Nassim Boudaoud

A Comparative Study of CUSUM and EWMA Charts: Detection of Incipient Drifts in a Multivariate Framework

This contribution to the field of statistical process control concerns a multivariate control chart in the case of incipient process drifts. The design of this chart is based on the assumption that an a priori knowledge of the process operation states is available. The detection is based on a multimodal T 2 hoteling statistic and a dynamic model of the drift. The generated diagnosis makes possible the anticipation of the direction of the drift (toward a known or unknown operation state). The performances of this chart are compared with the classical T 2 chart using a set of simulated data.

Statistical modeling and optimization for diesel engine calibration

The European standards concerning the pollutants emissions of automotive engine become more and more severe. Modern automotive engines are equipped with an increasing number of new technologies and controlling elements. The consequence of this evolution, is the increase of the number of the controllable parameters, the difficulty to understand the engine behavior, and to find the parameters settings that offer the best compromise across the entire engine map, especially between fuel consumption and emissions constraints. This paper deals with problem of engine calibration, using the minimum of experiments. The approach proposed consists in building a global emulator based on Kriging model, which was adapted to take into account a number of control parameters greater than 3, while existing software are limited to two control parameters. This model is used to predict an engine response, and is coupled with a genetic algorithm, in order to give a best setting of parameters, optimizing the fuel consumption within constraints on the emission of NO∞ (nitrogen oxide). The main advantage of this approach is, its capacity to take into account a considerable number of controllable parameters in the optimization process, without lost in accuracy of model prediction.

Case Study: Color Control in the Automotive Industry

Numerous manufacturers are concerned with ensuring the homogeneity in the colors of different parts that compose their products. This quality objective of primary importance is especially challenging when the parts are sourced by different suppliers. The technical difficulties are accentuated when metallic or pearly paints are used since the presence of metallic chips in the paint causes a variation in the color according to the angle of observation. Also, the development of metametric colors increases these difficulties even more, as these colors are sensitive to different illumination sources. The present study relates to the production of colored car bumpers. By using a spectrocolorimeter and color samples, the process is improved to better respect the customer specifications. To achieve this, critical colors and significant parameters affecting colors are identified, design of experiments is used to optimize the process settings, and a correlation study allows further process improvement. The results achieved are impressive: the quality index used has been improved by 67% in only 6 months. This case study therefore illustrates how simple quality tools can be used in a rigorous search for process improvement toward total color mastering with zero defect objective.

A new Bayesian technique for readjusting LOLIMOT models: example with diesel engine emissions

Constraints on diesel engine emissions have increased dramatically over the past ten years. In this situation, design of experiments (DoE) are generally used to model the engines exhaust emissions (EE) behaviour. The main drawback of parametric modelling is that, if the system evolves (e.g. new product development), then the model is no longer valid. Our proposition, focused on change management, is based on the Bayesian theory and presents two new algorithms. The aim of this paper is to outline a method for readjusting LOLIMOT models resulting from the DoE with as little data as possible, in order to optimise the EE of new engines. Two algorithms are presented: one use new data to readjust the model and the other, use both new data and expert judgement. We prove that Bayesian theory could be used to reduce the number of required test points and so, the cost of new product development.

Two stage approaches for modeling pollutant emission of diesel engine based on Kriging model

The automotive industry faces the competing goals of producing better performing vehicles and keeping development time with low costs. It is crucial for the manufacturers to be able to produce fuel-economic vehicles, which respect pollutant emissions standards, and which meet the customers expectations. Accordingly, the complexity of the engines responses we have to optimize and the number of the parameters to control during the design stage, have increased rapidly, in the later years. In order to deliver vehicles, which respond to these requirements, in a reasonable time scale. Companies use design of experiments (DOE) (Schimmerling et al., 1998) in one side, and modelling, in the other side. DOE is a power tool, but the cost of the experiments and their duration, particularly in the field of pollutant emissions, can be a limit to their use in automotive industry.

The Desirability Function in a Multiresponse Optimisation Framework: A Case Study

In this paper we present a review of the different approaches in a multi response optimisation design framework. Different criteria are reviewed and discussed. A comparative study is presented using different multi-responses optimization techniques. This study concerns the design of an insulation product

The T/sup 2/ control chart in a multimodal framework: detection of incipient trends

The aim of this paper is to propose a new statistic for monitoring multivariate trend processes. We focus on the possible choices of more sensitive statistics than the classical Hotelling T2 statistic in the case of processes where incipient trends are considered. A comparative study based on simulations is presented. The performances of detection are compared to the classical Hotelling control chart in terms of average run length ARL.