Applied Stochastic Models in Business and Industry

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.518

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.526

This paper uses estimated model parameters as inputs into multivariate quality control charts. The thickness of paper leaving a paper mill is measured at a high sampling rate, and these data are grouped into successive data segments. A stochastic model for paper is fitted to each data segment, leading to parameter estimates and information‐based standard errors for these estimates. The estimated model parameters vary by more than one can be explained by the information‐based standard errors, suggesting that the ‘true’ underlying parameters are not constant over time. A model is formulated for the true parameters in which the information matrix dictates the distribution for the observed parameters given the true parameters. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.521

In this paper we describe an approach for establishing control limits and sampling times which derives from economic performance criteria and a model for random shifts. The total cost related to both production and control is calculated, based on cost estimates for false alarms, for not identifying a true out of control situation, and for obtaining a data record through sampling. We describe the complete process for applying the method and compare with conventional procedures to real data from a Portuguese pulp and paper industrial plant. It turns out that substantial cost‐reductions may be obtained. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.525

Sheet metal spinning is a very complex forming process with a large number of quality characteristics. Within the scope of a joint project of the Department of Statistics and the Chair of Forming Technology the impact of process parameters (design factors) on important quality characteristics has been investigated both theoretically and experimentally. In the past, every response has been treated individually and uncontrollable disturbances (noise factors) have been neglected. Now this approach has been extended to robust multiresponse parameter design. For this, a review of common multivariate approaches for robust parameter design has been carried out, which also leads to the proposal of some new variants. In addition to the theoretical comparison, the methods were applied to data gained in the sheet metal spinning process. The obtained results were evaluated in terms of applicability, limitations and quality accuracy. Practical experiments confirmed the high degree of efficiency that the finally proposed method based on desirabilities promises. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.524

Optimal inspection and maintenance of complex systems in modern industry is important for safety and economic reasons. With appropriate statistical modelling, the utilization of inspection resources and quality of inferences can be greatly improved. Modelling and inspection of a full‐scale industrial furnace subject to corrosion will be considered. A suitable Bayesian spatio‐temporal dynamic linear model is developed for wall thickness, by eliciting the beliefs of experts and incorporating other relevant data for related systems. The model may be used to derive efficient inspection schedules for corrosion detection and we demonstrate the considerable reduction in the inspection burden which the model allows. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.519

The measurement process of the thermal interface resistance tester combines physical measurements and computer simulations to obtain the measured value of the thermal resistance of a material. The computer simulation model is calibrated by 14 parameters that have been estimated from various experiments. The estimation errors in these parameters contribute to the measurement error in the thermal resistance. The following research questions were raised: (1) What calibration parameter errors have a large contribution to the thermal resistance error? (2) How does this error depend on the reference value of the thermal resistance measured under standard conditions? The main point in this paper is to show the use of statistical modelling to estimate the effect of the calibration parameter errors on the thermal resistance measurement precision by means of a tolerance design procedure that is based on the model. Our final conclusion is that two out of 14 calibration parameters dominate the thermal resistance error, and that their effects strongly depend on the reference thermal resistance. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.522

Stemming from a consulting project about a gas distribution network, a new, Bayesian model is proposed to describe failures in a complex, expanding over time, repairable system, which is split into components installed over different years. Both exchangeable and independent Poisson processes, homogeneous in space but not in time, are used to model the components. The model takes also into account missing data, due either to unrecorded early failures or unknown installation dates of failed parts. Actual escape data from a gas distribution network illustrate the implementation of the model, which relies on the use of Markov chain Monte Carlo methods. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.523

We propose a multivariate time series model to forecast the returns and volatilities of 15 European financial markets. Using the approach of mean‐variance portfolios we develop several strategies which are based on the predictions of high‐dimensional VAR‐GARCH models for future volatilities. We explore the value of volatility timing strategies by simplifying the forecasting model. One approach for information blocking is based on factor analysis for the returns. Finally we discuss if multivariate volatility timing strategies are successful for beating the benchmark index (the MSCI Europe index). Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.520

In market research, such as for the measure of the customer satisfaction, data are collected through questionnaires. Responses are often classified into ordered categories, so observed variables are ordinal and the rate of missing data may be very high. In this paper, a method for the analysis of a categorical and incomplete data matrix is proposed. Our methodology is applied to data collected by a market survey of Fiat Auto in order to show the latent dimensions underlying the customer satisfaction with car dealers. After multiple imputation of missing values the polychoric correlation matrix, measuring the manifest variables correlations, is computed and used as a proper input to factor analysis. Two factors underlying the several judgement items are thus obtained and their weights on the global judgement ordinal variable are then estimated by ordered probit regression. Copyright © 2004 John Wiley & Sons, Ltd.

Ruggeri, Fabrizio; Polasek, Wolfgang

2004 Applied Stochastic Models in Business and Industry

doi: 10.1002/asmb.527

This paper presents a practical procedure for performing non‐parametric bivariate regression analysis. The procedure applies the Nadaraya–Watson local linear kernel estimator with associated bootstrap variability bands whenever the pseudo‐likelihood ratio test rejects the linear regression model hypothesis. Two case studies and simulations are used to demonstrate the proposed technique. Calculations have been performed using the shareware R software. Copyright © 2004 John Wiley & Sons, Ltd.