without sounding like you are guessing.
Worst-case clearance = nominal clearance - sum of the maximum geometric deviations that close the gap. If the number goes below the matrix minimum, the pair fails and needs a detail change.
Statistical tolerancing is the tempting replacement, and I use it for large production runs where the stack-up is real Gaussian noise. On a one-off machine, the sample is one, statistics are a fantasy, and the worst case is what actually gets built. Design the one-off for the worst case, and you avoid the phone call.
4. The Moving Parts: Interference Is a Motion Problem
The static interference check catches parts that touch at a fixed pose; the motion study catches the ones that would collide in between. Any part with a degree of freedom, a rotating pulley, a pivoting arm, a sliding drawer, needs a swept-volume or motion-envelope check, because the interesting collisions always happen at some intermediate angle where nobody looks.
I cannot tell you how many times a customer has handed me a machine where the arm clears in the home position and the parked position, and jams exactly at 60 degrees of travel, right where the torque is highest. A simple motion sweep from 0 to 90 degrees in coarse steps would have found the clash on the first day, and it takes minutes, not weeks.
- Identify every part with a degree of freedom and its travel limits.
- Run a swept-volume or motion-envelope check across the full travel.
- Check the envelope against all stationary parts and other envelopes.
- Add a generous margin for flex, especially in long arms and belts.
- Repeat the check after any mate or dimension change.
Step five is the one that gets skipped, and the skipped check is the one that gets built. The assembly changes three times during the detail phase, and the motion check ran once, on the first layout. Make the motion sweep part of the release gate, not a one-time adventure.
5. Fasteners, Cables and the Invisible Envelopes
The two most productive hours in interference checking are spent on the things the solid model does not know about. Fasteners: a screw head is a solid, but the socket drive, the wrench access, and the space needed to slide a bolt in from the side are not in the model. I model a tool-access envelope on critical fasteners as a transparent body, and suddenly half the ‘fine at nominal’ fits show up as impossible to assemble.
Cables and hoses are the other blind spot. A 10 mm cable routed through a 12 mm slot looks perfectly fine in the rigid model, and then the harness bundle is 22 mm and the slot is off by the bend radius. I route cables as tubular bodies with their real bundle diameter and a generous radius, and I check the bend against adjacent parts. The envelope eats the clashes before the electrician does.
The machine does not hit the classification ‘clash’ until it has been assembled. The rigid assembly misses every flexible thing, and the flexible things are exactly the ones that cause the field jam that stops production at 9 a.m.
6. A Release-Gate Routine You Can Run Every Time
Here is the routine that has served me well and that I run before every assembly release, in this order. First, the matrix review, confirming who may touch whom and the signed minimums. Second, the static interference sweep with the tolerance band applied. Third, the motion envelope sweep for every degree of freedom. Fourth, the fastener and cable envelope pass. Fifth, the wander check, rotating and translating every part to its extremes to catch the used-to-fit-then-changed case.
On a recent conveyor design the routine caught a real one: the belt tensioner arm, redesigned to take a bigger bearing, now swept through the motor terminal box during its travel. Nobody had touched the motion study after the bearing change, and the clash would have been found by the first commissioning electrician instead of by the release gate. The whole catch took forty minutes, including the fix, and it saved a field trip.
That is the honest pitch for this discipline: the release gate is boring, it is overhead, and it will infuriate the designer who is sure the model is fine. It is also the cheapest insurance a machine builder has, because the alternative is discovering the interference on the shop floor, where the fix costs a week and the argument costs a friendship.
7. Common Mistakes and How I Unlearned Them
Every interference habit has a bad version that lives comfortably in a busy office. Running the static check once and never again is the classic, and it is wrong the instant any part changes. Checking only the visible view, so the clash behind the panel is invisible until the panel ships. Trusting a clean interference report on a model with suppressed parts, because suppression hides the clash and the report is honest about what it cannot see. I have read more than one report claiming zero clashes where suppressed fixtures hid the entire conflict.
The correction is dull but decisive: check against the fully loaded and unsuppressed master assembly, run the report as part of the ticket, and make the interference report part of the definition of done. If the report is not attached, the release does not proceed. It sounds administrative, and the first reaction is grumbling, and the second reaction is gratitude when the machine assembles first time.
Finally, do not confuse the report with the judgment. A clean interference report is the beginning of the conversation with the workshop and the electrician, not the proof that the conversation is unnecessary. The geometric check finds what it can model; the skilled assembly technician finds the rest, and your job is to give them a machine whose modeled conflicts are already zero so the unmodeled ones have your full attention. That is the division of labor that actually keeps machines from jamming at 9 a.m., and in my experience it is the whole game.
8. Setting Up the Tools So the Routine Runs Itself
The discipline only survives if the tools make it cheap, so I set up the CAD environment so the routine runs itself. I create a named saved position for the assembly checks with the transparency rules applied, the subset groups isolated, and the interference settings saved, so opening the master and clicking one saved position runs 80 percent of the routine. The saved position is the difference between a routine people run and a routine people talk about running.
I also keep a short template do list in the project folder, with the matrix table and the five passes as checkboxes, and the release sign-off box on the last line. It is a paper check, but the account is on the release ticket, so it cannot be silently skipped. When a designer objects that the routine is slowing the project down, I remind them that the routine is why the last three machines assembled first time, and that the alternative is to let the workshop be the interference checker, which it has always been, and which always ends in a field trip and a better story.
One more tip that sounds too small to matter and turns out to matter a lot: color the checked-safe zones. Reserve a set of colors for the final passes, green for verified clearances and amber for envelope bodies, so a skim over the assembly tells you instantly where the checks have been and where they have not. The model becomes the checklist, and the checklist stops living in someone’s memory. On a large welded frame with forty subassemblies, that visual scan finds the forgotten corner in seconds, which is exactly where the interference was going to hide.
