StartLux: Compressed-Air Leak Localization Simulation

A compressed-air system has low header pressure and excessive idle consumption. This simulation follows an inspection vehicle as it gathers evidence across three branches, narrows the search to branch B, and checks the diagnosis by isolating that branch and measuring again. What happens in the video: - Read the header pressure and consumption baseline, then scan branches A, B, and C with a wide directional sensing mode. - Revisit the B1 and B2 connections with a narrow mode to collect more localized evidence. - Ask the PLC to isolate branch B. The vehicle performs sensing; a separate valve actuator performs the isolation. - Remeasure header pressure and consumption and scan branch B again before submitting the leak diagnosis. The final screen shows the simulated header at 6.04 bar and 12.0 NL/min, with the post-isolation acoustic reading down to 33.8 dB. These are values from this recorded example, not factory acceptance thresholds. The useful pattern is to choose a test, intervene, then obtain new evidence instead of treating the first suspicious signal as a confirmed fault. According to the supplied technical notes, StartLux receives public topology, collected measurements, vehicle state, and completed-test history; the injected fault is withheld from its input. The surrounding program generates legal candidates, checks response freshness, controls motion and sampling, and confirms that the selected valve has actually changed state in the simulation. The model selects among those candidates; it does not steer the wheels or calculate the pressure dynamics. This is a MuJoCo-based simulation with a simplified gas-network model and synthetic directional acoustic values, not real audio, an acoustic-camera test, or a repair demonstration. This publication reviewed the recording and supplied documentation without rerunning the simulation or auditing the raw inference traces. The original 48-second recording has Chinese on-screen labels and no audio. The displayed choice probabilities are not fault probabilities or a reliability guarantee.