A neural network model with filtering-based time-domain data augmentation method for predicting vibration discomfort.

Area of Science:

• Automotive Engineering
• Human Factors Engineering
• Computational Mechanics

Background:

• Occupant comfort is crucial in vehicle design, particularly in multi-purpose vehicles (MPVs) where second-row seating experiences unique vibration dynamics.
• Quantifying and predicting vibration-induced discomfort is challenging due to complex vibration transmission paths and subjective human perception.

Purpose of the Study:

• To investigate vibration-induced discomfort in the second-row seats of MPVs.
• To develop and validate a deep learning-based prediction model for occupant discomfort.
• To establish a framework for modeling the relationship between seat vibration and subjective discomfort.

Main Methods:

• Experiments were conducted on a four-poster test rig to record multi-directional accelerations at the backrest, seat pan, and armrests.
• A Long Short-Term Memory (LSTM) neural network was employed to model the nonlinear relationship between vibration data and discomfort ratings.
• Data augmentation techniques, including filtered white noise, were used to enhance the dataset and model generalizability.

Main Results:

• The deep learning model achieved high prediction accuracy for discomfort when using three-directional backrest acceleration as input.
• Including vibration data from other seat components (seat pan, armrests) as input features resulted in decreased prediction performance.
• The study successfully established a predictive framework linking objective vibration measurements to subjective discomfort levels.

Conclusions:

• Backrest acceleration is a key indicator for predicting second-row seat vibration discomfort in MPVs.
• The developed LSTM model provides an effective method for quantifying vibration-induced discomfort.
• Future research can refine the model by exploring additional vibration features and vehicle operating conditions.