Search

ABSTRACTSensor implementation together with data processing in manufacturing provides continuous measurements enabling process monitoring, optimization, and automation. Additive manufacturing (AM) technologies are shifting from one-off and prototyping production to batch, mass, and continuous production. Current AM technologies are mostly focused on small production devices, without monitoring systems. This study aims to bring AM closer to automation by proposing an online monitor system for bottom-up photopolymerization AM (VPP) systems. The sensor-generated data is used to capture the detachment error of built parts from the build platform that otherwise cannot be observed physically by the machine operator. The detachment itself will not stop the build job, which results in lost material and operating time. The online monitoring procedure consists of two phases, an offline and an online one, respectively. The offline phase is used for training a prediction model to be used in connection with a control chart for online monitoring. The monitoring control chart is advantageous as only the detachment predictions need to be recorded. The research carried out in this article brings novelty in both the vat photopolymerization set-up, as well as in an online monitoring procedure that can be easily extended to similar technologies.

ABSTRACTThe current work addresses an industrial problem related to injection moulding manufacturing with focus on mould wear-out prediction. Real data sets are provided by an industrial partner that uses a multitude of moulds with different shapes and sizes in its production. An analysis of the data is presented and begins with clustering the moulds based on their characteristics and pre-chosen running settings. Using the results of the clustering, the mould wear-out is modelled using Kaplan-Meier survival curves. Furthermore, a random survival forest model is fitted for comparison and model performance is assessed. The main novelty of the case study is the implementation of mould wear-out prediction in real-time with the outcomes presented in terms of conditional survival curves including a proposed early warning system. For visualization and further industrial implementation, an R Shiny dashboard is developed and presented.

The unprecedented growth of machine-readable data throughout modern industrial systems has major repercussions for process monitoring activities. In contrast to model-based process monitoring that requires the physical and mathematical knowledge of the system in advance, the data-driven schemes provide an efficient alternative to extract and analyze process information directly from recorded process data. This paper introduces the least squares sparse principal component analysis to obtain readily interpretable sparse principal components. This is done in the context of parallel coordinates, which facilitate the visualization of high dimensional data. The key contribution is the establishment of control limits on independent sparse principal component and residual spaces to facilitate fault detection, complemented by the use of the Random Forests algorithm to carry out the fault diagnosis step. The proposed method is applied to the Tennessee Eastman process to highlight its merits.

With the emergence of Industry 4.0 and Big Data initiatives, there is a renewed interest in leveraging the vast amounts of data collected in (bio)chemical processes to improve their operations. The objective of this article is to provide a perspective of the current status of Big-Data-based process control methodologies and the most effective path to further embed these methodologies in the control of (bio)chemical processes. Therefore, this article provides an overview of operational requirements, the availability and the nature of data, and the role of the control structure hierarchy in (bio)chemical processes and how they constrain this endeavor. The current state of the seemingly competing methodologies of statistical process monitoring and (engineering) process control is examined together with hybrid methodologies that are attempting to combine tools and techniques that belong to either camp. The technical and economic considerations of a deeper integration between the two approaches is then explored, and a path forward is proposed.

Principal component analysis (PCA) has been widely used for data dimension reduction and process fault detection. However, interpreting the principal components and the outcomes of PCA-based monitoring techniques is a challenging task since each principal component is a linear combination of the original variables which can be numerous in most modern applications. To address this challenge, we first propose the use of sparse principal component analysis (SPCA) where the loadings of some variables in principal components are restricted to zero. This paper then describes a technique to determine the number of non-zero loadings in each principal component. Furthermore, we compare the performance of PCA and SPCA in fault detection. The validity and potential of SPCA are demonstrated through simulated data and a comparative study with the benchmark Tennessee Eastman process.

Climate factors such as humidity and temperature have a significant impact on the corrosion reliability of electronic products. Given the huge geographical variability in climate conditions globally, a climate classification is a useful tool that simplifies the problem of considering climate when designing electronics packaging. Most current guidelines for electronic product design rely on the Köppen–Geiger classification first developed by Köppen over a century ago. Köppen devised a set of heuristics to separate climates to match different vegetation types. These climate classes are unlikely to be the optimal for electronic product design. This paper presents a new climate classification using parameters important for corrosion reliability of electronics. The classification is based on real climate data measured every 3 h during a 5-year period at over 9000 locations globally. A key step is defining relevant features of climate affecting corrosion in electronics. Features related to temperature are defined, but also the amount of time that the difference between Temperature and Dew Point is less than 1, 2 or 3 ℃. These features relate to the risk of condensation in electronic products. The features are defined such that diurnal, seasonal and yearly variation is taken into account. The locations are then clustered using K-means clustering to obtain the relevant climate classes. This data-driven classification, based on key features for corrosion reliability of electronics, will be a useful aid for product design, reliability testing and lifetime estimation.

