Research on efficient virtual load spectrum extraction method for heavy-duty commercial vehicle chassis based on novel AM-P-sLSTM model

Abstract

Virtual load spectrum extraction technology significantly enhances simulation efficiency, yet traditional methods suffer from critical limitations in generation efficiency. This study focuses on virtual load spectrum extraction for a heavy-duty truck chassis. First, physical vehicle tests were conducted based on a refined multi-body dynamics (MBD) model, measuring wheel forces, axle accelerations, damper draw-wire displacements, and frame strains at a proving ground. Second, by integrating weight mechanisms, exponential gating strategies, and patch-independent channel mechanisms into a Long Short-Term Memory (LSTM) framework, this paper creatively proposes an efficient and accurate AM-P-sLSTM model. The model demonstrates superior performance compared to six advanced models across five public datasets. Third, the AM-P-sLSTM was trained using MBD-derived draw-wire sensor displacements and axle-head acceleration simulation signals as inputs, with corresponding simulated wheel-center vertical excitation signals as outputs, to extract virtual load spectra for the heavy truck. Experimental results indicate that while the AM-P-sLSTM, traditional Virtual Iteration method, and other advanced models all achieve load spectrum acquisition, the proposed AM-P-sLSTM exhibits significantly enhanced efficiency without compromising accuracy. Compared to the VI method, it improves efficiency by 35.9% and 45.9% under different cutoff frequencies. This technology is expected to substantially shorten structural fatigue validation cycles for commercial vehicles, providing high-fidelity data foundations for intelligent design and lifecycle health management of heavy equipment, thereby advancing the automotive industry toward data-driven, efficient R&D paradigms.