Five-Axis CNC Shot Peening Machine for Turbine Blades

What does this machine do to a turbine blade?

Motionwell Automation designed and built a five-axis CNC shot peening machine for an aerospace turbine blade MRO services provider. The machine is a sealed blast chamber of roughly 2.5 m by 2.0 m by 2.5 m carrying a nozzle on X, Y and Z linear stages with a B-axis swing of +/-90 degrees, a C-axis rotary table turning the blade through 360 degrees continuously, an aluminium quick-change fixture with quick-locking clamps on that table, and a collaborative arm that loads blades through a window from comb-shaped racks holding 20 to 30 blades per side. Motion runs as a G-code program on Delta motion control, with an OMRON HMI at the operator station.

What the blade leaves with is a compressive stress layer, and on this machine that layer raises turbine blade fatigue life by 30 to 50%. The matte texture on the surface is a by-product of the treatment rather than the reason for it.

That one fact shapes every station on the machine. A blade peened at the wrong pressure, or with a clamp standing between the nozzle and a fillet, comes out of the chamber looking like a blade peened correctly and goes back into an engine that way. So the machine was not built to produce a finish. It was built to hold a defined set of inputs, to reach every treated surface at a controlled angle, and to leave a record that the run happened the way the process specification says it should have.

The rest of this page walks those stations in the order a blade meets them, starting with the axis that decides the result.

Why does the nozzle have to be presented to the blade rather than aimed at it?

Three axes reach every point on an aerofoil. What three axes cannot do is arrive at each of those points at the same angle.

Media leaves the nozzle as a directed stream, and what it does on landing depends on the angle at which it lands. A blade curves in two directions at once, so a nozzle set square to the surface at one point along the chord is at a glancing angle a short distance further along. Reaching a surface and presenting to a surface are separate requirements, and the second one is what produces an even result across a curved part. This is why the axis count was settled before any structure was drawn, in the way our mechanical design and simulation work treats kinematics generally: a machine cannot be given another degree of freedom afterwards without becoming a different machine.

Read the last two rows together and the machine’s logic shows. X, Y and Z decide where the nozzle is and how far off the surface it stands. B and C decide what the surface looks like from the nozzle’s point of view, which is the part of the geometry that governs the outcome. The C axis also earns its keep on handling: a blade turned under the nozzle does not have to be unclamped, reversed and re-located halfway through a cycle.

The path is a G-code program rather than a taught trajectory. That was a deliberate choice on an audited process. A program can be read, compared against its previous revision, archived beside the part family and reissued to a second machine. A taught path lives inside a controller and inside somebody’s memory.

Which process parameters are recorded, and which are the machine’s limits?

The machine carries capability windows. The process carries numbers, and those numbers live inside the windows. On this build the windows came out as follows, and all of them are set per part family at commissioning rather than fixed as machine limits.

Two rows deserve more than a table cell. Media is often treated as a consumable and it behaves as a process variable: steel shot, ceramic shot and glass bead at the same pressure do not produce the same result, and the 0.2 to 1.0 mm band spans materially different processes at its two ends. Whoever refills the hopper is not the person who gets to choose.

Cycle time is the row that inverts the usual instinct, because on a machine of this kind it is not a number to attack. It falls out of the coverage specification: if holding 100 to 200% coverage needs the nozzle over the surface for that long, that is the cycle, and shortening it by any route other than a re-qualified process is a quality decision wearing a productivity costume.

The sequence behind that table matters as much as the values in it. The parameter set was proved on Almen strips at commissioning and only then released to production, and proving it changed the status of every number: from a setting an operator could turn into a value the program holds and the machine records against the blade.

How is a blade held without spoiling the root?

Two requirements pull against each other. The blade has to be held rigidly enough that it does not shift under the stream, because a part that moves changes both standoff and impact angle and therefore changes the process. At the same time every surface that has to be treated must stay exposed to a nozzle trying to reach it from several attitudes.

