X-Bar/Range & Sd Charts are **not robust to non-normality**. For X-bar charts typical sub-group sizes are too small for central tendency to come into effect.

The standard X-bar/Range control chart** has many out of control points**, even though the process was in-control. The distribution optimized chart ** has the expected number of out-of-control points** for the number of sub-groups plotted.

The key feature of control charts are the control limits. Sometimes these are supplemented with run tests and warning limits, at the expense of increasing false alarm rates.

For this example, we can see how the key feature of control charts fails. The standard X-bar chart has only two out-of-control points. Operators will search for assignable causes in the wrong area and hence not find causes. Run tests in this example do show some changes corresponding to the steps, but not clearly defined. For other examples even run tests fail. The dynamic plateau chart shows clearly defined changes in the process which will enable operators to better find causes for the variation.

Dynamic Plateau Chart

**Target charts** (not shown here) are used when it is important to aim to a target. One example is for net weight control of packaged consumer goods. Weights and Measures legislation protects the consumer from receiving underfilled product, relative to the declared measure. Manufacturers must thus target to an average that not only ensures weights and measures legislation is met, but also to minimize give-away. Target Charts will warn when there is a significant departure from target, using change analysis technology.

**Process Performance Charts** show changes in the mean and variability on one chart. When individual results are compared with specification limits the analyst can readily determine if non-conformance was produced by changes in the process average, or process standard deviation or both.

Control charts are one of the seven basic tools of quality possibly introduced by Kaoru Ishikawa, who was influenced by W. Edwards Deming. The seven tools of quality were designed at an elementary level after recognizing that statistical complexity intimidated workers. However, times have changed. The work force is better trained and statistical complexity is handled by computers. The time to move beyond the seven basic tools is here. This is the age of Artificial Intelligence, Machine Learning and augmented predictive analytics.

Hybrid charts are a great way to transition from control charts to more modern technology. They provide the best of both worlds, the world of yesteryear and the world of the future

Quality Control, Quality Assurance, Process Improvements are all dependent on correct measurements. If the measurements are erroneous wrong decisions will be made. Measurement error can consist of bias, reproducibility and repeatability. Historically, prior to modern computing power these were quantified through control chart methods. More recently Analysis of Variance was introduced, but at a very elementary level, usually applied to just one measuring device at a time.

Fully utilizing machine power, BIS.Net Analyst has increased the applications more suitable for todays real-world. Included are Dart Boards, Bias (One Equipment, One Appraiser, Multiple Parts), Bias (Multiple Equipment, One Appraiser, One Part), Bias (Multiple Parts, One Equipment, Multiple Appraisers), Gauge R&R (Multiple Equipment, Multiple Appraisers, One Part), Multi Equipment Linearity, Bias (Multiple Equipment, one appraiser), Nested Gauge R&R (Multiple appraisers, multiple equipment, one part). These technologies reflect todays manufacturing environment where different devices are used to measure the same characteristics.

Inferences making is often ignored, which it should not be. It is not uncommon for operators to take a sample of 5 items and then make a decision based on the average. Statistical inference making ensures that sample sizes are selected such that sampling risks are controlled. Until recently inference making technology was constrained by the available technology. This required relying on assumptions and approximations. Confidence intervals and hypothesis testing on the proportion of defectives was subject to errors, resulting in wrong conclusions.

Modern machine powered algorithms have overcome many of the historical limitations, resulting in better decision making. They have made possible previously difficult inference making on process capability and performance indexes.

Multi-modal process performance analysis and stream process control

Used for stream processes such as confectionery extrusion enrobers, beverage filling heads. Equally applicable to analyze combined output of same product and characteristic from different machines and lines.

Multi-modal performance analysis includes performance indexes and theoretical estimates of non-conformance. Each stream can have a different probability distribution

Stream process control uses a stream significance chart to detect when there is stream to stream variation. A profile analysis helps identify which streams are causing problems during periods of changed stream variability.

Acceptance Sampling

Sampling plans used to rely on approximations to calculate hyper geometric probabilities, with considerable error. Today these approximations are used less, but there is still considerable error due to inferior technology used to determine sample sizes and acceptance values. New machine powered algorithms have now made it possible to obtain more reliable sampling plans through better optimization technology.

Process Potential

Conventional Process Capability Analysis, relying on within sub-group variability cannot be applied to non-normal data. Many processes have inherent time to time variation, making the concept of process capability based on within sub-group variability meaning less. Process potential analysis is a robust way of determining what the process can achieve. Machine powered algorithms are used to virtually stabilize the process potential to see what can be achieved if the process were stabilized to reasonable levels.

DISCLAIMER

No technology can be perfect when making estimates about the population, based on a random sample. Machine powered technology is not perfect and can also sometimes get it wrong. However, machine powered technology is a step forward, out of the time capsule.