Robot cell and work-cell design

A robot on its own does nothing useful. It becomes productive only as part of a cell: the robot, its controller, the end effector, the fixtures that hold parts, the feeders and conveyors that bring them, the machines it serves, the guarding that keeps people safe, and the control system that coordinates it all. Most of the engineering effort in an industrial robot project goes into designing that cell. This lesson follows the design sequence used by integrators: understand the process and the parts, lay out the cell, check reach and interference, estimate and then simulate the cycle, and plan for maintenance, changeover and safety from the start. Safeguarding gets its own lesson, Lesson 8, but it is a design input from day one, not an afterthought.

Start from the process, not the robot

Good cell design begins with a process study. What parts are handled, in what variants, at what rate, and with what tolerances? How do parts arrive: loose in bins, oriented in trays, on a conveyor, or on pallets? What happens to them in the cell: machining, welding, inspection, assembly, packing? How do they leave? What quality checks are required, and what happens to rejects? Which tasks are best left to people? An integrator usually captures this in a user requirement specification (URS) that the customer approves before a quotation, and later in a functional design specification (FDS) that describes how the cell will meet each requirement.

A frequent cause of failed projects is poor part presentation. A robot that must pick parts dumped randomly in a bin needs 3D vision, a gripper that can cope with many orientations, and error recovery for failed picks, as ROB-400 showed in its bin-picking capstone. If the supplier can deliver the same parts in a molded tray, the problem becomes a simple taught pick at known positions. Changing the presentation is often the cheapest engineering decision in the whole project.

Laying out the cell

A layout drawing places the robot, machines, conveyors, fixtures, operator stations, guarding, the robot controller and electrical cabinets on the plant floor. Designers work in plan view first and then check heights and clearances in 3D. Common layout patterns include a robot in the center serving machines arranged around it, a robot on a linear track serving a row of machines, an overhead gantry above a line, and a two-station layout in which the operator loads one fixture while the robot works on another.

A two-position turntable or a pair of sliding drawers is a classic way to keep the robot busy while a person loads parts. The operator loads one side outside the guarded area while the robot works on the other side, then the table indexes. Because the person never enters the robot's space during normal operation, the cell can run at full robot speed behind ordinary guarding. Lesson 8 discusses how such load stations are safeguarded.

Layout also determines access. Maintenance technicians need room to change grippers, clear jams and service the machines; quality inspectors may need to remove parts; forklifts must deliver material. Leaving at least the clearances required by the guarding standard between the robot's restricted space and the fence, and planning gates where people will actually need to go, prevents later modifications that weaken safety. Robot controllers and electrical cabinets are placed outside the safeguarded space so that troubleshooting does not require entry.

Reach, envelope and interference

The robot's work envelope, drawn from the datasheet, shows every point the wrist center can reach. With a tool fitted, the reachable volume shifts outward by the tool length but also gains orientation constraints: a point may be reachable only with the tool pointing in certain directions. Designers check every target with the actual tool orientation it needs, not just the position. Points near the edge of the envelope or near singularities, which ROB-210 described, are avoided because the robot slows, loses stiffness or must reconfigure its joints.

Three spaces are defined in robot safety standards. The maximum space is everything the robot, end effector and workpiece could sweep if the robot moved to every joint limit. The restricted space is the portion of the maximum space that is limited by mechanical stops or safety-rated soft axis and space limiting, which modern controllers provide through functions with names such as SafeMove, DCS (Dual Check Safety) or Safe Operation. The operating space is the part actually used by the program. Guarding is placed based on the restricted space, so limiting the restricted space with safety-rated functions allows a smaller footprint.

Fixturing and part location

Fixtures hold parts in known positions so that taught points remain valid. The classic 3-2-1 principle locates a prismatic part with three points on a primary face, two on a secondary face and one on a tertiary face, constraining all six degrees of freedom with the minimum of contacts. Clamps then hold the part against those locators without disturbing its position. Robot fixtures must also be robust to how robots load parts: lead-in chamfers, slightly oversized nests and compliance in the gripper allow small errors without jamming.