The root is where those two meet. It is the obvious place to clamp and it is also the load-bearing interface into the disc, which makes it the surface least tolerant of a witness mark. Inside the chamber the blade sits in a grid rack built to protect the root, and the process fixture on the C-axis rotary table is an aluminium quick-change unit closed by quick-locking clamps. Quick-change matters here for a reason beyond changeover time: the fixture is a per-family item, so the machine was designed on the assumption that fixtures would be swapped rather than adjusted.

Shadowing is the failure mode that looks like a cycle time problem and is not. Anything the fixture puts between the nozzle and the surface leaves a local coverage hole, and because the obstruction turns with the part on the C axis, more time in the chamber does not fill it. On a five-axis machine the fixture therefore becomes part of the path problem: the program has to arrive at each surface from an angle where no clamp is standing in the way. Fixture geometry and program are developed together, not in sequence.

How does the collaborative arm load the cell?

The loading duty is one blade in and one blade out against a cycle measured in minutes, from racks a person refills between batches. That duty was matched with a collaborative arm working through a window in the chamber wall, with the application tooling designed in house for a component that is aerodynamic, sometimes coated and unforgiving about where it is touched. The gripping problem is the one described on our end of arm tooling page, applied to a part that has no convenient flat.

The cycle runs in one loop:

1. The arm takes an untreated blade from the input rack.
2. It seats the blade in the fixture on the rotary table.
3. The chamber door closes and locks, and the G-code program runs the five axes over the blade.
4. The air wash clears media off the blade before the door is released.
5. The arm lifts the treated blade out and places it in the output rack.
6. The next blade starts, and the cell continues until the input rack empties or the output rack fills.

The racks carry a second job that is easy to overlook. Untreated and treated blades do not share a rack, with an intermediate rack available where parts have to flow in both directions, so the state of a blade is a physical fact about where it stands rather than a note somebody has to keep current. On a process whose result cannot be seen on the part, that is worth more than the handling convenience, and it is the reason the racks are shaped rather than generic: the comb profile follows blade geometry so stored blades do not touch each other.

Where this loading duty differs from ordinary machine tending automation is the interlock discipline. The arm must not reach through the window while the nozzle is live or the door is unlocked, so loading and containment run as one sequence rather than two, with door lock and machine lighting handled as machine functions inside the same program as the motion. The safety case follows the usual route for a guarded cell with an arm in it: a risk assessment sets the scope, ISO 10218-1 covers the robot and ISO 10218-2 the integrated system. The current third edition of Part 1 was published in February 2025 and is the first substantive revision since 2011, adding robot classifications with matching functional safety requirements, safety-related cybersecurity requirements and end-effector guidance, with most of the collaborative-operation material from ISO/TS 15066:2016 moved into Part 2. A collaborative arm standing beside a blast chamber does not make the chamber a collaborative space, and the guarding on this cell was scoped on that basis alongside our other machine safety and compliance work.

How is the media kept inside the machine?

Three streams leave the nozzle and each of them has somewhere it has to go.

Spent media falls to a collection trough in the chamber floor and returns through a recovery loop. Recovery is where a peening machine differs from a cleaning cabinet, because the loop is returning media that has already been fired at a hardened surface. Media that has broken down or changed size is no longer the media the process was qualified on, so how the loop screens what it returns, and what confirms it is doing so, is a specification question rather than an installation detail. It is also why media size is re-confirmed after recovery rather than only at delivery.

Airborne dust and fines leave through a dedicated extraction enclosure with a filter cartridge, standing as its own unit beside the machine along with the control cabinet. Siting abrasive dust collection away from the machine rather than sealing it harder is the same reasoning we apply to control panel design and build in dusty rooms.

The third stream is media riding out on the finished blade, which the air wash removes before the door opens. Without it, media leaves with the part, into the output rack and into the next operation.

The energy in the stream has to be absorbed as well, and it is, by wear-resistant liners on the chamber walls. Those liners are a maintenance item with a service life, and treating them as one keeps the wear from arriving later as an unplanned shutdown.

What makes containment structural rather than an accessory on this machine is that the part program switches it. An operator cannot forget to turn on extraction, because extraction is an M-code in the same program as the motion.

