Quality and Reliability Methods
What are quality methods?
Quality methods are a group of statistical tools used to 1) ensure consistency among all sorts of processes in a system, such as ensuring a stable pH in a chemical reactor or consistent wait times at a medical clinic, and 2) ensure product measurements are within specification limits. These tools include basic but useful graphs, control charts, and process capability analysis. All process measurements are made with gauges, which are imperfect and can add noise into process measurements. Measurement systems analysis is used to measure and control gauge variation.
What is reliability analysis?
Reliability analysis is the analytical process of determining quality over time, such as predicting failure times for electronic components or computer systems made up of electronic components. Product quality is typically assessed using a lower and upper specification limit. Reliability is quality over time. It is specified by the proportion that should go out of specification by a given time (for example, 9% of the failures by 5,000 hours) or the time when a given proportion should fail (for example, 1,205 hours until 10% of the failures).
When to use quality and reliability methods
Quality methods are used to monitor processes to make sure they stay within historical or customer-driven specification limits. Suppose you work for a company that prints books. You can use quality and reliability methods to make sure your processes are producing consistent output.
One widely used approach is statistical process control (SPC), which includes control charts. Started in industry to make sure that manufacturing processes were producing consistent product, control charts are used in other applications, including healthcare, laboratories, and even finance.
In our book printing example, control charts allow you to track performance and detect any unusual variation. As your company prints books, you measure your processes and use control charts to ensure the processes are stable over time.
Example of a control chart. Control limits are based on the natural variation of the data on the chart. Points within control limits are deemed part of the historical distribution, points outside the control limits are deemed to not belong to the historical distribution.
Process capability analysis helps ensure that the manufacturing process is producing good products that are fit for use. These methods are typically used after the process is fully developed, characterized, and improved. After a book is printed, its quality characteristics can be compared to specification limits or compliance limits before it is passed to shipping. By analyzing the data on all produced books, you can assess how capable the printing process is at meeting specifications.
However, before the process can be improved, it’s critical to trust the measurements that the gauge or measurement system is providing. For example, how consistently do the calipers measure a line whose length is supposed to be 16 cm? Gauge studies are used to understand the important sources of variation in the measurement process: repeatability (variation due to repeated measurements under identical conditions); reproducibility (variation due to changing, but typical, conditions); and bias (differences in measurements from truth). By conducting a gauge study on the tools used to measure book quality, you can determine how much of the total variation is due to the measurement system itself versus the actual process.
It is straightforward to decide if a product is fit for use at the time of manufacturing. But how can you determine if the product quality stays high or if it degrades over time? How can you determine expected failure times? These questions are answered with reliability analysis.
Reliability studies look at time-to-failure data to estimate quantities such as the average time to failure or the proportion of items that will fail within a certain timeframe. Reliability studies account for three different types of failure data: when failure times are known; when they are only known within a time interval; and when they have not yet been observed because the part hasn’t failed yet. Reliability analysis takes into account the partial information when exact failure times are not known. For example, you can collect data on whether the binding of a set of books has failed in order to predict failure times for one book, to compare the reliability of a new binding process to the old one, or for setting warranties.
The chart on the left shows that about 11.5% of items are predicted to fail by 1,825 hours. The chart on the right tells us that by about 3,641 hours, 50% of the items are predicted to fail. Other predictions can be made by changing the value of Life in the Distribution Profiler or Probability in the Quantile Profiler.