Bot will answer all plasma cutting common questions. It knows all cutting parameters and what consumables to use for cutting. like if I want to cut 1 inch mild steel with 300 amps using XPR 300, it knows all parameters for this process like speed, kerf etc.. and what consumables to use. If also knows about cut quality. If also knows all plasma error codes, it’s cause and resolutions. Also I have service tickets (like problems and solutions) I wan to feed that to bot and bot answer user questions using that service tickets as resource.
For amusement, let’s see how smart GPT-5 is if I throw all that text right into the API “developer” message to make a specialist AI application, bad spelling and unrefined prompting included.
Full response to my (AI-generated) question
Below is a solid “book” starting point for 1.00-inch (25 mm) A36 on a water table with Thermal Dynamics/ESAB Ultra-Cut XT 300, XT mechanized torch, O2 plasma / Air shield, 300 A, and iHC THC. I’ve also included the usual quality/THC tweaks to eliminate the top dross, correct the bevel, and stop the AVC/arc-voltage alarms.
Book setup (baseline)
- Process: O2 plasma / Air shield, 300 A
- Gas supply to console (at inlets):
- Oxygen: 115–130 psi (8–9 bar), 99.5% purity minimum
- Air (shield): 115–130 psi (8–9 bar), -40 F/C dew point (or better), oil-free
- Let the UC XT gas console manage flows automatically per “O2 MS 300A” process
- Torch height and timing:
- Pierce height: 0.250 in (6.4 mm)
- Pierce delay: 1.5–1.7 s (start at 1.6 s; extend slightly if splash/water blowback)
- Cut height: 0.090 in (2.3 mm)
- THC voltage: 198 V starting point
- Voltage sampling: delay THC on until: a) arc is steady, b) feedrate at programmed speed; sample after 0.75–1.0 in (20–25 mm) of straight cut, then enable AVC
- Anti-dive: disable AVC below 65–75% of programmed feedrate (set the threshold to match your controller capability)
- Cut speed:
- Programmed/book speed: 60–70 ipm (1500–1800 mm/min); start at 65 ipm (1650 mm/min)
- Kerf width for CAM:
- 0.095–0.110 in (2.4–2.8 mm); start comp at 0.100 in (2.5 mm)
- Minimum quality hole diameter:
- About 1.5× material thickness (≈1.5 in / 38 mm). For simple through-holes that only need to be drillable, 1.2–1.3× thickness can work with a slow hole routine, but edge quality will fall off.
- Lead-in/lead-out and motion:
- External profiles: use CCW (counterclockwise) to put the “good side” on the part edge for most XT torch/swirl setups
- Lead-in: 0.6 in (15 mm) arc/curved lead-in, 90° preferred
- Lead-out: short straight 0.2 in (5 mm) or overburn, then extinguish off-part if possible
- Holes: small radial lead-in 0.20–0.25 in (5–6 mm), 60–90° overburn, exit tangent
- Corner speed lockout: limit CNC to ~60–70% of programmed feed in tight corners; freeze THC/AVC during corners, small features, lead-in/lead-out
Consumable stack (XT mechanized torch, 300 A O2/Air)
- Use the 300 A Oxygen “mild steel” set for the XT torch:
- Electrode (O2, 300 A)
- Water tube (300 A)
- Swirl ring (O2 type, 300 A)
- Nozzle/Tip (O2, 300 A)
- Shield (Air, 300 A)
- Retaining cap (XT mechanized)
- Notes:
- Part numbers vary by torch revision/region and kit type (standard vs XtremeLife). If you send the torch model tag and gas console model/revision, I’ll return the exact ESAB/TD PNs for each piece or the correct kit PN.
- Verify the swirl ring is the O2 version and installed in the keyed orientation.
Addressing the reported cut quality (light top dross, slight positive bevel on CW cuts)
- Cut height/arc voltage:
- Top dross and mild positive bevel commonly indicate the torch is a bit high. Reduce arc voltage 3–6 V to bring the torch down ~0.015–0.035 in (rule of thumb ≈ 1 V changes height ~0.004–0.006 in, 0.10–0.15 mm). Re-test.