What record does the process leave?

The record is a deliverable of the machine rather than a report about it, for the reason set out at the top of this page: nobody can inspect a finished blade for the property that was bought.

Three things make it up. Process parameters are logged through the Delta motion control and OMRON HMI on every run, so each blade carries its own processing record. The operator interface shows live X, Y, Z, B and C positions beside the running G-code program list and the M-code function table, so the state of the machine is something an operator reads rather than infers, which is the design principle behind our HMI and SCADA integration work. And the machine shipped with a documentation package covering maintenance procedures, calibration schedules and operator training materials, assembled to survive a customer audit rather than a handover meeting.

One item belongs in that record that buyers rarely ask for. Because the path is a program, the program identity and its revision are part of the evidence alongside the parameters. A pressure captured against a part number says what the air was doing; the program revision says where the nozzle went. Both are worth holding under the same change control.

Constraints and trade-offs

It is a nozzle machine, not a wheel machine. The whole design premise is a controlled nozzle presented to a specific surface under a program. A centrifugal wheel machine throwing media at bulk parts is a different mechanism answering a different question, and nothing on this page transfers to it.

Unattended means rack-limited, not shift-limited. Rack capacity of 20 to 30 blades per side with dual-side access is what sets the length of a run without an operator, so the useful planning figure is blades per rack against blades per shift rather than seconds per blade. Somebody still walks over to refill.

The floor area is larger than the machine envelope. The chamber occupies roughly 2.5 m by 2.0 m by 2.5 m, and the extraction enclosure, the control cabinet and the input and output racks all stand outside it. Extraction routing is a building service the site has to accept somewhere.

The control stack spans two vendors. An OMRON operator panel over Delta motion control is a sound machine and it is also two engineering tools, two spares chains and two support routes for whoever owns it in year six. That is a decision to take deliberately at platform selection rather than during a fault, as we set out on our PLC migration and upgrade page.

Related work

The wider scope it belongs to, including force-controlled robotic grinding for CFRP composite panels and the nacelle paint removal work developed with A*STAR SIMTech, is on our aerospace and precision engineering page. The machine concept and how a peening cell gets specified are covered on our aerospace surface treatment automation page. Machines like this one are designed, assembled and tested at our Woodlands Link facility, where the company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3, and how a build of this shape is scoped from concept to commissioning is in our guide to special purpose machine design.

To scope a peening cell, send us the blade drawing with the treated and excluded surfaces marked, along with your intensity and coverage specification and how many part families share the process.

Frequently Asked Questions

How is peening coverage verified on this machine?

Against Almen strip intensity measurement, not by looking at the blade. A standard strip is exposed under the conditions the part sees, and the arc it takes reads as the intensity the process delivered. Coverage of 100-200% was verified against strip results during commissioning, and those results set the parameters that then went into production. A camera cannot substitute, because the property under control is a residual stress condition inside the metal rather than a surface appearance. Where the strip sits matters as much as what it reads: a strip proves what the process delivered at that position, so strip placement belongs in the verification design rather than in the machine manual.

What happens when a new blade geometry arrives?

Three pieces of engineering, and none of them is a setting change. The blade needs a fixture that holds it rigidly without marking the root, a G-code path that presents the nozzle at a controlled angle across the new aerofoil, and a proving run against Almen strips before production parts go through. The comb-shaped storage racks are shaped to blade geometry as well, so a family with a different profile can need its own rack. The consequence for planning is that the unit of cost on a machine like this is the part family rather than the part, and a shop carrying many families pays that engineering repeatedly.

Why is a collaborative arm fast enough to load a peening cell?

Because the duty is one blade in and one blade out against a cycle of 2-5 minutes, and nothing in that asks for speed. The arm is idle for most of every cycle by design, and that idleness is what makes the cell unattended rather than what makes it slow: it runs until the input rack empties or the output rack fills. What the duty does ask for is an arm that can share floor space with the operator refilling racks between batches, that can be re-taught when a new blade family arrives, and that is held out of the chamber by the same interlocks that hold the door.