Tolerance stack-ups decide whether a cell can rely on teaching alone. If the part position in the fixture varies by ±0.3 mm, the fixture is placed within ±0.1 mm, the robot repeats to ±0.05 mm, and the gripper fingers have ±0.1 mm play, the worst-case error is the sum, ±0.55 mm, while a statistical estimate using the root sum of squares is about ±0.34 mm. If the insertion requires ±0.2 mm, the cell needs better fixtures, chamfers, a compliant wrist or sensing.

# Worst-case vs root-sum-square vs Monte Carlo tolerance stack
import numpy as np
rng = np.random.default_rng(440)
tol = {"part in fixture": 0.30, "fixture placement": 0.10,
       "robot repeatability": 0.05, "finger play": 0.10}       # +/- mm
worst = sum(tol.values())
rss = np.sqrt(sum(t**2 for t in tol.values()))
# treat each tolerance as +/-3 sigma of a normal distribution
samples = sum(rng.normal(0, t / 3, 200_000) for t in tol.values())
print(f"worst case  +/-{worst:.2f} mm")
print(f"RSS         +/-{rss:.2f} mm")
print(f"share of parts within +/-0.20 mm: {np.mean(abs(samples) <= 0.20):.1%}")

worst case  +/-0.55 mm
RSS         +/-0.34 mm
share of parts within +/-0.20 mm: 92.6%

Cycle time analysis

Lesson 1 introduced takt time. During design, the cell's cycle time is built up from every robot motion, gripper action, machine operation and waiting period. Engineers draw a sequence chart, also called a Gantt chart or time-and-motion chart, showing what each resource is doing over one cycle. The chart reveals which resource is the bottleneck and where the robot is idle.

Common ways to shorten a robot cycle include dual grippers, so the robot can unload a finished part and load a new one without a trip to the outfeed in between; parallel motions, such as moving the robot while the gripper opens; avoiding wrist reorientation; reducing approach and retract distances; and using blended or fly-by points, called zones or continuous paths depending on the vendor, instead of exact stops at intermediate points. Each exact stop costs time to decelerate, settle and accelerate.

Simulation and digital twins

Robot simulation software, either from the robot vendor or from independent suppliers, models the robot's kinematics, its controller's motion planner, the tooling, fixtures and machines. Vendor tools typically include a virtual controller running the same software as the real one, which gives cycle-time predictions within a few percent. Simulation is used to choose robot models, verify reach and interference, estimate cycle time, generate and test programs offline (Lesson 4), and present the concept to the customer.

A digital twin goes further: it is a model connected to the real cell, updated with live data, used for virtual commissioning before installation and for diagnosing problems after. In virtual commissioning, the real PLC program and robot programs run against a simulated cell, so that logic errors are found at a desk instead of on the shop floor with the customer watching. Studies and vendor case reports commonly claim that virtual commissioning shortens on-site commissioning substantially, though the size of the benefit depends on the quality of the model. ROB-360 covered simulation fundamentals; here the emphasis is on using them to make a commercial decision.

Designing for changeover, maintenance and recovery

A cell that runs perfectly on one part is easy to design. A cell that changes over between ten variants in five minutes, recovers from a dropped part without an engineer, and lets a technician change the gripper fingers safely is much harder and much more valuable. Design for changeover includes quick-change fingers with poka-yoke features that prevent wrong installation, recipes in the PLC that select the right program and parameters, and sensors that confirm the right tooling is fitted. Design for recovery includes clear fault messages on the operator panel, defined home positions, and programs that can resume from any step after a stop. Lesson 10 returns to maintainability.

Integrators often say that half of the programming effort goes into the abnormal cases: missing parts, failed grips, machine alarms, power loss in the middle of a cycle, and an operator opening the door at the wrong moment. A sequence chart that only covers the normal cycle is therefore only half a design.