Mechanical Engineer Interview Questions

Not your career. This week.

Say what you produce — drawings, calculations, method statements, a machine that has to run on Monday — and at what scale. "I release about fifteen part drawings a month for a two-hundred-tonne press line" tells them more in one sentence than three minutes of narrative.

These are three different careers wearing one degree, and the interviews barely overlap. Design gets calculation reasoning and CAD depth. Manufacturing gets process, tolerance and takt. Maintenance gets failures, spares and the night shift.

Answer with the one you want, then name the crossover you have. Saying "any of them" reads as not having decided, and the first month exposes it.

Have the dates ready. FE passed, EIT status, years of qualifying experience, target sitting for the PE. In the UK and Ireland: IEng or CEng, which institution, whether you have a mentor, and when you expect to submit.

A mechanical engineer whose work sits under the industrial exemption may never license at all, and that is a fine answer said plainly. Vagueness is not — it reads as drift.

Separate the two piles out loud: what you have built real work in, and what you have opened. SolidWorks, Creo, NX, Inventor, CATIA on the modelling side; ANSYS, Abaqus, Fusion simulation, MATLAB on the analysis side.

Overstating here costs you two questions later, when someone asks which mesh you used and why.

Both answers are facts. Give the facts.

If you have been designing for more than about two years and your answer is that nothing has failed, one of two things is true and both are bad: you have not shipped anything, or you did not follow it into the field.

The interviewer is not measuring the failure. They are measuring the loop — how you found out, how fast, what you changed, and whether the change was in a drawing or only in your head.

Both versions below are true stories about the same bracket. Only one gets an offer.

They want to know whether you will sign off something you do not believe, because a mechanical engineer who does that is a liability with a stamp.

Give the sequence: what you were being asked to release, what was missing, what you offered instead — a partial release, a conditional sign-off, a test that could run in parallel. Saying no without an alternative is not a story about judgement, it is a story about being difficult.

The good answers here tend to be about a drawing. A tolerance nobody could hold, a datum scheme that made the part impossible to fixture, a surface finish called out on a face that never gets touched.

Say what you learned about how the part is actually made. Engineers who have been corrected by a machinist and remember it are worth more than engineers who have not.

Avoid the version where you were right and they were unreasonable. "I went and watched them set it up" is the sentence that lands.

One specific instance, with the number the model predicted and the number the rig gave you. This is a fast, quiet test for whether you treat analysis as evidence or as decoration.

Yours or someone else's — they will take either, and the interesting half is what you did about it once it was visible.

Scrapped tooling, a shipment recalled, a line down for a shift. Put a cost on it, because engineers who cannot cost their own errors tend not to weigh them properly either.

The honest answer is a list of artefacts: the model tree, the calculation file, the assumptions register, the open items. If your handover is a conversation, say so and say what you would add.

Do not recite definitions. Draw the curve in the air and point at the three places anybody designs against: the proportional limit, the yield point, and the ultimate tensile strength. Young's modulus is the slope of the first bit.

Then say which one you design to and why. For ductile metals under static load it is yield, with the factor of safety on top — and saying that unprompted tells them you have used this rather than memorised it.

This one separates people fast, because the weak answer is a number and the strong answer is a list of unknowns.

The factor is not a fudge. It is the accounting for everything you did not model: variation in material properties, uncertainty in the load, corrosion, temperature, the quality of your own analysis, and the consequence of the thing breaking.

So talk about what drives it up and down. A lifting accessory with a person under it, a fatigue-loaded weld, a casting with unknown porosity — up. A well-characterised material, a load you have actually measured, a failure mode that is a leak rather than a collapse — down.

Then, and only then, mention that the number is usually not yours to invent: the code has one. ASME BPVC, the lifting standard, the customer specification. Engineers who reach for the code first look experienced.

The trap is answering like a textbook. Energy is conserved; entropy sets the direction and tells you what you can never get back.

The follow-up is usually about efficiency, so have the practical version ready: why no heat engine gets everything, why a chiller is rated on a coefficient of performance rather than an efficiency, why waste heat rejects to somewhere.

Stress concentration — the local stress at a geometric discontinuity is several times the nominal stress, and fatigue does not care about your average.

Say what you do about it: radius the corner, polish the surface, put the weld somewhere else, shot-peen to get compressive residual stress in the skin. Naming a countermeasure is what turns this from a viva answer into a design answer.

Steels show a stress below which they will run essentially forever; aluminium does not, which is why aluminium structures are designed to a life rather than to infinity. That contrast is the whole question.

Reactions first, then shear, then moment — and say that out loud as you go. The maximum moment sits under the load; if the load is central it is PL/4. They are watching the order, not the arithmetic.

Reynolds number, and then say what it changes: pressure drop, mixing, and how much heat you can move through a surface.

