Improving Manufacturing Quality with Integrated Test and Statistical Analysis

The primary goal of any manufacturing process is to produce high-quality goods that execute on the functions they are designed to perform. Producing quality products creates value and establishes a brand name that consumers feel they can trust.

In manufacturing, some failures – product errors or defects – are inevitable; others are not. The first of the two kinds of failures is an infrequent, random failure that you may not be able to predict or prevent. For example, a computer may come across the manufacturing line with a crack across the monitor, although the computers before and behind the defective computer are correctly manufactured. The second kind of failure may affect an entire batch of products – for example, if every computer that comes across the line has a faulty power supply. Through testing and statistical analysis, not only can you diagnose the cause of this problem, but also you can predict the problem and prevent it from happening.

Why Test?

Producing low-quality products is expensive for the following reasons:

• The amount of materials wasted on products that fail functional or quality tests
• The amount of time spent repairing defects
• The amount of money wasted if a bad product is shipped and must be recalled

Just as you would not buy a cell phone from the same company that sold you a faulty one, when a company produces low-quality products, people begin to mistrust that brand name and are unlikely to buy from that company again. Testing is necessary to ensure quality and to safeguard your brand name.

Mistakes are inevitable on a manufacturing line. The cost of those mistakes depends on when you find them during the manufacturing process. If you catch a mistake early enough, it could cost you only pennies. But if the mistake makes its way through the entire manufacturing process undetected – and especially if it goes out the door – costs could skyrocket into millions.

During the first stage of testing, automated optical inspection systems can identify out-of-place components. It is common to have errors in this stage and you can easily and inexpensively correct these errors. After solder reflow, an automated optical testing system will reinspect the product. Correcting mistakes in this stage is more difficult and the cost goes up by approximately 10X. The third testing stage, in circuit testing (ICT), finds simple electronic problems such as open circuits, circuits that should be soldered down but are not; or short circuits, circuits that are connected but shouldn’t be. Mistakes at this stage, again, are about 10X the cost of mistake in a previous stage. This means that the same mistake that cost you a few pennies in the first stage of testing costs you a few dollars now. In complex products, you be able to test only 40 percent of the components on your board at this stage.

This point is where functional test comes in. Instead of looking for connections and proper placement of board components, this test determines if your product performs the tasks that it is intended to perform. These tests could include trying out the buttons and speakers on a cell phone, or inspecting the display screen to ensure that it lights up when the dial pad is in use. When failures occur at this stage, it can be difficult to pinpoint the cause of the failure. Statistical analysis expedites diagnosing problems found in functional test by pointing you to the most likely cause of the product failure. For example, if your display does not light up, the reason for the defect could be the battery, a bad chip, or the screen. Using statistical analysis, you would learn that, based on your previous tests, in 80 percent of the cases that the display was not working, the battery was the culprit. You then would know to test the battery first, saving time and money. Statistical analysis trains the system or the engineer regarding which fault to check for first.

If each function of your product passes the functional tests, the product is then put together and tested as a whole in a system test. If each of the functions is successful in the previous stage, but the product as a whole doesn’t work, the cost of replacement or repair goes up dramatically. Here, a component that may have cost one cent to replace could cost $100 to replace. The only more expensive time to discover a defect is after the product has been shipped to customers. Between public notices, recalls, replacements, repairs, and damage to consumer trust in your name brand, an error here could cost you millions of dollars.

Test and Test Management Software

Producing quality products that are free of flaws requires testing. For up-to-date manufacturing enterprises, engineers need to implement a modern test system that is both cost-effective and flexible enough to meet their individual application needs. They also need a modular testing architecture that encourages code reuse and is set up to expand easily with the enterprise. They need a test executive that is scalable enough that everyone in the company can use it, even if they program in different languages.

National Instruments addresses these needs with TestStand, a ready-to-run test executive that organizes, controls, and executes your automated prototype, validation, or production test systems. TestStand is completely customer-defined, so you can modify and enhance it to match your specific needs.

TestStand is compatible with all leading test programming languages, including LabVIEW, Measurement Studio, Visual Basic, Visual C++, and HT-Basic. TestStand also executes test code compiled as a DLL, ActiveX Server, or Windows executable, so you can easily handle a variety of test programming environments and legacy code. TestStand fully integrates with the LabVIEW and Measurement Studio programming environments so you can generate code and perform full debugging, including stepping into your test programs directly from TestStand.

TestStand provides the performance you need to keep up with increasing demands for faster test times, quicker test development, and more intelligent data sharing and test sequencing. TestStand uses a speed-optimized, multithreaded sequence engine capable of running parallel sequences. TestStand also features a flexible framework for sharing variables, so you can test more products faster.

Statistical Analysis

Quality is paramount in the manufacturing process, the target being to increase yields and lower scrap costs while meeting or exceeding budget goals. Testing is one important part in creating quality products. The second part, which reduces the costs associated with failures and repairs in the manufacturing process, is statistical analysis.

