Dynamic Mass Measurement in Checkweighers Using Optimized Singular Spectrum Analysis and Robust Subspace Identification

Le résumé fourni par la source

Conveyor belt-type checkweighers are widely employed to ensure that product mass on production lines stays within specified limits. However, mechanical vibrations, measurement noise, and other external disturbances can significantly degrade their weighing accuracy. To address this issue, a novel implicitly structured coupled dynamic weighing method that integrates optimized singular spectrum analysis and robust subspace identification (OSSA-RSI) is proposed. First, the load mass estimation problem is theoretically transformed into a state-space model identification problem for an equivalent check-weigher system, providing a modeling basis for the application of subspace identification methods. Subsequently, OSSA constructs a physics-prior-driven singular value selection criterion based on the energy–frequency distribution characteristics of load cell signals, extracts the weight signal subspace, and reconstructs the signal after suppressing multi-source interference. Then, based on the reconstructed signal, RSI is applied to identify the equivalent state-space model of the checkweigher system, and the steady-state response under constant excitation is calculated to determine the load mass. To verify the effectiveness of proposed method, a series of loading experiments with different masses under various operating conditions were conducted. Experimental results demonstrate that OSSA-RSI achieves an overall average relative performance index of 0.274, representing improvements of 34.0%, 12.7%, 38.8%, and 93.2% over the existing advanced dynamic weighing methods time-varying low-pass filter and improved morphological filter (TVLP-IMF), stationary wavelet forced threshold denoising and system identification optimized by the improved whale optimization algorithm (IWOA-SWFTD-SSI), finite impulse response filtering method based on debiased local models (FIR-DLMs), and improved backpropagation neural network with ADAM optimizer (IBPNN), respectively.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.