The useful follow-on is that turbulence is often what you want. In a heat exchanger you pay for it in pumping power and buy it back in transfer coefficient, and being able to describe that trade is worth more than quoting a transition value.

Head against flow, with efficiency and power laid over it. The operating point is where that curve crosses your system curve, and the answer they want is that you sit near best efficiency rather than out at either end.

Then say what goes wrong away from it: recirculation and vibration at low flow, cavitation risk and motor overload at high flow. If you have ever throttled a pump to move its operating point and watched the bearing temperature drop, lead with that.

Asked constantly, and the answers divide neatly into people who have heard a cavitating pump and people who have not.

The mechanism first: pressure at the impeller eye drops below the vapour pressure of the liquid at that temperature, bubbles form, they collapse violently a few millimetres later where pressure recovers, and the collapse takes metal off the impeller. It sounds like gravel.

Then the bookkeeping: NPSH available is a property of your system — static head, minus friction, minus vapour pressure — and NPSH required is a property of the pump at that flow. You need available to exceed required, with a margin.

Then the fixes, because that is the real question. Raise the suction level, shorten and fatten the suction line, cool the liquid, slow the pump down, or trim the duty back toward best efficiency.

Give the order you actually use, not a list of properties. What has to be true — strength, stiffness, temperature, corrosion, weight, whether it is food or medical contact. Then what it costs, then whether you can get it, then what it does to the process.

Availability is the one candidates forget and buyers never do. A grade with a fourteen-week lead time is the wrong grade.

Answer in terms of what you are trading. Annealing softens and relieves stress; normalising refines and evens out the grain after forming or welding; quenching gets hardness and brittleness together; tempering gives back toughness at the cost of some of that hardness.

The sentence that makes it land: you do not ship a part as-quenched, because you have made something that is hard and will shatter.

It is for saying what the part has to do, rather than what it has to measure. Position, flatness, perpendicularity, profile, and the datum scheme that ties them together.

The thing to say out loud is that a good datum scheme mirrors how the part is fixtured and how it locates in the assembly. Engineers who mention fixturing here get believed.

A stack-up question is really a question about whether you have ever had an assembly that would not go together.

Describe the method: identify the gap you care about, build the chain of dimensions that produces it, decide whether you are doing worst-case or statistical, and run it. Say which you would use and why — worst-case for a small production run or a safety-critical fit, statistical when the volume justifies it and you know the process capability.

Then the closing moves, in the order that costs least: change the nominal, redistribute the tolerance onto the features that are cheap to hold, add an adjustment feature, select-fit, and only then tighten everything and pay for it.

Pick one process and be concrete. Draft and uniform wall for moulding, radii and tool access for machining, bend reliefs and minimum flange for sheet metal, feed and riser thinking for casting, joint access for welding.

If you have sat with a supplier and changed a part after their feedback, that story is the answer. It beats any list.

Have one ready. Say what you had not prepared, and what you now bring to a review that you did not bring then.

Do not start calculating. Start listing what you need: the duty, both fluids and their flow rates, the inlet temperatures, the allowable pressure drops, the fouling you expect, and the material the fluids will let you use.

Then the shape of the method — energy balance to get the duty, log-mean temperature difference, an overall coefficient, area falls out — and be explicit that the coefficient is where all the uncertainty lives.

Break it into the buckets and the answer writes itself: envelope gains through fabric and glazing, solar, people, lighting, equipment, and fresh-air load, which in a hot humid climate is often the one that dominates and the one juniors leave out.

Name what you have used — a load calculation package, a spreadsheet against a published method — and say that you keep the diversity assumptions written down, because that is the number that gets argued about later.

This is a diagnosis question, and the mark is for having an order rather than a favourite cause.

Cheap and quick first: is it actually new, and what changed? Duty, suction conditions, a valve someone throttled, a strainer nobody cleaned.

Then the mechanical shortlist, in the order that is cheapest to rule out: misalignment, unbalance, bearing wear, looseness or soft foot, and cavitation — which you can often identify by ear before you put an instrument on it.

Then say how you would tell them apart, because that is the actual answer. Vibration at once per revolution points at unbalance; twice per revolution with axial content points at misalignment; broadband noise at the suction end points at cavitation; and a coupling that runs hot points at both alignment and looseness.

The second half is the question. Anyone can colour a part red.

Talk about validation: a hand calculation on a simplified case, mesh refinement until the result stops moving, checking the reactions sum to the applied load, sanity-checking the deflection against something you can feel. Say what your boundary conditions were and why they were defensible, because over-constraining is the standard way to get a beautiful and wrong answer.

Instrument before you theorise. Say what you would log and for how long, and admit that the honest first step is often proving the failure is real and repeatable rather than a reporting artefact.