- Speed:
- If dross persists, increase speed 5–10% and watch edge angle. If the positive bevel worsens as speed increases, revert speed and lower height a touch more (another 2–3 V).
- Water table level:
- Keep water 0.5–1.0 in (12–25 mm) below the plate bottom during cutting to avoid quench/top spatter and voltage instability. Avoid water contacting the shield.
- Gas quality:
- Confirm O2 purity ≥99.5%, no restrictions, and shield air is very dry (desiccant/refrigerated dryer OK). Wet air on thick O2 cuts promotes top dross.
- Motion strategy:
- Use CCW on outer profiles and CW on internal features to put the “good side” on the part. This directly addresses the sidewall angle you’re seeing on CW exterior cuts.
- Torch/perpendicularity:
- Re-square the torch to plate. Even 0.5–1° out-of-square shows up as consistent bevel.
- Consumable condition:
- Replace electrode when pit depth reaches ~0.040 in (1 mm).
- Check nozzle orifice for ovality or nicks; replace if in doubt.
- Verify the correct O2 swirl ring and that O-rings are intact and lubricated per ESAB spec.
Stopping the intermittent “arc voltage lost / AVC fault” mid-cut
On water tables, most mid-cut AVC faults trace to noise/grounding, water contact, or THC settings. Work through this list:
- THC enable/anti-dive:
- Delay THC enable until the cut is stable (see sampling above).
- Freeze THC below 65–75% of programmed speed (anti-dive). Also freeze on lead-in/lead-out, tight arcs, and holes.
- Work return and grounding:
- Clamp the work lead directly to the plate (not the slats or table frame). Use a second work lead for large parts.
- Bond the table and work return per ESAB recommendations; avoid shared grounds with servo drives/CNC where possible.
- Water and ohmic:
- Prevent water from bridging the shield to plate; if you use ohmic sensing, ensure it’s disabled once cutting starts and that wet slag isn’t causing false contact.
- Voltage signal path:
- Verify the iHC is using the correct divided voltage ratio and matches the power supply divider setting.
- Inspect voltage sense wiring for loose connectors/intermittents; check shielding and routing away from motor drives.
- Confirm the THC voltage calibration (some iHC units allow a quick check with a known divider).
- Set a modest voltage filter (e.g., 20–30 ms) to smooth noise without making the torch sluggish.
- Motion/feeds:
- Avoid decelerations that drop speed far below your anti-dive threshold. When necessary, program feed holds with THC frozen.
- Gas and cooling faults:
- Momentary O2 pressure dips or coolant flow alarms can show up as AVC faults. Check supply regulators/hoses and coolant level/flow.
Quick “try now” adjustments for your job
- Program 65 ipm, 0.25 in pierce height, 1.6 s delay, 0.090 in cut height.
- THC: set 198 V to start; enable after 1.0 in of straight cut; anti-dive at 70% feed; freeze on leads/corners/holes.
- Water level just below plate; verify dry air and O2 purity.
- Motion: CCW for external contours; 0.6 in arc lead-in; 0.2 in straight lead-out; corner lock at ~60–70%.
- If top dross shows: lower voltage 3–5 V and/or increase speed 5–10%. If bevel persists on CW cuts, keep the direction change and check torch squareness and swirl ring.
If you can share the torch model tag (XT torch variant), gas console model, and your consumable packaging, I’ll reply with the exact ESAB/TD part numbers for the full 300 A O2 stack (or the correct prepack kit) so you can stage the first run with the precise parts.
The basic issue - the AI might look smart, but it DOESN"T have a book. It will fabricate answers.
You’ll need to augment the built-in knowledge with a data retrieval layer, giving the AI some kind of tool where it can search extensive documents, and it also must be prompted to not answer if it doesn’t retrieve factual answers from “the book”.
Building such a product on OpenAI’s “Responses” endpoint, paying for each request to an AI model, you also have a ready-built basic document lookup tool that can be employed to search for knowledge that you provide. Of course, you must start with the “quickstart”, setting up an account organization and funding it – and the API is for programmers, not a ready-built product.
(note - gpt-5 was absolutely disobedient in making an original question, unable to do anything but repeat back it’s own developer message as a field application)