Robot Force Torque Sensor Mounts Machining

• 1
  Key takeaways
• 2
  Where the Sensor Stops and the Mount Begins
• 3
  Stiffness: The First Number to Get Right
• 4
  Flatness, Parallelism and Bolt Preload
• 5
  Why One Setup Decides the Tolerance Stack
• 6
  Material and Finish Choices That Age Well
• 7
  What to Put on the Drawing
• 8
  Material Trade-offs for Sensor Mounts

A six-axis force torque sensor outputs Fx, Fy, Fz, Mx, My and Mz. All six numbers come from strain gauges bonded to a machined flexure inside the sensor housing. That flexure deflects a few micrometers at full load. Anything outside the housing that also deflects gets added to the reading, and the sensor has no way to separate the two.

The mount is that outside part. It sits between the robot flange and the end effector, usually as a single plate or a short spacer block. From the sensor's point of view the mount is a spring in series with its own flexure. Series springs add their compliance. If the mount deflects 10 μm under a load that moves the flexure 40 μm, roughly a fifth of the reported signal never came from the sensor.

This is why robot force torque sensor mounts machining matters more than it first appears. The sensor may be accurate to 0.5 % of full scale on a calibration bench. Bolted to a soft or poorly seated mount, the same unit can drift several percent once the arm starts moving. The bench number is real. The installed number is what the control loop sees.

A useful first estimate: treat the mount as a cantilever and compare its tip deflection to the sensor's rated deflection at the same load. If the mount moves more than about 10 % of the sensor's own travel, it will show up in the data.

Stiffness is force divided by deflection, and it is geometry before it is material. A flat plate loaded out of plane is a poor shape. A plate with ribs, a boss around the bolt circle, or a closed box section can be five to twenty times stiffer at nearly the same mass.

Thickness helps, but not linearly. Deflection in bending scales with the cube of thickness, so going from 8 mm to 12 mm raises stiffness by roughly 3.4×. Adding a 6 mm rib can beat that while saving weight. For a wrist-mounted sensor on a collaborative arm, mass at the wrist costs payload, so ribs usually win over a thicker plate.

Aluminum 6061-T6, 7075-T6 and Ti-6Al-4V cover most robot sensor mounts. Aluminum has an elastic modulus near 69 GPa, titanium near 114 GPa, steel near 200 GPa. Stiffness scales with that number directly, so a titanium part is about 1.7× stiffer than the same geometry in aluminum, at 1.6× the density. Titanium helps when the geometry is already fixed by bolt patterns and you cannot add ribs.

Stainless 17-4PH sits between titanium and steel and machines cleanly in the H1025 condition. For food, medical or washdown cells it is often the practical choice, since it tolerates cleaning agents that would strip an anodized layer.

Two bolted faces do not touch everywhere. They touch at the high spots, and the real contact area can be a small fraction of the nominal face. Under load those high spots flatten, the joint rotates slightly, and the sensor sees a moment that is not in the real world.

Flatness and parallelism control how much of that happens. A practical target for a sensor interface face is 0.02 mm flatness over the full face and 0.02 mm parallelism between the flange side and the effector side. Tighten to 0.01 mm when the sensor is small, when the arm is fast, or when the measurement band goes below 10 N.

Surface roughness matters less than most people assume, but it is not irrelevant. A ground face at Ra 0.8–1.6 μm gives enough asperity contact for stable friction. A mirror-polished face can actually be worse in a bolted joint, because too little texture means less mechanical interlock and more reliance on friction alone.

Bolt preload is the other half. A joint holds its stiffness only while the bolts stay stretched. Use the bolt circle the sensor maker specifies, use class 12.9 or A2-70 hardware as appropriate, and torque in a star pattern in two or three passes. Thread lubrication changes achieved preload by 20–30 % at the same torque, so specify whether threads are dry or lubricated.

A sensor mount has datums on both sides: the flange face that bolts to the robot, and the effector face that carries the tool. If those two faces are machined in separate operations, the shop has to stack the two setups through a vise or fixture. Each re-clamp adds error, and the errors add up in the direction that hurts most.

Machining both faces in one setup removes that stack. On a simultaneous 5-axis machine with a trunnion or a Ø400 mm rotary table, the part can be turned over within the same program using probed datums, so the two faces share one coordinate system. This is the main reason five-axis work is common on these parts even when the geometry looks simple.

Bolt hole position deserves the same treatment. Holes should be bored or interpolated to size rather than drilled, and located from the same datum as the faces. Hole-to-hole position within 0.02 mm keeps the sensor body from being pulled sideways as the bolts tighten. A sensor forced off center reads a built-in offset that no calibration will remove.

Dowels or a precision pilot diameter are worth adding when the mount will be serviced in the field. They let a technician refit the sensor to the same position without a dial indicator, which matters on a production cell where every hour of downtime is measured.

Aluminum 6061-T6 is the default for most robot force torque sensor mounts machining projects. It machines fast, holds ±0.005 mm without drama, and hard anodizing (Type III) gives a 25–50 μm layer with decent wear resistance on the faces. The catch is that anodizing builds thickness, and on a sealing or locating surface it can shift the fit. Mask those surfaces or specify the coating on non-critical faces only.

Hardcoat on a full face also changes flatness slightly. For a sensor interface, we usually suggest clear or black anodize Type II with masking on the mating faces, and hardcoat only on the outer profile. If the user wants a conductive path for grounding, conductive anodizing or a masked bare patch is the way to keep continuity.

Titanium TC4 (Ti-6Al-4V) suits high-load or high-cycle wrists. It is stiffer per unit volume and has good fatigue behavior, but it is also the material that punishes a bad process plan. Thin walls deflect under cutting force, and the low thermal conductivity keeps heat in the cut. Light radial passes, sharp tooling and generous coolant keep dimensions in tolerance.

Stainless 17-4PH covers hygienic and corrosive environments. Electroless nickel on aluminum is another option when the customer wants a hard, uniform surface without the dimensional shift of anodizing. In all three cases, the finish is specified after the geometry, never before, because coating thickness and edge break requirements drive the drawing.

A drawing that a shop can hit on the first run names four things: the datum scheme, the flatness and parallelism limits on the two interface faces, the bolt hole position tolerance, and the surface finish on each face. If any of those is missing, the shop will pick a number, and the number may not match what the sensor needs.

Spell out the datum reference frame. Pick the flange face as datum A, the pilot diameter or a bored hole as datum B, and one hole as datum C. Then call out perpendicularity of the effector face to A within 0.02 mm and position of the bolt pattern to A|B|C within 0.02 mm. This is more useful than a general ±0.05 mm block tolerance, which says nothing about how the faces relate to each other.

Add a note about edge break. A sharp corner on a sensor face can raise a burr that holds the two faces apart by 30 μm. A 0.3 mm maximum edge break on all interface edges costs nothing and prevents a classic assembly problem.

If the part will be anodized, say which faces get masked. If it will be used in a cleanroom or a medical cell, say so, because that changes cleaning and packaging. Finally, ask for an inspection report with the flatness and parallelism values. A CMM report on two faces takes minutes and settles any argument at incoming inspection.