Strategies for Robust Designs

We have discussed in the previous sections how to define and analyse a factorial design in order to estimate the effect of a series of variables on the response 

It was discussed above that two types of factors can influence our process or system controlled factors design factors in Taguchi s terminology) and uncontrolled factors noise factors. The latter are inherent to the experimentation and can only be estimated by replication of runs. If the variability between replications is too large, any conclusion drawn from our study may have no meaning at all. Further, their variability can be as important as the mean of the replicates. Noise factors must be identified properly and, if possible, simulated in our experimentation. Sometimes noise factors are actually uncontrolled and, in such a case, we must be able to simulate them by means of some alternative parameter controlled during the experiments. For instance, temperature inside an oven can be a noise factor, but we can measure it at different locations, and accordingly, the effect of temperature inside is simulated by a location factor. 

Any experimenter is usually interested in obtaining an optimal response (a nominal value, a maximum or a minimum response) which is related directly to changes in the levels of the factors, but he/she needs also to guarantee a constant quality and, as a consequence, a minimum variability in the response. The main objective is then to control, although partially, the noise factors and obtain a system which is insensitive robust to them. In such a way, we can discard the noise factors in future studies. In short, a robustness design identifies the levels of the experimental factors that reduce the effect of the noise factors and consequently minimise the variability in the response in a simple and economical way. 

---

The strategy for robust design experiments that will be considered in Section 2.3 is based on the statistical techniques associated with response surface methodology. This section will give an overview of response surface methodology, presenting some of the more common experimental designs that have been developed in this area. 

The landscape of research in solar photocatalysis has been rapidly changing in recent years, with a flurry of activity in the development and analysis of catalysts for water oxidation and fundamental studies of photocatalysis based on semiconductor surfaces. Significant effort is currently focused on the development of more efficient catalysts based on earth-abundant materials and various strategies for the design of molecular assemblies that efficiently couple multielectron photoanodic processes to fuel production. The outstanding challenge is to identify robust materials that could catalyze the necessary multielectron transformations at energies and rates consistent with solar irradiance. 

Care for people with dementia is demanding of resources, while the outcomes of care are uncertain. To date, the economic analyses of care strategies have been limited to new drug therapies for people with Alzheimer s disease. Full economic evaluations to compare both the costs and consequences have only been conducted for one of these dmgs, donepezil. However, problems with the design and data used in these studies mean that they do not provide robust evidence to determine appropriate management strategies for dementia. 

Section 5.3 described a number of alternative design and implementation strategies for near-infrared analyzers, suitable for operation in a process analytical environment. However, none of these analyzers can operate without a robust, maintainable and repeatable sampling interface with the process sample under consideration. In addition to this question of the optical interface to the sample, there is a whole wider area of concern, which is how far the particular sample interface is representative of the sample in the process as a whole. This complex issue is not addressed here, and is dealt with separately in Chapter 3. 

In future research, the relation between the retrieved optical intensity distributions and the design of the nanometric structures may come to be xmderstood, including their environmental conditions. Such insights should allow us to propose, for instance, an optimized strategy for implementing nanophotonic codes, or a strategy that is robust against errors that possibly occur in the fabrication and/or retrieval processes (Naruse et al., 2009). 

---