Share this article
Continue reading with these related posts
One of the most overlooked costs in automation is the ongoing human cost of keeping it running. This post helps you assess your team's current capabilities, identify the training paths that actually work, build a preventive maintenance programme, and decide when a support contract is worth the money.
Building an automated cell without a proper safety assessment is like building a house without a foundation inspection — everything looks fine until something goes wrong. This post walks you through ISO 12100, physical safeguards, safety PLCs, and lockout/tagout procedures to keep your team and your operation protected.
Connecting a robot to the rest of your factory is often harder than programming the robot itself. Learn the communication protocols, handshake sequences, legacy compatibility techniques, and staged implementation strategies that make robot integration succeed.
In the previous post, we covered the in-house skills and maintenance practices that keep an automated cell running after the integrator has left. In this post, we set theory aside and walk through a real project: the cobot cell we designed and deployed for a digital print shop client.
This is not a polished success story. It is the honest account — the decisions that worked, the assumptions that did not hold up, the integration problems that cost us a week each, and the final numbers we measured after a year of production. If you are planning an automation project of your own, the gaps between expectation and reality described here will be more useful to you than any smooth case study.
The client is a mid-sized digital print shop producing custom-decorated items for the promotional products and corporate merchandise market. Their primary production equipment is a fleet of UV flatbed printers capable of decorating a wide range of substrates: rigid panels, moulded plastic items, metal plates, and wood blocks. The printers are fast — a high-resolution print cycle takes two to four minutes per bed — but loading and unloading them manually required an operator standing at the machine for most of the shift.
What Is UV Flatbed Printing
A UV flatbed printer uses ultraviolet-cured inks printed directly onto rigid substrates without contact. The substrate sits flat on a moving bed, the print head travels across it depositing ink, and UV lamps immediately cure the ink as it is laid down. The result is a durable, full-colour image on virtually any flat surface up to the machine's maximum bed size.
Before automation, each printer required roughly half an operator's attention during a production run. An operator would load a tray of items, start the print cycle, and need to return before the cycle finished to unload and reload — otherwise the printer sat idle waiting. With four printers running simultaneously, this meant two full-time operators whose entire job was walking between machines, loading, and unloading.
The owner's goal was straightforward: free up those two operators for value-added work — quality inspection, packaging, order preparation — while maintaining or improving throughput. The secondary goal was to reduce ergonomic strain; loading heavy panels repeatedly across an eight-hour shift had resulted in several repetitive strain complaints over the previous two years.
The economics made sense on paper. With the two operators fully redirected, the payback period on the automation equipment was under 18 months at current production volumes, with the potential to drop further if production grew — the cell could run unattended during breaks and shift changes, something manual operation could not.
Before any design work started, we documented the operational requirements and the hard constraints the cell had to work within:
Operational requirements:
Hard constraints:
Budget Caps Shape Everything
A firm budget ceiling is not a problem — it is a design parameter. It immediately eliminated several options (a vision-guided bin-picking system, a second cobot for unloading) and forced us toward simpler, more reliable solutions. Simple systems fail less, which often makes them better choices even without a budget constraint.
We evaluated three platforms: Universal Robots UR10e, FANUC CRX-10iA, and Doosan A0912. All three had adequate reach (the printer bed required approximately 900 mm of reach from a practical mounting position) and sufficient payload (the heaviest substrate was 4 kg; we needed at least 6 kg to leave margin for the end-of-arm tooling).
We selected the UR10e for three reasons. First, Universal Robots' Polyscope programming interface was the most approachable for the client's maintenance team — after two days of training, their lead technician was able to create simple programs independently, which was a stated requirement. Second, the UR10e's built-in force/torque sensing at the tool flange gave us the ability to detect substrate pick failures without adding an external force sensor. Third, the UR ecosystem has a large library of certified end-of-arm tooling from third-party suppliers, which simplified the gripper selection process.
The FANUC CRX-10iA was a strong technical contender, but its programming environment requires more background to use confidently, and the client's maintenance budget for ongoing training was limited.
The substrate variety was the hardest tooling problem. Rigid panels, moulded plastic items, and wood blocks have different surfaces — some smooth and flat, some textured, some with through-holes or recesses. A single gripper that handles all of them reliably does not exist off the shelf.
We settled on a modular vacuum gripper design: a rectangular aluminium manifold with four independently switchable vacuum zones, each serviced by a Schmalz suction cup. By activating only the zones that align with a particular substrate's flat area, we could handle most of the product range with one tool. The manifold was designed with a quick-disconnect on the robot flange, allowing the full EOAT to be swapped in under two minutes for the small number of substrates (heavy curved items) that required a different approach.
Fixture design for the input and output stacks was deliberately simple. Substrates are manually loaded by an operator into a gravity-fed magazine — a welded aluminium channel with adjustable side guides for different widths. The robot picks from the bottom of the stack, which maintains a consistent pick position regardless of how full the magazine is. The output side uses a simple inclined tray that gravity-sorts completed items away from the robot's path.
