Statistical quality control is a method for manufacturers to increase their profits by reducing manufacturing defects. The basic tool of statistical quality control is the control chart. It is used to show graphically if a process is in control or not and if corrective actions must be taken. They also show if there are patterns that are symptoms of assignable causes which can lead to more defects if they are not dealt with.
Interpreting control charts must be done by experts, because the patterns are sometimes hard to distinguish from random statistical noise and sometimes they overlap. Pugh (1989) introduces the use of artificial neural networks for the automated analysis of control charts. In the years that follow, different machine learning algorithms were used with varying results from one study to another. The problem is that the experimental protocols, the datasets and the algorithmic parameters are often incomplete or missing entirely.
This research compares the most widely used machine learning algorithms in control chart pattern recognition: decision trees, random forests, support vector machines and artificial neural networks. In addition to reporting performance results with their confidence intervals, this research provides a clear experimental protocol, the dataset used as well as the algorithmic parameters. A website address to a Git deposit that contains all the code used is given in the appendix.
| Date | 21 Mar 2019 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Alain April (Supervisor) |
|---|
Gagnon, I. (Author),
April (Supervisor),
21 Mar 2019Student thesis: Master's thesis › Master in Engineering: Engineering