How can we control production processes today?

• Statistical process control is one of the keys to quality assurance in production.

• What seems so simple is a challenging task.

• This is especially true when taking into account the large amount of data generated during the process.

Statistical process control (SPC) has long been an important method for ensuring high product quality. However, the complexity of modern production processes such as electronics manufacturing does not conform to the basic assumptions of SPC with regard to process stability. This makes traditional SPC largely worthless as a quality indicator, especially when combined with the increasing volume of data. Different approaches are therefore needed to identify and prioritise opportunities for improvement. These must be in line with the Lean Six Sigma philosophy, however, and allow greater leeway than SPC.

Basic limitations

SPC continues to play an important role with Original Equipment Manufacturers (OEM). It can be found in continuous production processes and in the definition of limit values. The same applies when it comes to detecting faulty process parameters that influence quality.

In theory, limit values of this type help to visualise quality changes. There is a difference between theory and practice, however: a basic assumption is that by using SPC, the causes of quality fluctuations can be eliminated from the process or at least allowed for. This means that all remaining process variations have specific causes. The parameters you should be concerned about are those that are beginning to drift.

An electronic product today can contain hundreds of components. It will undergo various changes, for example because components are no longer available or there are different assembly variants. The product is tested at various stages during the assembly process, different firmware versions are received or changes in environmental conditions are experienced.
 

High dynamics

An actual example of this is the company Aidon, a manufacturer of smart metering products. An average production batch for the company has the following characteristics:

• It contains 10,000 units.
• Each unit consists of over 350 electronic components.
• In each production batch there are more than 35 variants of the product.

This means that, on average, a new product is built or a new process occurs every 285 units. In addition, there are changes to the test procedure, the fixtures, test programmes and other components. This means a process change after every ten units, as an estimate. Or to put it another way: around 1,000 different processes result from the production of a single batch. How can it be at all possible to identify and eliminate changes in the process? What is to be done about this?
Even if you were able to identify the changes in the process, how would you implement the corresponding alarm system in production? An SPC-based method developed by the Western Electric Company as early as 1956 is known as Western Electrical Rules or WECO. It defines certain rules under which a deviation justifies a process investigation - depending on how far the current value is from standard deviations. One problem with WECO, however, is that - in principle - it triggers a false alarm on average every 91.75 measurements.
 

62 false alarms a day

Suppose you have an annual production of 10,000 units. Each piece is tested through five different processes, and each process includes an average of 25 measurements. Combining these, you get an average of up to 62 false alarms a day, assuming 220 working days per year.

In summary: assuming that SPC and WECO enabled you to eliminate the causes of the frequent variations, you would still get 62 alarms a day. People who get 62 false-alarm emails a day will soon start to ignore them and so miss potentially important alarms. SPC-savvy users will probably now argue that there are ways to reduce the number of false alarms through new and improved analysis methods.

Even if we were able to reduce the number of false alarms to five a day, could this really be a strategic alarm system for our production process? Can SPC provide a system that production managers can rely on when they bring the actual process dynamics into the mix?
 

Analysis of Key Performance Indicators

What most people do in this situation is make assumptions about a limited set of important parameters and then monitor them carefully. This is an attempt to separate the wheat from the chaff. These Key Performance Indicators (KPIs) are usually recorded and analysed in the manufacturing process after several units have been combined into one system.
One obvious consequence of this is that problems are not detected at the point where they occur. The origin of a problem could, for example, be one of the upstream components that were produced a month ago in a batch of 50,000 units.

The 10x rule for a manufacturing process states that for each step in which a defect is just passed onwards, the cost of correcting it increases by a factor of ten. A system-level failure may require technicians to completely disassemble and then reassemble the product, which in itself creates possibilities for new defects. If the failure first occurs at the customer's premises, the impact on costs can be catastrophic. There are numerous examples of companies that, due to huge recall actions, have had to file for bankruptcy or take appropriate steps to protect themselves. A current example is the insolvency of the Japanese automotive supplier Takata following a massive recall of defective airbag components, which could involve more than 100 million units.

According to the standards of modern approaches such as Lean Six Sigma, one of the major shortcomings of SPC is that assumptions are made about where the problems come from. This is a consequence of the basic assumption that the process is stable. But as mentioned above, this is not the case, as dynamic factors influence the production process. Trending and tracking of a limited number of KPIs amplifies this error. This in turn results in a number of improvement initiatives that may fail to focus on the most urgent or cost-effective problems.

Modern approaches

All the points mentioned above are taken into account in modern methods of quality management.

In electronics production, this begins with transparent detection and monitoring of the First Pass Yield (FPY), known as the real FPY. Real means that any kind of fault must be taken into account, even if the fault occurs due to a cable not being plugged in, for example. Any test that follows an unsuccessful test is a waste of time. The resources that would be needed for this can be better used by the company elsewhere. The real FPY is probably the most important KPI, but most OEMs cannot identify it.

Real-time insight

If you know your FPY, you can split it in parallel across different products, product families, plants, fixtures or operators. You view this data in real time as dashboards. This gives you a powerful overview, known as the Captains View. In this way, you can quickly see the area(s) where production output is insufficient, and so intervene on the basis of economic considerations. The fact that these insights are provided as live dashboards for all participants also contributes to improved quality accountability. A rule of thumb for the dashboard says: if the correct information is not passed on, there will be no reaction. We often don't have the time to search for the essential data again and again.

It is vital that you are able to quickly get a Pareto view of your most common defects in all these dimensions. It might be useful to use the classical methods of SPC to capture more details. You now know that you are using them in the right place. This is a decision that is not based on conjecture and speculation. You suddenly find yourself in a situation where you can prioritise and initiate measures based on a realistic cost-benefit ratio.
 

The integrated repair data

It is important that the repair data is recorded in your system. It is not enough that this data is stored exclusively in a Manufacturing Execution System (MES) or an external repair system. Integrated repair data provides context-related information, which offers you many advantages. They improve the analysis of causes, for example. From the personnel's point of view, this repair data and repair information can also define whether a product needs to be re-tested because, for example, normal process fluctuations have affected the measurement, or whether the product needs to be removed from the production line and repaired as intended. But don't be under any illusions: it often happens that products have to be tested several times within an hour.

In summary, it can be said that quality-enhancing measures can only be achieved through sound decisions. If you have a data management system that gives you a complete picture of the situation, you can optimise your product and process quality and therefore the success of your company as well.

You can only repair what you test.