I work for a smart phone manufacturer. We have internal manufacturing tests that ALL our smart phones go through. Each of these tests have lower-upper specs for test results to fail when outside of the specs. With lots of internal test data on our hands, we are trying to pro-actively identify any anomaly detection in the trends of ANY of our test’s results that may still be within the spec (and hence pass the overall test for that phone), but could be drifting on one side OR an anomaly compared to general overall trend of majority of units. Such phones can potentially go out in the field and fail. We are trying to find out if HTM offers any specific algorithm or solution that may help us pro-actively identify anomalies in test results of passed units and prevent them from shipping out? Is anyone aware of any such practical application of anomaly detection where HTM may have been used?
We have a paper about it:
It is important that your contain contains temporal sequences. There must be some trend over time to find in the data.