Everyone’s Talking About Fast 3D Printing, What Matters is Throughput

Why speed is important and how to talk about it

FFF 3D printing can be a slow, tedious manufacturing process. Understandably buyers of printers are concerned about buying the fastest system for their use case. Unfortunately, how machine builders talk about ‘speed’ fails to describe performance differences between machines.

In FFF 3D printing, “print speed” often refers to the max feedrate achieved by the print system during extrusion.

Users ultimately don’t care about how fast the machine can move; they care about how quickly they get their parts. Max feedrate is, however, a poor proxy for cycle time.

These metrics are inherently misleading when comparing machines on their specs alone. We’ve observed a 700–1000% cycle time difference between jobs on 2 machines running ostensibly at the same feedrate.

Here we’ll attempt to explain why this metric fails and suggest alternative ways of measuring cycle time in real-world scenarios.

In addition, we’ll talk about the compromises speed often causes for operators.

Why Care About Cycle Time?

FFF 3D printing is a fantastic manufacturing tool, but one inherently limited by long cycle times. For many users, the benefits of high-speed printing are not immediately apparent, beyond the fact that faster cycles make it so you can make more stuff in a day.

The value of reduced cycle times is unclear until you are 5–10x faster than current popular systems. That is until a 3-day job becomes an overnight job.

5–10x lower cycle times come with several less-obvious benefits for all users:

• Parts may be made 5–10x stronger within a given cycle time
• Large, completely solid parts with high detail are viable in realistic cycle times
• Lower time-cost of mistakes and design errors
• Less floor space and power for the same productivity, generally by the same factor that cycle time is reduced.
• Lower chance of failure due to outside disturbances like power failure

R&D Teams also benefit greatly:

• Staff are more likely to try ideas without fear of wasted time
• Teams may produce daily iterations similar to the agile software development model
• Engineers and designers switch off-task less often and for shorter periods, leading to staying ‘in the zone’ on a particular problem or project

Serial production sees additional benefits:

• Even shorter lead times enable carrying less WIP/inventory of printed parts, meaning design changes can be propagated to production more quickly and with less waste
• Even higher degrees of customization is possible, especially for goods with fast turnaround
• Parts can be produced on-demand in a true “digital inventory” production system
• Jigs and fixtures can be produced and put into use within the same shift where a need was identified

Kinematics of Speed

Fundamentally, the problem with using max feedrate to compare 2 FFF machines’ performance is that machines rarely spend any time at their top speeds.

As a rough rule-of-thumb, the print head only spends the majority of its time cruising at top speed for moves 4 times farther than it needs to get to top speed. For popular belt-based systems on the market, this distance is 30–60mm. Aside from long, straight sections in larger parts, all other moves’ duration is set by acceleration.

FFF printers are acceleration-limited for most jobs. That is to say, most of the time in any part is spent speeding up and slowing down, rather than cruising at the maximum feed rate.

You can get an intuition for this by analyzing Gcode files. We lent a machine to an R&D job shop and collected 100 Gcode files from their machine. The two plots below show the distribution of move lengths across 169km of travel. 50% of all distance traveled is in moves under 8mm in length, and 75% of all distance traveled is in moves under 42mm. It is safe to say that max feedrate is not a predictor of cycle time for most print jobs.

Instead, acceleration is what FFF printers spend most of their time doing. Motion control strategies improve performance by chaining moves together, so the head does not come to a complete stop, and every relevant system today uses these strategies. However, control systems innovation cannot work around the fundamental kinematics of the underlying system.

Of course, talking about maximum acceleration is not much better than max feedrate, as pushing speed or acceleration can lead to unacceptable quality problems in finished parts. The following section discusses some of these problems and how they present themselves.

Quality, Functionality & Reliability

For any FFF printer, as you reduce cycle time, you reduce quality until, at some point, you aren’t producing parts; you’re producing plastic waste. These issues can ultimately result in a scrapped part and wasted production time.

Operator Problems:

• Reduced resolution
• Reduced strength

These are less machine design issues and more an emergent behavior in staff using slow machines. The user attempts to reduce cycle times by compromising on part quality. This can be done by increasing nozzle size or layer height or reducing wall thickness and infill. Often when users say they’ve “halved their cycle time” on a part, they’ve reduced resolution and wall thickness. This strategy may be viable for toys and trinkets but is rarely acceptable for end-use goods.

Motion system errors:

• ringing (i.e., overshooting)
• rounding (i.e., undershooting/cutting corners)

These issues result in poor surface finish and missed tolerances. Low stiffness, inadequate motor torque, control loop tuning, or motion control strategy cause these issues. They are exacerbated by pushing accelerations higher than the motion system can realistically handle.

