Robot makers: accuracy, vibration, cycle time, control skills

Absolute accuracy falls short of the task and calibration is manual

What happens. A robot repeats well but does not hit a programmed point exactly. The real link lengths, joint offsets and gearbox play differ a little from the model, so the tip is a bit off. For precise tasks the cell is taught point by point or calibrated by hand.

What it costs. Rework, taught points that must be redone after every change and tasks that cannot be given to a robot because it is not accurate enough.

Why the usual fix fails. Stiffer and heavier arms cost speed. A laser tracker once in the factory is accurate and has to be repeated whenever something is changed.

How a fixed-price remote diagnosis checks it. From measured positions against commanded ones at a set of poses, I identify the kinematic parameters and check how much of the error a better model removes. The fixed-price remote diagnosis tells you what accuracy is reachable with your robot.

Background: Precision control for robots and machine tools

Mechanical and electronics teams, but no control lead

What happens. A hardware start-up has mechanical designers and electronics engineers, and the control is written by whoever has time. Nobody owns the topic: the tuning is done by feel and models do not exist.

What it costs. Product delays, prototypes that behave worse than the design promised and investors who ask when it will work.

Why the usual fix fails. Hiring a lead takes months and costs a high salary before the product exists. A student or a generalist can write a controller, but not take the responsibility for a robust design.

How a fixed-price remote diagnosis checks it. From the prototype data, the mechanical design and the code in use, I check what control architecture suits the product and what is missing. The fixed-price remote diagnosis gives you a plan and a list of what an interim lead would do first.

Arms flex at full reach

What happens. A bigger robot with a long reach and a heavy payload is more flexible. At full stretch the arm bends and swings after each move, and the tip arrives later than the controller thinks.

What it costs. Lost precision and speed, because the robot must wait for the swinging to stop. The larger the reach, the larger the loss.

Why the usual fix fails. A stiffer structure costs weight, material and energy. Slowing the arm down is the easy way out and wastes the capacity of the drives.

How a fixed-price remote diagnosis checks it. From joint and motor data and a few test motions, I identify a flexible-body model and check how much damping and compensation the controller could add. The fixed-price remote diagnosis tells you what gain is realistic for your arm.

Background: Precision control for robots and machine tools

Robots follow paths with errors at speed

What happens. At low speed the tip follows the programmed path. When the speed goes up, corners are cut, circles turn into ovals and the path sags. Gluing, welding and cutting suffer.

What it costs. Seams in the wrong place, parts that need rework and speed limited by the quality of the path rather than the drives.

Why the usual fix fails. Reducing the speed is the cure that costs cycle time. Teaching the path corrected by hand works for one program and one robot.

How a fixed-price remote diagnosis checks it. From path records at several speeds (commanded and measured), I identify the tracking error and check how much a model-based trajectory tracking could remove. The fixed-price remote diagnosis tells you what path accuracy to expect at speed.

Background: Contour accuracy versus speed

Unknown payloads degrade the control

What happens. The controller assumes a payload. When the tool or workpiece weighs more or less, or the centre of gravity differs, the robot vibrates, overshoots or drifts. Users change tools and products often and do not enter the right values.

What it costs. Vibration and wear, trips on overload and accuracy that depends on whether the payload data are right.

Why the usual fix fails. Asking users to enter the payload is error-prone. Tuning for the heaviest payload makes the robot sluggish for the light ones.

How a fixed-price remote diagnosis checks it. From joint torques and motions during a short test movement, I check whether the payload can be estimated automatically and the control adapted. The fixed-price remote diagnosis tells you how well that works on your robot.

Background: Control that adapts to drift and wear

Long, light arms vibrate

What happens. Light arms are fast and cheap in energy, but they are springy. After each fast stop the arm keeps swinging, and the vibration shows in the tool.

What it costs. Cycle time lost while waiting for the vibration to die away, and wear on the gearboxes from the shaking.

Why the usual fix fails. Notch filters work for one posture and one payload. Mechanical damping adds weight, which defeats the lightweight design.

How a fixed-price remote diagnosis checks it. From vibration measurements in several postures, I identify a flexible-body model and check how well damping in the controller can work across the workspace. The fixed-price remote diagnosis tells you what to expect.

Background: Precision control for robots and machine tools

Robot motions are not time-optimal

What happens. Motions are programmed point to point with default speeds and accelerations, far below what the drives could do. The programmer chooses safe values because the real limits are not shown.

What it costs. Cycle time that is longer than necessary on every part, which means fewer parts per hour or one more robot than needed.

Why the usual fix fails. Raising speed by hand finds a limit by trial and error and rarely finds the best combination for the whole motion. Standard planners do not use the real torque limits.

How a fixed-price remote diagnosis checks it. From the robot model, drive limits and the motion programs, I check how much shorter the motions could be if they were planned within the true limits. The fixed-price remote diagnosis tells you what cycle time gain is realistic.

Background: Optimal paths and setpoint plans

Robots move inefficiently

What happens. A robot reaches the same point along many paths, with many speeds and postures. The programmed one is rarely the one that uses the least energy. Motors work against gravity and brake against springs.

What it costs. Power for motion that could be avoided, heat in motors and drives and a cell that needs more cooling and a bigger supply than necessary.

Why the usual fix fails. Operators optimise the cycle time, not the energy. Energy-saving modes in the controller shut the robot down when it waits, but do not change the motion.

How a fixed-price remote diagnosis checks it. From the robot’s power records and its model, I check how much energy a different path, posture or timing would save without changing the cycle. The fixed-price remote diagnosis tells you what to expect for your cell.

