This robot-tended lathe cell turns porous bronze and cast iron castings, and insert life swings between roughly 5 and 20 parts. The cell needs a stop signal that fires before the next tool enters the cut, not only a force reading. Start with load monitoring derived from the spindle and axis drives. Many CNC controls include it natively, and systems such as Marposs, Caron Engineering TMAC, and 3PM Technologies Machine Boss can add it. Back it up with a tool-probe check for gross breakage and a tool-life counter as a floor. Treat a bolt-on force sensor as the last option to evaluate.
Failure chain from casting porosity to multiple broken tools
A pore, crack, or hard inclusion turns a continuous cut into an interrupted one. The insert takes an impact load, and the carbide edge chips or fractures. The damage does not stop at one tool. A broken roughing insert leaves stock uncut. The next tool in the toolpath, such as a finishing, grooving, or threading tool, is programmed for a finish allowance. It meets roughing stock instead and breaks too. This chain is why a single fracture during an unattended run wrecks the rest of the turret.
Tool monitoring means comparing a real-time load signal against limits taught for each tool and operation. Fracture produces two signatures:
- A spike at the moment of breakage. An upper limit catches it.
- A step down afterward. With the edge gone, no chip forms and the load falls toward the air-cut level. A lower limit, or missing-tool limit, catches it.
A monitor that watches only for load increases misses the second signature. That signature is often the more reliable of the two.
Check 1: insert life spread from baseline data
Reading to take: parts produced per insert for each tool across a representative run. Record the minimum, maximum, and average. Also log which tool and which operation failed each time.
- Tight spread: set the control's tool-life counter (parts or cutting minutes) below the observed minimum. Stop the cell for an operator inspection when the counter expires. This branch ends here.
- Wide spread, as here (5 to 20 parts): a counter set to the minimum discards inserts that could run up to four times longer. A counter set any higher lets random fractures through. A counter alone cannot protect this cell. Go to Check 2.
The failure log also narrows the scope. Usually one or two roughing tools account for most fractures, and those are the tools to monitor first.
Check 2: controller load-monitoring capability
Reading to take: the control model, its installed options, and whether spindle load and axis load are available as monitored signals with a configurable reaction. A load meter shown on screen does not qualify.
| Outcome | Meaning | Next step |
|---|---|---|
| Native tool load monitoring with alarm reaction | The control can compare spindle and axis load against per-tool limits. | Check 3, then configure natively. |
| Native monitoring plus adaptive control | On Sinumerik 840D installations, spindle and axis load have been used to adjust feedrate override automatically and to call a tool change when forces rise or fall excessively. | Check 3. Use adaptive feed for hard spots and limits for breakage. |
| Load display only, no reaction | Not usable for unattended running. | Evaluate a third-party system (Marposs, Caron Engineering TMAC, 3PM Technologies Machine Boss) that interfaces with the control. |
Check 3: load signal margin per tool
Reading to take: a trend of spindle load and X/Z axis load for each monitored tool. Capture it once as an air cut with no stock and once on a known-good casting. Compare the cutting-load delta with the scatter in the air-cut trace.
Spindle load reflects tangential cutting force times workpiece radius, expressed as a fraction of spindle motor capacity. A large spindle taking a light pass in free-cutting bronze moves only a small part of its range. Spindle acceleration and constant-surface-speed changes can bury that signal. Axis load reflects feed and radial force on the X and Z servos, and it often discriminates better for small tools and light cuts.
| Method | Catches | Misses | When it reacts |
|---|---|---|---|
| Tool-life counter | Predictable wear | Random fracture | At count expiry |
| Spindle load limits | Fracture on heavy roughing, missing tool | Light cuts where the delta sits inside the noise | During the cut |
| Axis load limits | Feed-force change, missing tool | Axes dominated by friction or gravity load | During the cut |
| Adaptive feed override | Load rise at hard spots (slows the feed) | Fracture that has already occurred | During the cut |
| Tool probe after the operation | Large fracture, missing insert | Wear, small edge chips | Before the next tool call |
| Dedicated force sensor | Small force changes | Depends on mounting location | During the cut |
If the cutting delta clearly separates from air-cut scatter, set limits on that signal and go to Check 4. If the margin is marginal, monitor axis load instead. Otherwise, add a dedicated sensor from a monitoring supplier or rely on the probe branch.
