An Edge AI Test Bench for Unsupervised Anomaly Detection

Sammanfattning

Condition monitoring of rotating machinery requires time-series representations thatare compact enough to run on edge hardware, stable enough to generalize acrossproduction runs, and informative enough to support downstream anomaly detectionwithout labeled fault examples. This thesis addresses all three requirementson a purpose-built, low-cost CNC-inspired test bench equipped with a brushlessDC spindle and three stepper feed axes, instrumented for synchronized current,vibration, and speed sensing at approximately 91 Hz. A three-tier edge platform—ESP32-P4 acquisition node, Raspberry Pi 3 gateway, and Raspberry Pi 5 inferencenode—acquires a dataset of 94 636 samples across ten labeled operating cycles.Four encoders spanning a wide capacity range—per-channel summary statistics,FFT amplitude bins, the self-supervised TS2Vec encoder, and three sizes of thepre-trained MOMENT transformer—are evaluated on 20 public UCR and UEAdatasets as a cross-benchmark reference and on the CNC bench as the target domain,using four axes: a supervised linear probe, unsupervised clustering, six mode-awaregeometry metrics, and a CPU edge-deployment benchmark.The main findings are as follows. First, encoder rankings are dataset-dependent:TS2Vec leads on cross-benchmark accuracy but is outperformed on CNC by bothMOMENT-large (0.850 accuracy) and the parameter-free Summary baseline (0.844),a reversal explained by the small CNC training set and the high mode-discriminabilityof the raw sensor channels. Second, geometric stability and label-aware accuracyrank encoders differently: MOMENT’s embedding space is roughly an order of magnitudemore stable across production runs than Summary’s, making it the betterfoundation for run-disjoint anomaly detection despite similar classification scores.Third, post-training INT8 quantization collapses MOMENT’s accuracy to chancewhen applied naively; restricting INT8 to the FFN linears preserves FP32 accuracyat all three model sizes with a 1.33–1.82× disk reduction and 1.25–1.55× latencyreduction, and the small and base variants run comfortably within the per-windowbudget on the reference CPU.