Our client, nervous about letting faulty cars through their production line, was detecting production faults with a very high false positive rate. It takes a long time and a lot of resources to check all vehicles; therefore, if you can be more discerning about which ones are not actually defective, that would signify a significant OPEX reduction.
250-300k rows of data from multiple sensor readings.
Logical Glue built Neural Network models to recognise a number of different specific faults using interactions within the sensor data, rather than a catch-all generic fault.
By bucketing the results of these models according to the level of fault risk, it was possible to greatly increase the number of cars that cleared production testing without letting any more faulty cars through. This allowed the manufacturer to free up significant resource from the checking process.