Abstract:Highly accurate time-series vibration prediction is an important research issue for electric vehicles (EVs). EVs often experience vibrations when driving on rough terrains, known as torsional resonance. This resonance, caused by the interaction between motor and tire vibrations, puts excessive loads on the vehicle's drive shaft. However, current damping technologies only detect resonance after the vibration amplitude of the drive shaft torque reaches a certain threshold, leading to significant loads on the shaft at the time of detection. In this study, we propose a novel approach to address this issue by introducing Resoformer, a transformer-based model for predicting torsional resonance. Resoformer utilizes time-series of the motor rotation speed as input and predicts the amplitude of torsional vibration at a specified quantile occurring in the shaft after the input series. By calculating the attention between recursive and convolutional features extracted from the measured data points, Resoformer improves the accuracy of vibration forecasting. To evaluate the model, we use a vibration dataset called VIBES (Dataset for Forecasting Vibration Transition in EVs), consisting of 2,600 simulator-generated vibration sequences. Our experiments, conducted on strong baselines built on the VIBES dataset, demonstrate that Resoformer achieves state-of-the-art results. In conclusion, our study answers the question "Can Transformers Forecast Vibrations?" While traditional transformer architectures show low performance in forecasting torsional resonance waves, our findings indicate that combining recurrent neural network and temporal convolutional network using the transformer architecture improves the accuracy of long-term vibration forecasting.

What problem does this paper attempt to address?

The paper aims to address the torsional vibration prediction problem encountered by electric vehicles (EVs) when driving on rough roads. Specifically, when an electric vehicle is traveling on an uneven surface, the interaction between the motor and the tires causes torsional vibration on the drive shaft, which results in excessive loads on the drive shaft and may lead to drive shaft failures, vehicle performance degradation, and even driving hazards. Current damping technologies can only detect resonance phenomena on the drive shaft after the torsional torque vibration magnitude reaches a certain threshold, which means that significant loads have already been applied to the shaft by the time the detection occurs. To solve the above problem, the paper proposes a novel approach called Resoformer, which is a Transformer-based model used for torsional vibration prediction. Resoformer takes the time series of motor speeds as input and predicts the magnitude of torsional vibration occurring on the drive shaft after the input sequence. By calculating the attention between recursive and convolutional features extracted from the measurement data points, Resoformer improves the accuracy of vibration prediction. To validate the effectiveness of the model, the researchers conducted experiments using a vibration dataset called VIBES (Vibration dataset for EVs simulation), which consists of 2,600 simulated vibration sequences. The experimental results demonstrate that Resoformer achieves state-of-the-art performance compared to strong baseline models, proving that combining recurrent neural networks (RNNs) and time convolutional networks (TCNs) with the Transformer architecture can enhance long-term vibration prediction accuracy. In conclusion, the paper answers the question of whether Transformers can predict vibrations, pointing out that while the traditional Transformer architecture performs poorly in predicting torsional vibrations, the combination of RNNs and TCNs with the Transformer architecture significantly improves the accuracy of vibration prediction.

Can Transformers Predict Vibrations?

Enhanced Prediction of Transformers Vibrations under Complex Operating Conditions

A Fusion Transformer for Multivariable Time Series Forecasting: the Mooney Viscosity Prediction Case

Transformer Fault Prognosis Using Deep Recurrent Neural Network Over Vibration Signals

A novel Vision Transformer network for rolling bearing remaining useful life prediction

Variational autoencoders and transformers for multivariate time-series generative modeling and forecasting: Applications to vortex-induced vibrations

A Convolutional Transformer Architecture for Remaining Useful Life Estimation

Wireless Vibration Signal Prediction of Power Transformers Based on Contrastive Adversarial Sparse Model

Can Transformers Transform Financial Forecasting?

Prediction of Electric Vehicles Charging Demand: A Transformer-Based Deep Learning Approach

Bearing Remaining Useful Life Prediction based on TCN-Transformer Model

Are Transformers Effective for Time Series Forecasting?

Convolutional Neural Network-Based Transformer Fault Diagnosis Using Vibration Signals

A Transformer-based Framework For Multi-variate Time Series: A Remaining Useful Life Prediction Use Case

Transformer Conformal Prediction for Time Series

Spatial Attention-Based Convolutional Transformer for Bearing Remaining Useful Life Prediction

A transformer model with enhanced feature learning and its application in rotating machinery diagnosis

A predictive analytics framework for rolling bearing vibration signal using deep learning and time series techniques

Transformers for Energy Forecast

A Novel Transformer Model Based on Dynamic Convolution and ProbSparse Self-Attention for RUL Prediction of Bearings