The customer had a large number of microphones (over 100) installed on its acoustic test bench. These microphones were being measured at high sampling rates over a long period of time, so the probability of a transducer problem was high. It was difficult to examine each sensor by hand.
Automatic failure detection was needed to enable rapid remeasurement, as well as identification of the type of failure to assess its criticality.
DiagFit was used as a default detector and default classifier, so that the measurement could be redone as quickly as possible.
Excellent detection and class prediction on labeled data compared to the reference. Detection of real defaults not seen by the customer. Finally, detection of tagging errors made by the user: the model’s prediction was more reliable than the “ground truth” given by the customer.