Analisis y diseño de experimentos avanzado_DSSMP1_introduction.pptx

masterclass of experiment desing and analysis introduction UTFSM

1. Statistical analysis of processes Introduction Luis Bergh Chemical and Environmental Engineering Department University T. F. Santa Maria Valparaíso, Chile

2. Introduction General recommendations: Use the non-statistical knowledge of the process (problem). Use design and analysis as simple as possible Experimentation is generally iterative Recognize the difference between statistically significant and practical

3. Introduction. Affirms H1 Hypothesis 1 Consequence from H1 Data Induce Modify H1 Objective of Statistics Make it more efficient Process of Learning guided Scientific research

4. Window Hi Consequence H1 Design New Data Available Data Hi+1 STATE OF NATURE Noise Guess object

5. Game features What does it take to play well? Object-matter knowledge and intellect Knowledge of strategy ( e.g. guessing a number from 1 to 100) Strategy is equivalent to knowledge of statistical methods 1 is more important (essential) 2 helps to be more efficient (lower cost) No noise (Yes or No) Noise equivalent to: maybe, almost always, rarely, etc....

6. Analogy of the game If noise is present , they are needed: Efficient experiment design methods (answers that answer the question unambiguously and as little affected by experimental error as possible). Sensitive data analysis, what can be legitimately concluded from the data

7. Example A new bubbler is available on the market, whose characteristic is to generate bubbles whose size distribution is smaller in diameter than those commonly used. It is postulated that it would improve the recovery of fine ore . Experimental Design : Replaced in a flotation equipment There are no data on its use in similar plants

8. Example Basic objectives in the comparison of treatments. Estimate difference (between whom? The samples? The population?) Measure the accuracy of the estimate (how far does the difference extend?) Population: all data Sample: a few data

9. Shall we change bubblers ?

10. Experiment problems Compares results obtained at different times on the same equipment Different feeding (F, x, %s, dp , dosage...) Different operating conditions (Faire, z, ...) Confuses the effect of the bubbler with changes in other variables that also affect R Experimental error variance increases Reliability intervals are increased Decreases the sensitivity of observing the possible differences to be demonstrated

11. How to improve the experiment? Reducing the sources of variability of R Different feeding (F, x, %s, dp , dosage...) Decrease variation of F, x, %s, dp , dosage ... The result would be valid only for that condition Different operating conditions (Faire, z, ...) Maintaining the same conditions Faire, z, ... The result would be valid only for that condition

12. How do we analyze the data? Keeping some of the other variables fixed (partial elimination of sources of variability)

13. How to improve the experiment? Reducing the sources of variability of R Different feeding (F, x, %s, dp , dosage...) Use the same power supply in parallel equipment, one with each type of bubbler , varying F, x, ..... The result would be valid for more conditions Different operating conditions (Faire, z, ...) Replicating the same time-varying conditions, in parallel The result would be valid for more conditions

14. How do we analyze the data? If we keep the variability, but operate in parallel, we analyze the difference to the same condition

15. Moralities A bad design delivers data that do NOT contain the information needed to test the consequences of the hypothesis, regardless of the method of analysis employed. DESIGN vs. ANALYSIS What is more important? Knowing how to analyze data or how to design experiments? The design defines the information that will exist in the data, the method of analysis will make the extraction of information more efficient.

16. Main difficulties in the analysis The most common difficulties when designing and analyzing an experience are: 1. Experimental error (which includes: measurement error, effect of other variables simultaneously, other sources...). 2. Confusion of correlation with causation. 3. Complexity of estimated effects.

17. Correlation vs. causation Y population in Strasbourg X number of storks H1: Babies are brought by stork Y and X are both functions of time, they show correlation not causation.

18. Complexity Study of the effect of alcohol and coffee on reaction time ( Tr ): 1 glass of pisco increases Tr in 0.45 s 1 cup of coffee decreases Tr in 0.20 sec. Are the effects linear and additive? Tr = average value + 0. 45*vpp - 0.20*tc