Publication informationVehicle System Dynamics, 46 (6) AbstractRoad roughness is a broad term that incorporates everything from potholes and cracks to the random deviations that exist in a profile. To build a roughness index, road irregularities need to be measured first. Existing methods of gauging the roughness are based either on visual inspections or using one of a limited number of instrumented vehicles that can take physical measurements of the road irregularities. This paper proposes the collection of data from accelerometers fixed in a specific vehicle type and the use of this data to estimate the road condition. While the estimate is approximate, accelerometers are being increasingly used by car manufacturers to improve suspension performance and the proposed method is relatively inexpensive to implement and provide road managers with constantly updated measurements of roughness. This approach is possible due to the relationship between the power spectral densities of road surface and vehicle accelerations via a transfer function. This paper shows how road profiles can be accurately classified using axle and body accelerations from a range of simulated vehicle-road dynamic scenarios. This is an electronic version of an article published in Vehicle System Dynamics, 46 (6): 483-499. Vehicle System Dynamics is available online at: http://www.informaworld.com/smpp/title~db=all~content=t713659010 2

Multi-modal pre-training models have been intensively explored to bridge vision and language in recent years. However, most of them explicitly model the cross-modal interaction between image-text pairs, by assuming that there exists strong semantic correlation between the text and image modalities. Since this strong assumption is often invalid in real-world scenarios, we choose to implicitly model the cross-modal correlation for large-scale multi-modal pretraining, which is the focus of the Chinese project 'Wen-Lan' led by our team. Specifically, with the weak correlation assumption over image-text pairs, we propose a twotower pre-training model called BriVL within the crossmodal contrastive learning framework. Unlike OpenAI CLIP that adopts a simple contrastive learning method, we devise a more advanced algorithm by adapting the latest method MoCo into the cross-modal scenario. By building a large queue-based dictionary, our BriVL can incorporate more negative samples in limited GPU resources. We further construct a large Chinese multi-source imagetext dataset called RUC-CAS-WenLan for pre-training our BriVL model. Extensive experiments demonstrate that the pre-trained BriVL model outperforms both UNITER and OpenAI CLIP on various downstream tasks.

No abstract

No abstract

Publication informationJournal of Sound and Vibration, 293 (1-2): 125-137Publisher Elsevier ABSTRACTThe paper considers the influence of the surface profile on the dynamic amplification of a simply supported bridge when subject to a quarter car vehicle model. The effect of the profile irregularities on the bridge dynamic amplification is characterized with a 'response surface' giving dynamic amplification due to a 'unit ramp' at any location. Even though the dynamic interaction problem is non-linear, the effects of all ramps which together make up a road profile can be calculated separately and added using the 'response surface'. This superposition process achieves reasonable accuracy for 'good' (moderately smooth) surfaceprofiles. An accurate estimate of dynamic amplification for bridges is demonstrated with a wide range of good profiles.Keywords: DAF, DAE, dynamic, roughness, bridge, vehicle.