Split-plot designs are commonly used in industrial experiments when there are hard-to-change and easy-to-change factors. Due to the number of factors and resource limitations, it is more practical to run a fractional factorial split-plot (FFSP) design. These designs are variations of the fractional factorial (FF) design, with the restricted randomisation structure to account for the whole plots and subplots. When all the experimental runs cannot be performed under the same conditions, the designs are split into blocks. We discuss the formulation of blocked fractional factorial split-plot (BFFSP) designs using integer programming (IP) to achieve various design criteria.

Split-plot designs are commonly used in industrial experiments when there are hard-to-change and easy-to-change factors. Due to the number of factors and resource limitations, it is more practical to run a fractional factorial split-plot (FFSP) design. These designs are variations of the fractional factorial (FF) design, with the restricted randomisation structure to account for the whole plots and subplots. We discuss the formulation of FFSP designs using integer programming (IP) to achieve various design criteria. We specifically look at the maximum number of clear two-factor interactions and variations on this criterion.

One of the main features that distinguish split-plot experiments from other experiments is that they involve two types of experimental errors: the whole-plot (WP) error and the subplot (SP) error. Taking this into consideration is very important when computing measures of adequacy of fit for split-plot models. In this article, we propose the computation of two R2, R2-adjusted, prediction error sums of squares (PRESS), and R2-prediction statistics to measure the adequacy of fit for the WP and the SP submodels in a split-plot design. This is complemented with the graphical analysis of the two types of errors to check for any violation of the underlying assumptions and the adequacy of fit of split-plot models. Using examples, we show how computing two measures of model adequacy of fit for each split-plot design model is appropriate and useful as they reveal whether the correct WP and SP effects have been included in the model and describe the predictive performance of each group of effects.

Split-plot designs are effective in industry due to time and/or cost constraints, restriction on randomization of the treatment combinations of the hard-to-change factors, and different sizes of experimental units. Some of the results of fractional factorial split-plot experiments can be ambiguous and a need may arise to conduct follow-up experiments to separate effects of potential interest by breaking their alias links with others. For completely randomized fractional factorial experiments, methods have been developed to construct follow-up experiments. In this article, we extend the foldover technique to break the alias chains of split-plot experiments. Because it is impractical or not economically possible to foldover the whole-plot factors, as their levels are often hard or expensive to change, the focus of this article is on folding over only one or more subplot factors in order to de-alias certain effects. Six rules are provided to develop foldovers for minimum aberration resolution III and resolution IV fractional factorial split-plot designs.

SYNOPTIC ABSTRACTD-optimal designs have proved useful in analyzing common factorial experiments involving multilevel categorical factors. When analyzed by ANOVA, they allow the estimation of coefficients in a regression equation and the contributions to the variance by the main effects and interactions. If the measurement of contribution to variance is necessary but the estimation of all interaction coefficients in the regression equation is not, it is possible to reduce the number of experimental runs below a minimum D-optimal design, using what we call truncated D-optimal screening designs. D-efficiency calculations are not available due to the singularity of the design matrix; another method must be used to pare down the matrix while maintaining reasonable estimation of the original full factorial data. Covering arrays are adapted to guide this reduction. Combining properties of D-optimal designs and covering arrays produces designs that perform well at estimating full factorial results. A method is then developed to target specific interactions prior to the design of the experiment when process specific knowledge is available to indicate which interactions are least important.

A model-based approach to the forecasting of short-range product demand within the semiconductor industry is presented. Device-level forecast models are developed via a novel two-stage stochastic algorithm that permits leading indicators to be optimally blended with smoothed estimates of unit-level demand. Leading indicators include backlog, bookings, delinquencies, inventory positions, and distributor resales. Group level forecasts are easily obtained through upwards aggregation of the device level forecasts. The forecasting algorithm is demonstrated at two major US-based semiconductor manufacturers. The first application involves a product family consisting of 254 individual devices with a 26-month training dataset and eight-month ex situ validation dataset. A subsequent demonstration refines the approach, and is demonstrated across a panel of six high volume devices with a 29-month training dataset and a 13-month ex situ validation dataset. In both implementations, significant improvement is realised versus legacy forecasting systems. [Received 11 May 2007; Revised 5 September 2007; Accepted 15 October 2007]

AbstractAn overview of statistical forecasting methodology is given, focusing on techniques appropriate to short- and medium-term forecasts. Topics include basic definitions and terminology, smoothing methods, ARIMA models, regression methods, dynamic regression models, and transfer functions. Techniques for evaluating and monitoring forecast performance are also summarized.

Split-plot experiments are commonly used in industry for product and process improvement. Recent articles on designing split-plot experiments concentrate on minimum aberration as the design criterion. Minimum aberration has been criticized as a design criterion for completely randomized fractional factorial design and alternative criteria, such as the maximum number of clear two-factor interactions, are suggested (Wu and Hamada (2000)). The need for alternatives to minimum aberration is even more acute for split-plot designs. In a standard split-plot design, there are several types of two-factor interactions, not all of them equally interesting. However, minimum aberration is not designed to distinguish among the different types of two-factor interactions. It should be noted that this criticism is valid not only for the minimum aberration but also for any other design criteria originally proposed for completely randomized designs. Consequently, we provide a modified version of the maximum number of clear two-factor interactions design criterion to be used for split-plot designs.