Search

A computer model can be used for predicting an output only after specifying the values of some unknown physical constants known as calibration parameters. The unknown calibration parameters can be estimated from real data by conducting physical experiments. This paper presents an approach to optimally design such a physical experiment. The problem of optimally designing physical experiment, using a computer model, is similar to the problem of finding optimal design for fitting nonlinear models. However, the problem is more challenging than the existing work on nonlinear optimal design because of the possibility of model discrepancy, that is, the computer model may not be an accurate representation of the true underlying model. Therefore, we propose an optimal design approach that is robust to potential model discrepancies. We show that our designs are better than the commonly used physical experimental designs that do not make use of the information contained in the computer model and other nonlinear optimal designs that ignore potential model discrepancies. We illustrate our approach using a toy example and a real example from industry., 25 pages, 10 figures

A manufacturing system collects big and heterogeneous data for tasks such as product quality modeling and data-driven decision-making. However, as the size of data grows, timely and effective data utilization becomes challenging. We propose an unsupervised data filtering method to reduce manufacturing big data sets with multi-variate continuous variables into informative small data sets. Furthermore, to determine the appropriate proportion of data to be filtered, we propose a filtering information criterion (FIC) to balance the tradeoff between the filtered data size and the information preserved. The case study of a babycare manufacturing and a simulation study have shown the effectiveness of the proposed method.

A/B tests have been widely adopted across industries as the golden rule that guides decision making. However, the long-term true north metrics we ultimately want to drive through A/B test may take a long time to mature. In these situations, a surrogate metric which predicts the long-term metric is often used instead to conclude whether the treatment is effective. However, because the surrogate rarely predicts the true north perfectly, a regular A/B test based on surrogate metrics tends to have high false positive rate and the treatment variant deemed favorable from the test may not be the winning one. In this paper, we discuss how to adjust the A/B testing comparison to ensure experiment results are trustworthy. We also provide practical guidelines on the choice of good surrogate metrics. To provide a concrete example of how to leverage surrogate metrics for fast decision making, we present a case study on developing and evaluating the predicted confirmed hire surrogate metric in LinkedIn job marketplace.

Computer experiments may involve not only continuous input factors but also nominal factors, discrete numeric factors, and ordinal factors. Most existing literature in designing computer experiment...

In multistage manufacturing systems, modeling multiple quality indices based on the process sensing variables is important. However, the classic modeling technique predicts each quality variable one at a time, which fails to consider the correlation within or between stages. We propose a deep multistage multi-task learning framework to jointly predict all output sensing variables in a unified end-to-end learning framework according to the sequential system architecture in the MMS. Our numerical studies and real case study have shown that the new model has a superior performance compared to many benchmark methods as well as great interpretability through developed variable selection techniques., Comment: Accepted by Journal of Quality Technology

To identify the robust settings of the control factors, it is very important to understand how they interact with the noise factors. In this article, we propose space-filling designs for computer experiments that are more capable of accurately estimating the control-by-noise interactions. Moreover, the existing space-filling designs focus on uniformly distributing the points in the design space, which are not suitable for noise factors because they usually follow non-uniform distributions such as normal distribution. This would suggest placing more points in the regions with high probability mass. However, noise factors also tend to have a smooth relationship with the response and therefore, placing more points towards the tails of the distribution is also useful for accurately estimating the relationship. These two opposing effects make the experimental design methodology a challenging problem. We propose optimal and computationally efficient solutions to this problem and demonstrate their advantages using simulated examples and a real industry example involving a manufacturing packing line.

Problem: The prediction of cutting forces is very important in designing mechanical micromachining to ensure geometrical accuracy of the machined feature and avoid tool breakage. These predictions can be done either via analytical models or finite-element models. The finite-element models are precise but usually time consuming to run. Analytical models, on the other hand, are less accurate but computationally much cheaper. The problem here is to integrate these two types of physics-based models and obtain an easy-to-evaluate statistical model that can approximate the machining forces. Approach: We propose performing a sensitivity analysis using the computationally cheap analytical models prior to conducting the computationally intensive finite-element simulations. With the elicited prior knowledge from the sensitivity analysis, a two-stage strategy is presented for designing the finite-element simulations in which customized number of levels can be assigned for each input factor. The finite-element simulation data can then be integrated with the analytical models in developing the final metamodel. Results: We show that the initial sensitivity analysis can reveal critical information about the underlying system and guide us to more efficiently extract information from the finite-element models. The proposed design for the finite-element simulations is comprised of two subarrays and overall can achieve desirable orthogonality and space-filling properties. Compared with using n different levels for all input factors, as in the traditional space-filling design, the new design is more capable for estimating factor interactions while still maintaining the ability to capture necessary nonlinear effects. By using fewer levels, it can also improve the efficiency of estimating the effects when the simulation is subject to stochastic noise. In model validation, our numerical results indicate that the fitted integrated metamodel can more precisely approximate the machining forces than either using the analytical models or the traditional metamodel based on the finite-element simulations alone.

In this paper we propose a novel feature extracting method based on spatiotemporal slice analyzing. To date, video features are focused on the character of every single video frame. With our method, the video content is no longer represented with every single frame. The temporal variation of visual information is taken as an important feature of video in our method. We examined this kind of feature with experiments in this paper. The experiment results show that the proposed feature is effective and robust for variant video content and format.

This paper presents a system for automatic detecting and recognizing complex individual actions in sports video to facilitate high-level content-based video indexing and retrieval. This is challenging due to the cluttered and dynamic background in sports video which makes object segmentation formidable. Another difficulty is to fully automatically and accurately detect desired actions from long video sequence. We propose three techniques to handle these challenges. Firstly, an efficient approach exploiting dominant motion and semantic color analysis is developed to detecting the highlight clips which contain athlete’s action from video sequences. Secondly, a robust object segmentation algorithm based on adaptive dynamic background construction is proposed to segment the athlete’s body from the clip. Finally, to recognize the segmented body shape sequences, the hidden markov models are slightly modified to make them suitable for noisy data processing. The proposed system for broadcast diving video analysis has achieved 96.6% detection precision; and 85% recognition accuracy for 13 kinds of diving actions.