A High-Fidelity Symbolization Method for Reciprocating Pump Vibration Monitoring Data

Abstract

The condition monitoring of reciprocating pumps is challenged by severe operating conditions and massive data. Symbolization has emerged as a promising solution by reducing data volume, enhancing noise resilience, and preserving key diagnostic features. However, the symbolization process inherently compromises the preservation of time-frequency characteristics inherent in the original data. This article introduces a high-fidelity symbolization method that ensures effective symbolization while maintaining high-accuracy recovery. First, a robust entropy-optimized statistical method categorizes the data into positive impulses, nonimpulses, and negative impulses, reflecting the reciprocating dynamics of pumps. Next, refined symbolization is achieved through fuzzy clustering-based modeling, transforming the data into final symbolic sequences. Finally, a Lagrangian optimization function minimizes reconstruction errors, enabling iterative recovery through gradient descent. The effectiveness and superiority of the proposed method were validated through simulation and real data. Compared to the existing approaches, the proposed method achieves a substantial reduction in recovery deviation, exceeding 94.44%, in both time and frequency domains. Additionally, the time-frequency correlation coefficients improve by over 0.25 times, reaching values greater than 0.98, demonstrating its high fidelity in preserving the amplitude and distribution characteristics of the original signals. Moreover, the method’s performance is significantly influenced by the number of symbols, with diminishing marginal utility as symbols increase.