Real-Time Weld Width Prediction and Recording Program

• Desktop program for weld measurement experiments at a research institute
• Both a high-speed camera and a spectrometer connected and controlled from one screen
• Camera view, emission spectrum, and predicted width visible while welding is in progress
• One operator handles the full run: connect, observe, record

• Two measurement devices of different nature to be operated within a single experiment
• Different acquisition rates on the two devices → frames and spectra must be paired by time before any estimation
• A welding run lasting only seconds → the preceding window must be captured whole, without missing the start
• Repeated runs under changing capture conditions → settings to be stored and restored exactly

• Single connect action opening both devices and starting the live view
• Camera view, emission spectrum, and predicted width trace shown side by side
• Current predicted width reported with the mean and standard deviation for the window
• Recording length set between 1.0 and 9.9 seconds, with the preceding window saved by one button
• Device-side errors surfaced on screen at once, with the connection released safely

Workflow

• Set output paths → connect devices → observe live → start recording → run saved

Data

• Recent data from both devices held in buffers while connected → the preceding window pulled out at save time
• Frames and spectra paired by nearest timestamp for estimation
• One measurement producing a video file, a spectrum table by time, and a predicted width table by time

time,width
-4.98,3.07
-4.97,3.05

• Each device acquiring at its own rate, with screen refresh on a separate fixed interval → the faster device not held back by the slower one
• Capture running above the configured frame rate, then thinned to even intervals at save time → jitter in frame spacing evened out
• Region of interest drawn within the full sensor area as values change → capture area confirmed before connecting
• Three video quality presets → a choice between image quality and file size
• Estimation model kept as a replaceable external file → a retrained model drops in without program changes, and a default model is generated when the file is absent
• Value ranges checked when a settings file is loaded → connection attempts with values the devices cannot accept are blocked

• Predicted width trend readable on screen while welding is in progress
• Video, spectra, and predicted width kept for the same window → cross-referencing possible in later analysis
• Experiment conditions stored as settings files → identical runs repeatable
• Reuse for other conditions through a model file swap alone
• Supported device models stated inside the program → less confusion when other hardware is connected