Single-Channel Neyman-Pearson Detection of Impulses in Tainter Valve Machinery Systems

Abstract

Unexpected mechanical failures in lock fill/empty Tainter valves can significantly disrupt waterway traffic. Accurate diagnosis and prognosis of fault mechanisms are essential for timely interventions by operations and maintenance teams. To support this, an in-service lock operated by the U.S. Army Corps of Engineers was instrumented with a comprehensive sensor suite—including accelerometers, strain gauges, inclinometers, thermocouples, and electrical current transformers—spanning the entire drive system from the motor to the final gear shaft. Data were collected during valve opening and closing operations using an event-driven approach. Notably, impulses were observed in the accelerometer signal at the sector gear during valve opening, coinciding with anomalies in the sector gear inclinometer data. To detect these impulses, a single- channel binary hypothesis testing framework based on the Neyman-Pearson theorem was implemented, modeling the signal as either zero-mean Gaussian noise or a non-zero mean Gaussian distribution with deterministic impulse magnitude incorporated. To address signal non-stationarity, an optimal sliding window size was selected to maintain stationarity and statistical reliability. Results from the Neyman-Pearson detection model are discussed, offering valuable insights into the operational health of the Tainter valve system and supporting the development of more robust diagnostic tools for lock infrastructure.