Modified inception V3 based blockage fault diagnosis of centrifugal pump

Abstract

Centrifugal pumps play a vital role across numerous industries—including water treatment, manufacturing, oil and gas, and chemical processing—but remain vulnerable to hydraulic faults that can severely degrade performance and precipitate unplanned shutdowns. Among these, blockage faults in the suction line, discharge line, or both are often overlooked in diagnostics despite their potential to trigger catastrophic pump failures. Prompt and accurate detection of blockage severity is therefore essential for maintaining pump reliability, ensuring operational safety, and minimizing maintenance costs. In this work, we present a deep learning–based framework that relies solely on suction pressure signals for the classification and severity assessment of blockage conditions. Suction pressure time-series are first transformed into two-dimensional time–frequency scalograms via the Continuous Wavelet Transform (CWT), which captures the transient and nonstationary features of the signal with higher resolution than traditional FFT or STFT approaches.