Background: Economic predictive control

Force control becomes unstable on stiff parts

What happens. When the robot presses, grinds or inserts, it regulates the contact force. On a soft part this is calm. On a stiff part, the same gain makes the contact bounce and the force oscillate.

What it costs. Damaged parts and tools, rejected work and slow, cautious contact speeds. Programs are written again for each new material.

Why the usual fix fails. Lowering the gain makes the control sluggish. A tuned value for one stiffness fails for the next part.

How a fixed-price remote diagnosis checks it. From force and position recordings during contact, I identify the contact stiffness and check how a model-aware control keeps the force steady across materials. The fixed-price remote diagnosis tells you what to expect.

Background: Precision control for robots and machine tools

Robots keep conservative speeds near people and obstacles

What happens. A robot that shares space with people or moving obstacles slows down far ahead of any danger, because it does not know what the others will do. It stays slow even when the space around it is clear.

What it costs. Throughput lost in every cycle and cells that cannot be used at the speed their drives allow.

Why the usual fix fails. A fixed safety distance is simple to certify and costs speed everywhere. A bigger safety zone makes the cell larger.

How a fixed-price remote diagnosis checks it. From the cell layout, the safety settings and logs of the motions, I check how a planner that takes the uncertainty about others into account could keep the safety and gain speed. The fixed-price remote diagnosis tells you what throughput gain can be expected.

Background: Obstacle avoidance under uncertainty

Control is tested only on hardware

What happens. New control software goes straight onto the robot. Each change is tried on the real arm, which is also needed by others. A mistake can mean a collision, a broken gearbox or a damaged tool.

What it costs. Damage to hardware, days of waiting for the robot and bugs found late, when they are expensive.

Why the usual fix fails. Testing in a simple simulation hides contact and friction effects, so the team does not trust it. Spare hardware for everyone is costly.

How a fixed-price remote diagnosis checks it. From the robot’s data sheets and a few measurements, I check how good a simulation model must be to catch most mistakes before the hardware, and what it takes to build. The fixed-price remote diagnosis tells you if simulation-first testing pays off.

Algorithms are too heavy for the robot controller

What happens. Better control methods exist on paper, but they are too slow for the robot’s controller within the cycle time. The team simplifies them until they fit.

What it costs. Control that is simpler than the robot could use, lost accuracy and speed, and a new hardware generation that is held back by the old software.

Why the usual fix fails. A faster computer adds cost and a new platform. Simplifying an algorithm by hand costs the benefits that made it attractive.

How a fixed-price remote diagnosis checks it. From the controller specification, cycle times and the algorithm, I check how a reduced form can run in real time and what it loses. The fixed-price remote diagnosis tells you what fits on your hardware.

Background: Predictive control on embedded hardware

Gearbox and joint wear goes unnoticed until the robot stops

What happens. Gearboxes and joints wear slowly. Backlash grows, friction changes and noise appears. The robot goes on working until a joint fails, often in a running production line.

What it costs. Unplanned downtime, an emergency repair and parts that are not in stock. The cell has to be stopped at a bad time.

Why the usual fix fails. Replacing gearboxes by hours of operation is too early for most and too late for some. Listening to the noise depends on the person who walks by.

How a fixed-price remote diagnosis checks it. From motor currents, positions and temperatures that the controller already records, I check whether changes in friction and backlash can be seen and a replacement scheduled. The fixed-price remote diagnosis tells you how early the signals show it.

Background: Computed measurements

Drive concepts and motor sizing for lab-automation robots rest on rules of thumb

What happens. Lab-automation robots move small loads very fast and often, and must be quiet and precise. The motor, gearbox and controller are chosen from a rule of thumb and then tuned on the bench.

What it costs. Drives that are too large or too small, more heat than planned and tuning cycles that delay the product.

Why the usual fix fails. Catalogue values are for typical duty, not for this motion profile. Trying several motor sizes costs prototypes.

How a fixed-price remote diagnosis checks it. From the planned motion profiles, loads and motor data, I check by model which motor and gearbox satisfy the profile with margin, and how the control should be set. The fixed-price remote diagnosis tells you whether a smaller drive would do.

Background: Predictive drive control

Advanced robot control, calibration and signal processing are hard to hire

What happens. A robot maker needs better controllers, calibration routines and signal processing for its sensors. The people who can do all three are few, and the roles stay open.

What it costs. Features wait on the roadmap, the same people carry every project and the quality of the robot stays behind what the hardware could do.

Why the usual fix fails. A general software developer does not have the background. Training takes longer than a product cycle.

How a fixed-price remote diagnosis checks it. From the control code, calibration procedures and sensor data, I check where identification and model-based control would give most, and which part can be done outside the team. The fixed-price remote diagnosis gives you a plan in parts.

Background: Precision control for robots and machine tools

Robot models for simulation need physical parameters that nobody has tuned

What happens. A simulation of a robot is only as good as its masses, frictions and stiffnesses. These come from drawings and defaults, and the simulated robot behaves differently from the real one.

What it costs. Results from simulation that are not trusted, learning methods trained on a model that does not match and a gap that appears when the code moves to the real robot.

Why the usual fix fails. Adjusting parameters by hand until the plots look similar fits one motion. A model from the drawings gives masses but no friction.

How a fixed-price remote diagnosis checks it. From motion recordings of the real robot, I identify the parameters that make the model reproduce them and check how far the model can be trusted. The fixed-price remote diagnosis tells you whether your recordings are enough.

Background: Computing unmeasured states for control