Extrusion problems:

• Delamination
• Under-extrusion
• Porosity

Inadequate heater power or hydraulic-flow limits in the hotend coupled with slipping in the extruder cause these problems. This results in poor surface finish and mechanically unusable parts. The material is also a significant factor in these problems. Materials must be designed for high flow to work reliably as you push deposition rates higher.

Reliability problems:

Very few 3D printers are designed for industrial use, to be run 24/7 for years. Increased speeds exacerbate these problems.

The way subsystems of 3D printers are often not up to the task of printing 5–10x faster. At higher speeds, you rack up mileage quickly, and components that used to last 2–3 years wear out after a couple of months.

In a production system, things need to be built with maintenance and upkeep in mind. Machines are rarely designed for easy maintenance and fast swapping of parts to minimize downtime. Ultimately, bearings wear out, cables fatigue from bend cycles, and filament paths wear through friction. From our experience, when a 3D printer breaks, it’s days before it’s back online.

Something else often left out of “cycle time” is how much machine maintenance goes into it. If a machine takes 1 hour to make a part and requires a day of downtime, can you say that part took an hour? A top-fuel dragster does a quarter-mile in 4.4 seconds, but how long will it take to get from New York to LA?

Benchmarks

Comparing based solely on maximum feed rate and acceleration tells you nothing about the real-world performance of any given machine on its own and forms no basis of comparison when talking about cycle times or productivity.

Seeing the success of the speedboat race, it’s clear that there’s an appetite for speed-based benchmarks for FFF 3D printers. However, speed benchies don’t test for the needs of commercial & industrial users.

What is needed is a standard speed benchmark that includes requirements for part quality and usability. This benchmark should represent a realistic manufacturing scenario and be long enough to verify that no part of the print system thermally saturates or otherwise experiences fatigue during the job.

We propose defining a benchmark with features to test and verify:

• Tolerancing & surface finish
• Mechanical properties in all axes
• Adequate rendering of small details
• Successful printing and removal of supports.

This benchmark should include specifications for:

• Line width/nozzle diameter
• Wall, floor, and ceiling thickness
• Infill type and density
• Total printed volume

Ultimately, the reported values should be:

• Cycle time to a complete, ready-to-use part, including powerup, warmup, and cooldown of the machine.
• Normalized cycle time (NCT) based on hours to deposit a liter of material.

No one part can test the full sweep of use cases; instead, a hybrid benchmark should be created, similar to EPA City/highway fuel economy.

A printing benchmark based on one large, relatively simple 0.5–1kg part and one based on a run of several 100–200g complex parts gives the best impression of real-world user experiences.

Here are the 2 build trays we use for internal benchmarks right now:

As a point of comparison, here are the cycle times for these parts for several other popular machines on the market¹:

For the electronics enclosure, a job started at 4 PM on Wednesday is either ready the following day on the fastest system, or Friday afternoon for the rest. In the R&D context, this means getting 2 more iterations in a week.

For the auto brackets, the variability is much higher. Some machines are finished on Thursday, some Friday, and some need the weekend to run. This is how deadlines are made or missed.

All these machines have comparable maximum feed rates, but while printing, these parts perform wildly differently. Looking at normalized cycle time (NCT), we get the numbers below for these parts:

We propose a Many Small/One Big benchmark, analogous to the city/highway MPG. At a glance these 2 numbers give a far better intuition for how fast a particular machine will do a variety of jobs.

Normalized Cycle Time should be published nominal density of 1.1g/cm3 as a rough average of the most popular engineering materials used in FFF: ABS, PETG, PA and PC. By standardizing density we produce a more intuitive benchmark.

This value may be calculated for a job using the following formulae²:

We chose PCT in hrs/kg because we found it the most intuitive number to use in our heads. If a part is ~500g, you just multiply it to get the job time. It makes it simple to compare how long jobs take between machines, and leads to better intuition than the other popular metric, material deposition rate in mm³/s.

Conclusion

Speed and reliability are the keys to making FFF 3D printing an essential manufacturing tool with the same broad adoption as machining, molding, and other production processes.

Part of accelerating this adoption is using performance specifications that are meaningful to manufacturers. The question users want answered is “How many parts of acceptable quality can I make in a day?” This is the ultimate question that decides if a manufacturing process is viable or not.

Many Small/One Big benchmarks and normalized cycle time are tools that can help professional users get a better intuitive understanding of the performance differences between systems and guide the industry in improving speed in easier-to-understand ways that more directly impact the user experience.

Footnotes:
[1]
Based on slicer-reported print times for each vendor’s default material. F370 runs the bracket job in 2 batches, Stratasys slicer only permits 0.007” (0.178mm) layers.
[2]
The magic number 909,000 is derived from 1/(1mm³/hr*1.1g/cc) in hr/kg
The magic number 55.5 is derived from 1/(1in³/hr*1.1g/cc) in hr/kg