Unveiling the Power of Response Surface Methodology in Process Optimization

As we navigate an era of rapid growth, optimizing our time has become increasingly essential. Streamlining analysis processes and maximizing results from every step of production or research are now critical. Imagine a method that not only enhances efficiency in product testing but also clarifies the complex interactions between factors in a process. This is precisely where Response Surface Methodology (RSM) comes into play — a sophisticated technique poised to transform our approach to experimentation and optimization.

When creating top-quality products or reducing production costs without compromising the final result, RSM allows us to explore various possibilities through efficient experimental design, visualize how variables interact, and ultimately discover the optimal point that delivers the best outcomes. My experience in learning and applying RSM could be the key to making significant breakthroughs in data analysis work.

In this article, we will explore RSM’s benefits, the specific data structure we use, how it works, the challenges you may encounter, and examples of its application. Prepare to understand one of the most effective techniques in process optimization and see how RSM can transform the way you tackle complex challenges.

What is Response Surface Methodology?

Response Surface Methodology (RSM) is a statistical method consisting of experimental design and regression components. It is used to optimize and analyze the response (output) influenced by several independent variables.

Benefits of Response Surface Methodology

In short, the benefits of Response Surface Methodology are:

1. Efficiency: Reduces the number of experiments needed to find the optimal combination of variables.
2. Accuracy: Provides a model that can predict the performance of the response.
3. Optimization: Capable of performing processes to find the optimal value (maximum value).

Data Structure

The data structure for this method is built with the help of experimental design. Many data structures can be used in RSM, but here I will briefly explain two common ones:

1. Box-Behnken Design (BBD): A type of experimental design that uses combinations of midpoints between the high and low limits of factors. Advantages of BBD: It requires fewer experiments, making it time- and resource-efficient.
2. Central Composite Design (CCD): A type of experimental design that uses a combination of factorial points (low and high levels), axial points (outside the factorial limits), and center points to build an effective quadratic model.

This article will focus more on the Box-Behnken Design, as illustrated in Fig 1, for three factors and the data matrix. The design consists of Factorial and Center points, with the Center point data collected three or five times, depending on the source. This gives us 15 rows of data (12 points for the factorial and 3 for the central point). When we have four factors, we will have 23 rows of data; with five factors, we will have 27 rows, and so on.

Figure 1. Box-Behnken Design

The optimal point can only be determined in the second or quadratic order in optimization analysis. The first order is limited to understanding linear relationships, specifically how independent variables affect the response, while the Quadratic order estimates where the optimal point lies.

Quadratic order can determine critical points and assess whether those vital points are maxima.

Challenges in Determining the Optimum Point

The optimum point can be determined using a Response Surface plot, as shown in Fig 2. However, we may face dimensional limitations in the plot, which can only represent up to 3 dimensions. This means we can only use one response and two factors. In reality, more than two factors can influence a single reaction. Should we separate the factors, such as using factors 1 and 2 first, then factors 2 and 3, and finally factors 1 and 3? This could result in different optimization points and introduce an error.

Figure 2. Response Surface plot

The solution to this issue is to use the Stationary Point. This point analyzes quadratic functions and determines optimization in high-dimensional space.

Example of Application

One of the applications I have previously worked on is the **Response Surface Model with Comparison of OLS Estimation and MM Estimation**. This research used data from the snack-making process, where observations were made on frying temperature (X1), frying time (X2), and the percentage of peanut flour used in the mixture (X3). The purpose of observing these three variables was to optimize the snack’s expansion capacity.

For a more straightforward application, data will be created with two factors and one response. Many tools are available for this analysis, such as Matlab, Design-Expert, and the R language. This article will implement it using the R language and the “rsm” package.

#install.packages("rsm")
library(rsm)

# example response values
dataRSM <- data.frame(
  x1 = c(-1, 1, -1, 1, 0, 0, 0),
  x2 = c(-1, -1, 1, 1, 0, 0, 0),
  y  = c(552, 549, 521, 595, 504, 502, 505) 
)

# Fitting Response Surface Model
fit <- rsm(y ~ SO(x1, x2), data = dataRSM)
summary(fit)

# Response Surface 3D plot
persp(fit, ~ x1 + x2, zlab = "Response", col = "lightpink")

This syntax will generate the RSM model and Response Surface plot (Fig 3). The model can be evaluated using R-square and adj-R-square.

Figure 3. Response Surface plot in case study

For a case with three factors, the Stationary Point will be used to determine the optimal point due to the limitations of dimensional space.

Explanation: Visualizing and finding the exact optimal point in a three-dimensional response surface can be challenging due to the complexity of the space. The Stationary Point provides a mathematical approach to pinpointing the optimal condition by finding where the gradient of the response surface is zero. This technique helps efficiently locate the optimal settings for the factors involved, even when direct visualization is difficult.

> fit <- rsm(y ~ SO(x1, x2, x3), data = dataRSM)
> summary(fit)

Call:
rsm(formula = y ~ SO(x1, x2, x3), data = dataRSM)

            Estimate Std. Error t value  Pr(>|t|)    
(Intercept)  917.667     30.951 29.6486 8.185e-07 ***
x1            53.625     18.954  2.8292  0.036707 *  
x2           -75.625     18.954 -3.9900  0.010427 *  
x3            24.000     18.954  1.2662  0.261215    
x1:x2         74.250     26.805  2.7700  0.039364 *  
x1:x3          4.000     26.805  0.1492  0.887207    
x2:x3        -15.000     26.805 -0.5596  0.599891    
x1^2        -113.458     27.899 -4.0667  0.009665 ** 
x2^2         -49.958     27.899 -1.7907  0.133350    
x3^2         -22.708     27.899 -0.8139  0.452684    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Multiple R-squared:  0.9129, Adjusted R-squared:  0.7562 
F-statistic: 5.826 on 9 and 5 DF,  p-value: 0.03333

Analysis of Variance Table

Response: y
                Df Sum Sq Mean Sq F value  Pr(>F)
FO(x1, x2, x3)   3  73366 24455.4  8.5092 0.02078
TWI(x1, x2, x3)  3  23016  7672.1  2.6695 0.15868
PQ(x1, x2, x3)   3  54299 18099.7  6.2978 0.03762
Residuals        5  14370  2874.0                
Lack of fit      3   6453  2151.1  0.5434 0.69906
Pure error       2   7917  3958.3                

Stationary point of response surface:
         x1          x2          x3 
-0.04940772 -0.91778145  0.82720934 

Eigenanalysis:
eigen() decomposition
$values
[1]  -19.76071  -35.57803 -130.78626

$vectors
         [,1]       [,2]       [,3]
x1 -0.1365847 -0.4003069  0.9061451
x2 -0.3936901 -0.8174461 -0.4204640
x3  0.9090395 -0.4141693 -0.0459462

Conclusion

In summary, Response Surface Methodology (RSM) provides a powerful tool for optimizing processes by modelling complex relationships between factors. Using RSM, as demonstrated with the RSM package in R, allows for practical analysis and visualization of these relationships, enhancing decision-making and process efficiency.

RSM remains a crucial technique in statistical modelling, offering valuable insights for improving systems and processes across various fields.

We hope this article proves useful and insightful. Thank you for reading.