Blog topic
Features, models, validation and hardware for AI designed to run at the machine.
9 articles
Spiking networks and event-driven chips promise anomaly detection on microwatts. What neuromorphic computing is, what vibration research shows and how it compares with a tuned microcontroller.
Clemens SaurA loose accelerometer looks like a machine fault to a model. How self-monitoring sensors detect their own faults and keep them out of the fault statistics.
Saichand GourishettiRaw vibration, sound and thermal data quickly outgrow any uplink. A practical comparison of edge and cloud evaluation, with numbers, regulations and a decision table.
Saichand GourishettiWhat a Qualcomm QCS6490-class processor brings to a multi-sensor edge hub: CPU, GPU, ISP and NPU, read from the datasheet and translated into sensing workloads.
Clemens SaurThere is no magic number of recordings. What really sets the amount of training data for acoustic inspection, and how to estimate it from a small pilot campaign.
Saichand GourishettiA model can score near 100 % and still have learned nothing useful. How shortcut learning creeps into industrial data and how to design a validation that catches it.
Saichand GourishettiIs it abnormal, what is it, how bad is it? Three model types answer three different questions and need very different data. How to climb the ladder in order.
Saichand GourishettiHow to turn raw vibration, sound and thermal signals into features a model can use: physics first, then time-domain, spectral and envelope features, then validation.
Saichand GourishettiEdge AI in manufacturing means evaluating sensor data with trained models at the machine. What it is used for, what hardware it needs and where projects fail.
Saichand GourishettiNext step
Write to us and we pass your question to the author.
