Linear Regression for Manufacturing Process Analysis

From Scatter Plot to Regression Model

In the previous lesson, you saw how data points on a scatter plot can form a linear pattern. Simple linear regression is the technique we use to draw a single straight line that best represents those data points. This line is called the regression line. Its purpose is to describe the relationship using a simple mathematical equation.

This equation is a core concept in Six Sigma and is used extensively in the Analyze phase to formally test the hypotheses you generate about your process.

As the article explained, the equation for a simple linear regression model is:

$Y={\beta }_0+{\beta }_{1}X+ϵ$

Let's break this down in a manufacturing context:

• $Y$ is the Dependent Variable: The process output or quality characteristic you are trying to predict (e.g., tablet hardness, coating thickness, shear strength).
• $X$ is the Independent Variable: The process input you are manipulating or observing (e.g., compression force, nozzle pressure, curing time).
• ${\beta }_0$ (beta-naught) is the Intercept: The predicted value of $Y$ when $X$ is zero.
• ${\beta }_1$ (beta-one) is the Slope: This is often the most important part. It tells you how much $Y$ is expected to change for a one-unit increase in $X$.
• $ϵ$ (epsilon) is the Error or Residual: The difference between the actual observed value of $Y$ and the value predicted by the line. No model is perfect, and this term captures that "unexplained variation."

The goal of regression analysis is to find the values for ${\beta }_0$ and ${\beta }_1$ that make the line fit the data as closely as possible, which means minimizing the total error.

Building a Simple Linear Regression Model in Excel

Given your preference for practical application, we'll use a common tool you likely have access to: Microsoft Excel. Excel's Data Analysis ToolPak can perform regression analysis quickly.

Let's work with a practical manufacturing scenario. Imagine you're working in a chemical plant and you suspect that the pressure in a reactor vessel (Pressure Level) has a significant impact on the final Purity of the product. You've collected data and created a scatter plot, which showed a strong positive correlation. Now, you want to build a model.

Running this tool will generate a summary output table containing all the key statistics we need to interpret our model.

Interpreting the Regression Output

The Excel output can seem overwhelming at first, but it's structured into three main parts. Understanding these parts is essential for any Green Belt or Black Belt.

We'll use a comprehensive guide to walk through the interpretation. The example in this guide is about a manufacturing process where Temperature is predicted by Pressure and Fuel Rate. For our simple linear regression lesson, let's imagine we are only modeling the relationship between Pressure (X) and Temperature (Y).

Let's dissect the three key tables from the output you just read about.

1. The Regression Statistics Table

This table tells you about the overall "goodness of fit" of your model.

• R Square ($R^2$): This is a critical value. It represents the proportion of the total variation in the dependent variable ($Y$) that is explained by the independent variable ($X$). It's the square of the correlation coefficient ($r$) we learned about previously.
  • An $R^2$ of 0.85 means that 85% of the variability in your output (e.g., Temperature) can be explained by the changes in your input (e.g., Pressure). In manufacturing, a high $R^2$ is desirable, but even moderate values can indicate a meaningful relationship.
• Standard Error: This is the typical distance between the observed values and the regression line, measured in the units of $Y$. A smaller standard error means your predictions are, on average, closer to the actual values.

2. The ANOVA Table

This table tests the overall significance of your model.

• Significance F: This is the p-value for the entire model. It answers the question: "Is my model, as a whole, better at predicting the outcome than just using the average of the Y values?"
  • The Rule: If Significance F < 0.05, we reject the null hypothesis and conclude that our model is statistically significant. It means your input variable $X$ has a genuine, non-random effect on the output variable $Y$.

3. The Coefficients Table

This table provides the details for your regression equation and is where you'll spend most of your interpretation time.

• Coefficients (Intercept and Slope): These are the values for ${\beta }_0$ (Intercept) and ${\beta }_1$ (the variable's slope). You use these to write your final prediction equation.
  • For example, if the Intercept is 10 and the coefficient for Pressure is 4.79, your equation is:
    Predicted Temperature = 10 + 4.79 * Pressure
  • The slope of 4.79 is highly valuable: it means that for every one-unit increase in pressure, the temperature is predicted to increase by 4.79 degrees.
• P-value: Each coefficient has its own p-value. The most important one is for your independent variable ($X$). This tests the null hypothesis that the slope (${\beta }_1$) is zero.
  • The Rule: If the p-value for your X-variable < 0.05, you can conclude that the variable has a statistically significant effect on $Y$. Its slope is not zero, and therefore it is a meaningful predictor.

Using the Model for Prediction

Once you have built your model and confirmed that it's statistically significant (low p-values) and explains a reasonable amount of variation (a decent $R^2$), you can use it to make predictions.

Using our example equation: Predicted Temperature = 10 + 4.79 * Pressure

If you want to know the expected temperature when the process pressure is set to 20 psi, you simply plug in the value:

Predicted Temperature = 10 + 4.79 * (20) = 10 + 95.8 = 105.8 degrees.

This predictive capability is what makes regression a powerful tool in the Improve phase of DMAIC. You can use the model to determine the optimal settings for your process inputs to achieve a desired output.

Conclusion

In this lesson, we transitioned from simply identifying a relationship to building a predictive model. Simple linear regression provides a mathematical equation that describes the linear relationship between a key process input (X) and a critical quality output (Y).

Key Takeaways:

• Simple linear regression models a relationship with the equation $Y={\beta }_0+{\beta }_{1}X$.
• Interpreting the model output is a systematic process:
  1. Check Significance F (< 0.05) to ensure the model is statistically significant overall.
  2. Check the p-value for your X-variable (< 0.05) to confirm the input is a significant predictor.
  3. Examine R-squared ($R^2$) to understand how much of the output's variation the model explains.
  4. Use the coefficients to write the prediction equation and understand the specific impact of your input variable.
• A validated regression model is a powerful tool for process prediction and optimization.

Preview of the Next Lesson:

Simple linear regression is powerful, but manufacturing processes are rarely this simple. Often, an output is influenced by multiple inputs working together. What if both temperature and pressure affect product purity? In our next lesson, we will expand on today's concepts to learn how to develop a multiple regression model to predict a quality outcome from several process parameters.