arXiv subjects

Wenbo Sui

Publications and source records attributed to Wenbo Sui.

5 recordsLinked to original sources

arXiv2026

LIBAD: A Multimodal Anomaly Detection Benchmark for Li-Ion Battery Electrode Manufacturing

Multimodal industrial anomaly detection has largely focused on discrete products using strongly correlated RGB and 3D observations, leaving continuous process manufacturing and weakly correlated sensing modalities underexplored. We introduce LIBAD, the first multimodal anomaly detection benchmark for Li-ion battery electrode manufacturing. Collected from real roll-to-roll production lines, LIBAD provides aligned double-sided visible-light imaging, high-resolution X-ray radiography, and inline-compatible low-resolution X-ray radiography. Electrode patches in LIBAD exhibit highly homogeneous material appearance, while defect evidence can be strong in one modality but weak or absent in another, resulting in pronounced cross-modal anomaly inconsistency. Benchmarks of representative methods under the inline-compatible visible-light and low-resolution X-ray setting exhibit limited transferability and consistently high false-positive rates. We therefore propose DA-Core, a memory-based method that jointly considers feature-space coverage and local density of normal features during coreset selection, allowing compact memory banks to better preserve fine-grained normal variations. With a coreset ratio of 0.05, DA-Core reduces FPR95 from 60.4% to 54.3% compared with standard farthest point sampling. At this ratio, DA-Core also outperforms the best standard coreset result (obtained at 0.20) while reducing inference time by 43.9%. These results suggest that both the data distribution of normal features and the modality relationship itself require explicit consideration when designing anomaly detection methods for process manufacturing.

cs.CV↗

arXiv2025

Incomplete Multimodal Industrial Anomaly Detection via Cross-Modal Distillation

Recent studies of multimodal industrial anomaly detection (IAD) based on 3D point clouds and RGB images have highlighted the importance of exploiting the redundancy and complementarity among modalities for accurate classification and segmentation. However, achieving multimodal IAD in practical production lines remains a work in progress. It is essential to consider the trade-offs between the costs and benefits associated with the introduction of new modalities while ensuring compatibility with current processes. Existing quality control processes combine rapid in-line inspections, such as optical and infrared imaging with high-resolution but time-consuming near-line characterization techniques, including industrial CT and electron microscopy to manually or semi-automatically locate and analyze defects in the production of Li-ion batteries and composite materials. Given the cost and time limitations, only a subset of the samples can be inspected by all in-line and near-line methods, and the remaining samples are only evaluated through one or two forms of in-line inspection. To fully exploit data for deep learning-driven automatic defect detection, the models must have the ability to leverage multimodal training and handle incomplete modalities during inference. In this paper, we propose CMDIAD, a Cross-Modal Distillation framework for IAD to demonstrate the feasibility of a Multi-modal Training, Few-modal Inference (MTFI) pipeline. Our findings show that the MTFI pipeline can more effectively utilize incomplete multimodal information compared to applying only a single modality for training and inference. Moreover, we investigate the reasons behind the asymmetric performance improvement using point clouds or RGB images as the main modality of inference. This provides a foundation for our future multimodal dataset construction with additional modalities from manufacturing scenarios.

cs.CV↗

arXiv2019

Combustion Phasing Modelling and Control for Compression Ignition Engines with High Dilution and Boost Levels

Because fuel efficiency is significantly impacted by the timing of combustion in internal combustion engines, accurate control of combustion phasing is critical. In this paper, a nonlinear combustion phasing model is introduced and calibrated, and both a feedforward model-based control strategy and an adaptive model-based control strategy are investigated for combustion phasing control. The combustion phasing model combines a knock integral model, burn duration model and a Wiebe function to predict the combustion phasing of a diesel engine. This model is simplified to be more suitable for combustion phasing control and is calibrated and validated using simulations and experimental data that include conditions with high exhaust gas recirculation fractions and high boost levels. Based on this model, an adaptive nonlinear model-based controller is designed for closed-loop control, and a feedforward model-based controller is designed for open-loop control. These two control approaches were tested in simulations. The simulation results show that during transient changes the CA50 (the crank angle at which 50% of the mass of fuel has burned) can reach steady state in no more than 5 cycles and the steady state errors are less than +/-0.1 crank angle degree (CAD) for adaptive control, and less than +/-0.5 CAD for feedforward model-based control.

eess.SY↗

arXiv2019

Cylinder-Specific Model-Based Control of Combustion Phasing for Multiple-Cylinder Diesel Engines Operating with High Dilution and Boost Levels

Accurate control of combustion phasing is indispensable for diesel engines due to the strong impact of combustion timing on efficiency. In this work, a non-linear combustion phasing model is developed and integrated with a cylinder-specific model of intake gas. The combustion phasing model uses a knock integral model, a burn duration model and a Wiebe function to predict CA50 (the crank angle at which 50% of the mass of fuel has burned). Meanwhile, the intake gas property model predicts the EGR fraction and the in-cylinder pressure and temperature at intake valve closing (IVC) for different cylinders. As such, cylinder-to-cylinder variation of the pressure and temperature at intake valves closing is also considered in this model. This combined model is simplified for controller design and validated. Based on these models, two combustion phasing control strategies are explored. The first is an adaptive controller that is designed for closed-loop control and the second is a feedforward model-based control strategy for open-loop control. These two control approaches were tested in simulations for all six cylinders and the results demonstrate that the CA50 can reach steady state conditions within 10 cycles. In addition, the steady state errors are less than +/-0.1 crank angle degree (CAD) with the adaptive control approach, and less than +/-1.3 CAD with feedforward model-based control. The impact of errors on the control algorithms is also discussed in the paper.

eess.SY↗

arXiv2019

Modeling and Control of Combustion Phasing in Dual-Fuel Compression Ignition Engines

Dual fuel engines can achieve high efficiencies and low emissions but also can encounter high cylinder-to-cylinder variations on multi-cylinder engines. In order to avoid these variations, they require a more complex method for combustion phasing control such as model-based control. Since the combustion process in these engines is complex, typical models of the system are complex as well and there is a need for simpler, computationally efficient, control-oriented models of the dual fuel combustion process. In this paper, a mean-value combustion phasing model is designed and calibrated and two control strategies are proposed. Combustion phasing is predicted using a knock integral model, burn duration model and a Wiebe function and this model is used in both an adaptive closed loop controller and an open loop controller. These two control methodologies are tested and compared in simulations. Both control strategies are able to reach steady state in 5 cycles after a transient and have steady state errors in CA50 that are less than 0.1 crank angle degree (CAD) with the adaptive control strategy and less than 1.5 CAD with the model-based feedforward control method.

eess.SY↗