Deep Learning SPIN Pattern Outlier Detection for Integrated Dynamic Rotary Machine

Abstract

Recently, the intelligent and advanced IT technology such as IoT, sensor, network and computer vision developed the rotation machinery outlier detection and condition diagnosis technology from vibration sensor data for various industrial environment. However, the rotary machine consists of complex system with various parts and operates under the dynamic environment, almost anomaly detection is not focused on the in–out combined information of rotary machine. With not according to multiple information, anomaly detection doesn’t process in the fluent way and be difficult to instantaneous decision making. This paper suggests Spectrogram Power Integrated Pattern (SPIN Pattern) Outlier Detection available to detect outliers based on integrated and multiple frequency patterns of rotary machines. SPIN Pattern extracts vibration frequency patterns from spectrogram image (Spectrogram Pattern), inside vibration attributes and then rotary capacity power frequency patterns (Power Pattern) which is external information. Considering integrated vibration frequency pattern of inside information and power pattern for outside information at the same time, SPIN Pattern is to derive subdivided pattern for fluent outlier causes. After deriving SPIN Pattern, CNN multi-classification model performed outlier detection based on SPIN Pattern and resulted in 85% high accuracy which is confirmed to stable outlier detection and a cause derivation. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.