Post-Processing is where the build actually fails
I spent three days last month trying to get a resin-printed titanium alloy bracket to hold tolerances within 0.05mm. The printer itself was fine. The sintering cycle was within spec. What killed it was the post-cure shrinkage I hadn't accounted for in the slicer settings. You set the scale factor to 1.03 to compensate for the ~3% linear shrinkage during debinding, but the bracket geometry had thin walls on one side and a massive cross-section on the other, so the shrinkage was uneven and the part warped during the second sintering pass. I ended up printing it with a symmetric lattice infill pattern that distributed thermal mass more evenly, then re-machined the bearing seats on a CNC mill afterward. Total time from design to final part: about 6 hours including the failed first attempt. That's the thing nobody puts in the marketing brochures about New Additive Manufacturing Technologies. The hardware got good enough around 2019 that the bottleneck shifted entirely to process knowledge. You need to understand the material behavior at each stage, not just push a print button.
What New Additive Manufacturing Technologies Actually Means Now
The term has drifted from its original meaning. Five years ago it basically covered FDM printers buying parts on MakerBot and Dremel machines doing quick prototypes. Today it refers to production-grade processes like direct metal laser sintering, binder jetting, continuous liquid interface production, and electron beam melting. The common thread isn't that they make things layer by layer. It's that they decouple geometry from tooling, which changes how you design and where you absorb cost in the workflow. I work mostly with metal AM for aerospace brackets and fluid systems. The processes I touch regularly are LPBF, DMLS, and binder jetting. Each has a completely different cost curve and defect profile. LPBF gives you the best mechanical properties but costs roughly $8 to $15 per cubic centimeter of build volume depending on machine utilization. Binder jetting runs closer to $2 to $4 per cubic centimeter but requires infiltration or secondary sintering that can introduce porosity if you don't control the cycle precisely. The slicer settings for metal AM look similar to what polymer users see, but the parameters mean something completely different. Layer thickness in metal is usually between 20 and 60 microns, not the 100 to 300 you see on resin printers. Hatch spacing controls the energy density directly, and the wrong setting here causes either lack of fusion pores or keyhole porosity, both of which show up as stress concentrations in fatigue testing. A typical setting is 100 to 130 microns hatch spacing with 110 to 150 watts laser power on a 1070 steel alloy.
The workflow most people get wrong
They design the part in CAD, export STL, load it into the slicer, and hit start. The part prints fine on the first go, sits in the chamber, gets removed, and then fails inspection. Half the rejects I've seen over the last two years came from this sequence. The issue wasn't the printing. It was that the part didn't have support structures designed for thermal management, the orientation put critical stress planes parallel to the build plate instead of at the recommended angle, and the surface finish on functional features was left as-built when it needed to be machined. I now run a design-for-AM review before anything goes to the machine. It takes about 20 minutes for a simple bracket and maybe an hour for a complex assembly. The checklist is straightforward. First, verify the minimum feature size. Most LPBF machines handle 0.3mm walls reliably, but anything below 0.5mm needs a support structure or it'll float during scanning. Second, check overhang angles. Free-standing surfaces above 45 degrees need supports unless you're using a process like CLIP that doesn't require them. Third, orient the part so the primary load path is vertical. Tensile strength along the build direction is typically 5 to 10 percent lower than transverse strength in most metals due to the columnar grain structure. Support design is another area where beginners lose money. Most slicers have a default support setting that's way too aggressive for production parts. I use minimal contact points with a density of about 30 percent on flat surfaces and switch to tree supports for overhanging features. Tree supports reduce material usage by roughly 60 percent and cut post-processing time from maybe 45 minutes per part down to 10 or 15 minutes. The tradeoff is slightly rougher contact points, which you file or grind off anyway.
Material selection is more important than machine selection
I've seen people buy time on a 3D Systems DMP 320 and expect it to print AlSi10Mg the same way a SLM Solutions 280 would. It won't. The DMP 320 has a 400-watt fiber laser and a smaller build envelope. It can print AlSi10Mg, but the scan strategy and parameter set from a 1kW machine won't transfer directly. You'll get different grain sizes, different residual stress profiles, and different surface quality even if the nominal settings look identical in the file. The material databases from OEMs are useful starting points, not final answers. I always run a small test coupon set when switching materials or machines. A 20x20x20mm block printed at three different energy densities gives you the data you need to adjust the main part settings. Print at nominal, then scale the laser power by plus and minus 15 percent while keeping scan speed constant. Measure the resulting relative density with Archimedes method. The sweet spot is usually where density peaks before cracking starts, which for most aluminum alloys is around 60 to 80 joules per cubic millimeter of volumetric energy. Post-processing choices matter more than people realize. Shot peening after printing can improve fatigue life by 30 to 50 percent in most aerospace alloys because it introduces compressive residual stress at the surface. But if you shot peen a binder-jetted part that hasn't been fully densified, you can actually drive surface pores deeper and make the problem worse. Know your process before you apply treatments.
When additive manufacturing is the wrong answer
CNC machining still wins for high-volume production of simple geometries. If you need 500 identical brackets with tight tolerances on two holes, a CNC mill with a fixture will do it in a fraction of the time and at lower per-part cost after the first unit. AM excels at low-volume, high-complexity parts where the geometry justification outweighs the unit cost. A single prototype bracket with internal cooling channels might cost $300 in AM and $2000 in CNC with custom tooling, but 500 units flip that equation completely because the AM per-unit cost stays relatively flat while CNC costs scale with machining time. Surface finish on as-printed metal parts is another place where expectations get unrealistic. Ra values of 10 to 20 micrometers are typical on LPBF surfaces. That's rougher than a milled part and rougher than most powder-coated polymer prints. If your application needs a smooth sealing surface, plan for secondary machining or EDM finishing. Budget 1 to 2mm of stock removal on critical faces. Quality certification is another practical limitation. If you need AS9100 or NADCAP documentation on every batch, the per-part cost jumps significantly. Material certificates, process documentation, and destructive testing on witness samples add time and cost that don't appear in the machine hourly rate. I budget roughly $50 to $150 per part for documentation overhead on certified builds, depending on the customer requirements.
Practical resources and what to look for
Open-source parameter sets for common alloys exist but are unreliable without validation. The NASA AM handbook and the SAE AM standards library are better references for baseline parameters. If you're just getting started, look for machine vendors that include application engineering support rather than those that sell bare equipment. A good vendor will spend two days helping you optimize parameters on a test block and save you weeks of trial and error. The equipment is commodity at this point. The knowledge is the value. For software, Meshmixer is fine for basic orientation and support addition, but Simplify3D and proprietary slicers from vendors like Desktop Metal and Markforged give you more control over scan strategies and thermal management. If you're doing production work, invest time in learning the thermal simulation features. Predicting warpage before you print saves more money than any post-processing fix. The field moves fast. What was cutting edge in 2020 is standard now. In 2023, multi-laser systems became common on mid-range machines, which cut print times roughly in half compared to single-laser equivalents. By 2025, AI-driven closed-loop monitoring started appearing on new machines, which catches defects in real time rather than after the build completes. The underlying physics hasn't changed. The speed of iteration has.
