Test Data Anomaly Analysis Agent

A bad sensor reading can make a good part look like a failure, and sorting real problems from data glitches takes hours. After each test run, this agent loads the data and checks sensor channels for dropouts, spikes, drift or calibration problems, marking bad segments. It then compares results with the acceptance limits and with simulation predictions. Results outside limits or far from predictions are flagged. Before calling any result a failure, it checks whether a sensor problem could explain it. If the data is too poor to judge, it recommends a re-run instead of guessing. It drafts a test summary with plots and flagged items. The engineer reviews and makes the call. Edge case: a strain reading far from simulation near a weld is flagged for both a possible model issue and a real defect.