mqr-quality 0.1.0

Manufacturing quality improvement toolkit

mqr

Tools for learning and practising manufacturing quality with python.

Installation

pip install mqr

See examples directory in the source code for notebooks with sample code.

Introduction

This library is called mqr which stands for manufacturing quality and reliability. Most of the tools are related to quite general statistical methods that are commonly used in fields outside manufacturing.

Most of this library falls into one of two categories:

1. Code that automates commonly used plots and tables. These are elementary, in that they don't require much special knowledge of the subject area and are easy to interpret, but they require an understanding of libraries like numpy, pandas, matplotlib, etc, and quite a few lines of boilerplate code. The goal of including these in the library is to make common plotting and descriptive activities fast.
2. Code that provides a uniform interface to functionality in other libraries. Existing libraries provide good coverage of hypothesis tests, particularly numpy, scipy and statsmodels. However, because the tests come from statistical literature, and are organised varously by name, purpose or application, they can be difficult to navigate, especially for students with limited experience with python. Further, each library has a different interface. The purpose of wrapping these tools is (1) to organise tests by goal or application (eg. tests on means, tests on proportions), and (2) to provide a uniform interface that is easy for students to navigate and use in jupyter notebooks. The doe module is another example. It provides an interface for designing experiments. It can be used with pyDOE3 (and provides convenience functions for that), but also provides extra features that help with the practicalties of designing experiments, collecting and analysing data. For example, experimental designs can be easily built up in a few lines of code by composing smaller sets of runs.
3. Several example notebooks showing how to do common quality tasks. The examples (in the examples directory) provide a summary of the functionality and arrangement of this library, and also demonstrations of some common activities in quality improvement. For example, the basic-inference notebook shows how the interface to hypothesis tests are arranged, while the probability notebook gives examples of calculating useful probabilities using the very good scipy library.

Overall, the code is intended to reduce the burden of understanding python and its various libraries, so the user can focus on understanding processes, designing experiments, and analysing results.

A note on wrapped code: excluding a few tests that we implemented (because we couldn't find an existing implementation), the confidence interval code and hypothesis tests call through to other libraries. We think there are benefits to wrapping the interfaces when using a notebook. However, if you use these tests outside of a notebook, for example in scripts or automated routines, or if you need more advanced functionality that we didn't expose in our interface, you should look up the original functions and then call the statistical libraries directly.

Code Organisation

The code is organised into the following modules.

• summary: summary statistics
• process: describing and presenting processes, including capability
• inference: basic confidence intervals, hypothesis testing and sample size calculation
• anova: tools for working with ANOVA calculations and results (statsmodels provides the regression)
• msa: measurement system analysis; an automatically constructed ANOVA for gauge reproducibility and repeatability studies
• control: plots for showing processes under statistical control
• doe: tools for designing experiments
• plot: various plots (below)

The plotting module has routines for generating a range of plots. The plots are mostly arranged like this:

• plot.Figure (context manager): create subplots (calls matplotlib.pyplot.subplots)
• plot.confint (function): visually show the process of an hypothesis test on the mean of a data set
• plot.ishikawa (function): draw an Ishikawa (fishbone) diagram
• plot.summary (function): draw a histogram, box plot and confidence interval with shared x-axes
• plot.grouped_df (function): draw data from a dataframe grouped by column and drawn adjacent to each other, left to right
• plot.control (module): draw X-bar and R charts
• plot.correlation (module): plot a detailed correlation matrix, including histograms, scatter plots with best-fit lines, and confidence intervals on coefficients
• plot.msa (module): plot the results of measurement system analysis, including a GRR summary
• plot.probplot (module): probability plots of subsets of data, with shared summary statistics
• plot.process (module): graphically represent a process, including capability
• plot.regression (module): draw residuals from a regression analysis

Inference module

The inference module is mostly an interface to other libraries. The module contains functions that calculate sample size, confidence intervals and hypothesis tests.

The functions are arranged with the following naming scheme.
Parametric tests:

• sample size: mqr.inference.<statistic>.size_<sample>(...)
• confidence interval: mqr.inference.<statistic>.confint_<sample>(...)
• hypothesis test: mqr.inference.<statistic>.test_<sample>(...)

Non-parametric tests have the same form, but are in the nonparametric module:

• hypothesis test: mqr.inference.<statistic>.test_<sample>(...)
• etc.

Where:

• <statistic> is the statistic of interest, for example, "mean", "correlation" coefficient, etc.
• <sample> is a description of the samples involved in the calculation, for example "1sample", "nsample". Some routines don't have a sample description.

For example, hypothesis tests that deal with the difference between the means of two unpaired samples are at:
mqr.inference.mean.test_2sample(...)

Plotting module

The library never renders a plot automatically, but instead expects users to provide axes to draw into. This choice means the plotting libraries have no side-effects, and it also allow the to change layout and plotting backend easily.

In the example notebooks, plots are wrapped in a with plot.Figure(...) context manager, which creates figures, shows them, then closes them automatically. The (fig, ax) that the context manager creates is the return value of matplotlib.pyplot.subplots(...). The context manager reduces the boilerplate code required for the user (especially those unfamiliar with matplotlib) to show a plot, and helps with a few other activities, like changing backends and saving files. It is possilbe, for example, to quickly switch from showing plots in a notebook to writing them into a backend that produces images for Word or PGF/TikZ for LaTeX.

Of course, you can always create and manage axes directly, by calling ax, fig = matplotlib.pyplot.subplots(...) and passing ax to the plotting routines.

License

This package is provided under the BSD License (3-clause).

Credit

Copyright (c) 2024 Nikolas Crossan, Kevin Otto

Supported by the University of Melbourne
Department of Mechanical Engineering