Check 4: breakage geometry and probe-after-cut
Reading to take: inspect broken inserts from the baseline run. Did the edge lose a large piece, or did it only chip?
- Large pieces: after each roughing operation, touch the tool on the tool setter. Compare the result with the stored offset, and alarm when the deviation exceeds a tolerance set just above normal wear. This catches fracture before the next tool runs into full stock. The cost is added cycle time per part.
- Small chips only: the probe will not see them. Load limits must carry detection.
The probe also misses gradual wear. It supplements load monitoring and does not replace it.
Stop reaction and robot interlock
Do not wire a load monitor to the emergency stop. An E-stop removes drive enable and halts the spindle and axes without a retract. That can leave the broken tool buried in the part and forces a recovery that needs an operator. Configure a controlled reaction instead: feed hold, retract, spindle stop, NC alarm, and cycle stop.
The robot must read the machine's alarm and cycle-complete status. It must not load the next blank while the lathe is in alarm. Otherwise the cell faults cleanly on one part and then loads another one.
Monitoring limits the damage from bad castings but does not eliminate it. Every trip still needs someone to replace an insert. Compare that labor against the cost of better-quality castings.
Configuration procedure and verification for the load-monitoring branch
- Select the tools to monitor from the failure log in Check 1.
- For each tool and operation, define the monitoring window. Open it after the spindle reaches speed and the tool engages. Close it before the retract. Exclude rapids, tool changes, and spindle acceleration.
- Run several known-good castings and capture the load envelope for each window.
- Set limits for each window:
- An upper breakage limit above the highest good-part peak.
- A lower missing-tool limit above the air-cut level.
- A wear or trend limit, if the system supports one.
- Map the reactions. Breakage and missing-tool events trigger feed hold, retract, and alarm. A wear event lets the current part finish, then changes to a sister tool or stops the cycle.
- Wire machine alarm and cycle status into the robot program as a load interlock.
- Re-learn the envelopes after any change to speed, feed, depth, insert grade, or casting supplier.
Verification checks, each with its expected reading:
- Known-good casting, all tools: no alarm, and every trace stays inside its band.
- Air cut with no stock, monitoring active: a missing-tool alarm in the first monitored window of each tool.
- Pre-chipped insert on a scrap casting at reduced depth and feed: an alarm and retract before the next tool call.
- Probe branch, with an insert offset or removed: an offset-deviation alarm at the post-operation touch-off.
- Lathe held in alarm while the robot program runs: the robot waits at the load handshake and does not pick or load a new blank.
FAQ
How do I detect a broken insert on a CNC lathe without an operator?
Monitor spindle or X/Z axis load against per-tool limits. Use an upper limit for the fracture spike and a lower limit for the missing-tool drop, with a feed-hold and retract reaction. Add a tool-setter touch-off after roughing operations to catch large fractures before the next tool cuts.
How do I set tool load limits for porous cast iron and bronze?
Learn the load envelope from several known-good castings for each tool and operation. Set the breakage limit above the highest good-part peak and the missing-tool limit above the air-cut level. Exclude rapids and spindle acceleration from the window, and re-learn after any change to cutting data or casting source.
How do I stop a robot-tended lathe when tool load monitoring trips?
Configure the monitor to issue feed hold, retract, and an NC alarm rather than an emergency stop. Interlock the robot on the machine's alarm and cycle-complete status. Confirm the robot waits at its load handshake and does not load a new blank while the lathe is in alarm.