Getting the Most out of the AI Assistant

The AI Assistant is most useful when you treat it as an analyst who can drive the software for you: it can explore your data, run native analyses with the exact options you ask for, read specific values off the results, compare and batch files, check limits, set up recording channels, and assemble reports — all from plain-language requests. This page shows how to phrase requests and gives a library of ready-to-use prompts, grouped by what you want to do.

The golden rule: every number the assistant states comes from the software's own engines. It will not estimate or invent metrics, peaks, or trends. If a request produces no result, it tells you why instead of making something up. You can trust the figures it quotes.

How to write a good prompt

Name the file and channel when you know them (“channel 1 of the first file”). If you are not sure, just ask it to list them first.
State the options you care about — weighting, window, spectrum size, RPM step, loudness standard, time range. The assistant looks up the exact parameter and confirms it applied.
Say what you want to see — “show it on a chart”, “give me a table”, “what are the top peaks”. One clear goal per message works best.
Iterate. Follow-ups keep the context: “now A-weighted”, “same but 2–5 s only”, “add that to the report”.
Ask in your own language. Chinese or English both work; the analysis names accept either.

Prompt library

1. Explore your data

Drag a .hdf / .dpsa file into the panel, then:

What files do I have loaded, and what channels and sample rate does the first one have?

List the analyses you can run.

2. Run an analysis, with options

Compute the 1/3-octave spectrum of channel 1 and show it.

Level vs. Time for channel 1, A-weighted, Fast time weighting — plot it.

FFT average of channel 2, 8192 lines, Hanning window, and show the chart.

Loudness of channel 1 using ISO 532-1, in sone, and plot it.

If you ask for an option an analysis does not support, the assistant tells you it is not adjustable rather than silently ignoring it. After running, it echoes the effective settings so you can confirm.

3. Analyze only part of a recording

FFT average of channel 1, but only from 2 to 5 seconds (the steady-state part).

Level vs. Time of channel 1, skip the first 3 seconds of run-up.

4. Order & RPM analyses

Run an order spectrum on channel 1 — the file has a tacho channel.

FFT vs. RPM for channel 2, RPM step 50.

Order and vs-RPM analyses need a speed reference. The assistant auto-detects the tacho channel by name. If it reports “no data” because your speed channel has an unusual name, tell it which one to use: “use channel 4 as the tacho and try again”.

5. Read specific values off the result

What are the three dominant peaks of that spectrum?

What is the level at 1000 Hz?

What is the value at t = 3 s on the Level vs. Time curve?

Which order dominates, and at what level?

6. Compare several files

Run FFT average on channel 1 of all five files and overlay them on one chart.

Now give me a comparison table of the overall levels with the delta versus the first file.

7. Batch a whole folder & find outliers

Run the 1/3-octave spectrum on every file in D:\tests\batch and summarize the overall levels in a table.

Are any of those samples outliers? Show the mean, standard deviation and flag anything beyond 2 sigma.

8. Limit / pass-fail check

Check the FFT spectrum of channel 1 against a limit of 80 dB and tell me where it exceeds it.

Apply this limit line and report pass or fail: 90 dB below 500 Hz, 80 dB above 500 Hz.

9. Export the numbers

Export that spectrum to a CSV file.

10. Listen to a channel

Play channel 1 of the first file so I can hear it.

Play it A-weighted. … Stop.

11. Continue in the workbench

Open that analysis in the workspace so I can tweak it.

This adds the file and an “AI:” analysis node (with the same settings) to the visible pipeline and runs it, so you can keep adjusting it by hand.

12. Set up recording channels

Before capturing, let the assistant configure the Recorder channel table for you:

Enable channels 6, 7 and 8 for vibration, sensitivity 50 mV/g, ICP coupling.

Show me the current recorder channel setup.

Set channel 1 as a microphone, 50 mV/Pa.

This configures the live recording table exactly as if you edited it by hand (and pushes to the hardware when a device is connected). It is separate from analyzing an already-recorded file.

13. Build a report

Add a report section titled "Baseline" with the spectrum chart and metrics table.

Show me the report outline before you generate it.

Generate the report as PowerPoint.

A full example, end to end

1.  "List the channels of the first file."
2.  "Level vs. Time of channel 1, A-weighted, from 2 to 8 seconds. Plot it."
3.  "What's the maximum level, and at what time?"
4.  "Do the same for the other four files and overlay them."
5.  "Give me a comparison table with deltas versus file 1, and flag any outliers."
6.  "Add all of this to a report with a short summary, preview the outline, then generate it as Word."

What the assistant will not do

Invent or estimate any measured value — every figure comes from a real engine run.
Read a peak or a level off a chart image — it uses the underlying data instead.
Use comparison results from other tools as analysis input — native results stay independent.

Data privacy: your prompts and the numeric summaries needed for a request are sent to the AI provider you configured. Choose a domestic provider (DeepSeek / Qwen) when data must stay in-region. See AI Assistant & Reporting for provider and key setup.