Design the Fixture Before the Robot Program
The fixture determines the robot program, not the other way around. Lock down fixture geometry — pick position, stack height, orientation tolerance — before writing a single line of robot code. Chasing a moving target in the fixture while simultaneously debugging the program multiplies your commissioning time.
The printer itself sits inside its own enclosure with an interlocked access door — this was already compliant with provincial machinery regulations. Our task was to define the safe operating zone for the cobot and how it would interact with operators during loading.
We chose collaborative operation with speed and force limiting rather than full perimeter guarding, for two reasons. First, the available floor space made a full cage impractical without significantly complicating operator access to the printer for maintenance. Second, the UR10e is certified for collaborative operation under ISO TS 15066 up to its rated payload.
The safety system consists of:
We completed a full ISO 12100 risk assessment before finalising the layout, which resulted in one significant design change: the original magazine position would have required the robot to move over the operator's head during loading. We relocated the magazine to the side, adding 200 mm of reach to the robot's cycle path but eliminating the overhead motion hazard.
Challenge 1: Printer communication protocol. The printer had no documented external control interface. The manufacturer's support team confirmed the printer had a serial command interface that was "not officially supported for third-party integration." After two weeks of working through the protocol documentation we eventually obtained, we were able to send job-start commands and receive cycle-complete signals. The lesson: assume undocumented interfaces will take twice as long as documented ones.
Challenge 2: Substrate surface variation. During commissioning testing, vacuum gripper performance was inconsistent on matte-finished wood substrates — the surface porosity caused gradual pressure loss over the pick-and-place cycle. We resolved this by switching from a flat suction cup to a bellows-style cup on those zones, which maintained seal under slight surface irregularities, and by adding a vacuum pressure check at the pick point before the robot committed to the move. The check added 0.4 seconds to every cycle, which was acceptable.
Challenge 3: Static charge on plastic substrates. UV-printed plastic panels accumulated static charge during the print cycle, causing them to stick together in the output tray. The robot would occasionally pick two panels at once, deposit both, and the printer would run a second cycle on an already-printed item. We resolved this with an ionising air bar mounted at the printer exit — a $400 fix that took half a day to install after we identified the cause. The diagnostic process took three days because the problem was intermittent and position-dependent.
Challenge 4: Training time underestimated. We planned two days of training for the client's lead technician. The technician was engaged and capable, but by the end of day two it was clear that two days was not enough for independent operation. We returned for a third training day three weeks after commissioning, which was ultimately more effective — the technician had spent three weeks observing the cell in production and arrived with specific, targeted questions. Build this second-visit training into your project from the start.
After twelve months of production operation, the measured outcomes against the original goals:
| Metric | Target | Actual |
|---|---|---|
| Operator touches per shift at target printer | Eliminate continuous attendance | 8–12 per shift (magazine refill, fault recovery) |
| Print throughput (beds per shift) | Maintain or increase | +18% (reduced idle time between cycles) |
| Ergonomic strain incidents | Reduce | Zero incidents at target printer post-deployment |
| Unattended run duration | 20 minutes minimum | 35–40 minutes average |
| Payback period (projected at time of sale) | 18 months | On track at 12 months |
The throughput increase was a pleasant surprise. We expected to roughly maintain the manual throughput by eliminating the operator as the bottleneck. What we did not fully account for was that manual operators naturally cluster their attention — they might load the printer slightly late, or take slightly longer at an adjacent task. The robot loads at exactly the right time, every cycle, which accumulates into meaningful throughput improvement over a shift.
The 8–12 operator touches per shift was higher than the original goal of "run unattended," but in practice this was not a problem. The client found that the operators appreciated having a clear, bounded task (magazine refill) rather than continuous monitoring, and the freed attention was immediately applied to quality inspection — which caught several job errors that would previously have been discovered only in packaging.
The Numbers Held Up
Twelve months in, the cell is running in every production shift, the maintenance technician handles fault recovery and minor program edits independently, and the projected payback is on track. The unglamorous work of fixture design, safety assessment, and staff training made the difference between a working cell and an expensive piece of equipment that sits idle.
Invest more in the fixture early. We underspecified the input magazine's substrate height sensor, which caused occasional jams when the stack ran low and the robot attempted a pick on an empty magazine. A simple ultrasonic distance sensor at the bottom of the magazine would have caught this. We added it as a field modification in month two.
Document the integration protocol immediately. The serial communication protocol we reverse-engineered from the printer took significant effort. We captured it in a technical note during the project, but that note needs to live in the client's documentation package, not just in our project files. Two of the three fault calls in year one involved communication faults that the client's team could not diagnose without calling us.
Plan for substrate variety growth. When we designed the system, the client's product range was 85% flat panels. During the first year, a new product line with thicker, heavier items was introduced. We accommodated it with a EOAT swap, but the process took a week of on-site work that would have taken a day if we had designed the tooling change procedure in from the start.
The collaborative safety approach was the right call. In hindsight, hard guarding would have created more problems than it solved in this application. The operators interact with the cell frequently enough that a full cage would have generated constant gate openings and cycle interruptions. The scanner-based approach required more careful risk assessment work upfront, but the operational result is a cell that feels natural to work alongside rather than adversarial.