Manufacturing operations that use statistical analysis to monitor the daily production of their machines on the manufacturing line can increase their product yields, improve their product quality, and boost their profit margins. With statistical analysis, the monitoring produces charts that reveal trends from which you can detect serious problems before they occur. In addition, engineers can study these trends to reveal the sources of the problems and remedy them before the problems cost the company thousands of dollars.

For example, an excessively hot soldering oven could be responsible for the damage in a batch of cracked computer screens. With statistical analysis, you can see the trends of production on a chart, which would show signs of the excessive oven temperature before it got hot enough to crack the computer screens and cause unnecessary scrap.

There are two kinds of statistical analysis. The first, statistical process control (SPC), performs point-by-point analysis – it analyses each product as it comes through the line. This kind of analysis reveals trends in real-time. You can detect a problem as it is happening and correct that problem immediately. The second, statistical quality control (SQC), runs an entire batch of products and then analyses the entire batch after it has run.

Northwest Analytical specializes in statistical analysis and is dedicated to providing essential analytical tools to help engineers understand processes and improve quality. Their Quality Analyst software provides the statistical process control to monitor and report both time-based parametric and attribute (pass/fail) data from any ODBC-compliant database or from ASCII files.

Integrating Test Systems and Statistical Analysis

By combining the best aspects of test management and statistical analysis software, it is now possible to "close the loop" on process data and deliver flexible factory management systems that maximize test coverage when processes are tracking out of control and minimize test coverage when processes are under control. This process maximizes throughput and reduces the cost of test.

National Instruments TestStand test management software and Northwest Analytical Quality Analyst statistical analysis software bring this efficiency and convenience to your systems. Together, these programs integrate easily into your system to bring you the best of test management and statistical analysis software. This powerful combination closes the loop on process data and delivers a flexible factory management systems that helps you increases yields, maximize test throughput, minimize manufacturing defects, and reduce your cost of test.

By integrating test management systems with statistical analysis, you create test systems that work more efficiently and produce quality products. And these integrated systems are automated, which means that you do not have to wait on a technician to start applications. The tests will automatically run each time a new board is produced. They keep your manufacturing process in control by revealing early signs that a process is not working correctly.

With these integrated systems, you can build intelligent test systems capable of testing that adapts to the condition of your manufacturing line. This kind of adaptive testing inspects what is likely to fail, instead of wasting time testing items with flawless histories. For instance, if there is a part of your product that never fails, it may not be necessary to test it 100 percent of the units. For items such as this, you could test every tenth unit to maximize throughput. Throughout the testing cycle, the system continually creates a history of each product and knows which parts of a product are prone to errors and need to be tested each time. The system can adapt to the current situation. For example, if a usually trouble-free part of the system begins to fail tests, the testing frequency automatically goes up to 100 percent. As it begins to pass the test consistently, the frequency goes back to its original testing frequency.

Finally, these integrated systems save you money. Catching errors before they ruin entire batches of products reduces scrap and increases productivity of your manufacturing process. Statistical analysis guides repair, reducing the time it takes to repair defects by pointing you to common causes for failures. Statistical analysis also frees you from having to take machines off line for periodical repair. Instead, you can take the machine off line only when the trends suggest that it needs repair.

Calculating the Cost

Without the integration of a test system with built-in quality analysis, the production system is, in effect, "out of control," with no direct feedback loop in place. This condition is found in many organizations where all of the analysis takes place after the production of a batch of products. We can better understand the cost savings of integrating test and analysis by examining a fictional manufacturing line before and after the introduction of an integrated solution.

Before
If we imagine that we are producing a batch of 100 communications subsystems for a military product, we may find the manufacturing line, the product, or components develop a fault while we are mid-batch. For example, the solder paste template may become misaligned, or incorrectly programmed FPGAs may be mixed into the production process. The net effect of these faults may be that 80 percent of the products fail in the batch. Because of no feedback loop being in place, this issue is not properly understood until the batch has completed the manufacturing process and the results subsequently analyzed.

The cost of such a problem is often highly understated. In this case, if we imagine that the value of the board is $3,000, made up of $2,500 of components and $500 for the manufacturing process, an 80 percent failure rate would constitute $240,000, plus lost revenue at an average 3X, which would bring the total to approximately $1 million.

Because this amount of scrap is intolerable, these boards would require repair. However, this repair process does in itself have a variable cost in terms of man-hours required, replacement components, and the loss of irreparable products. The time taken to repair each of these products will vary, but if we again assume that they will average four hours, or approximately $200 each, plus component costs of up to $500, we would have a repair cost of some $56,000, providing it is possible to repair all of the bad boards.

It is normally expected that a number of boards will be irreparable, because of damage in the production process or the repair process, or excess heat from unsoldering and resoldering components. If this number is even as high as five percent of the failed products, then we would see a total of four scrapped boards, which would total an additional $12,000. This amount brings the total cost of the bad production run to more than $65,000.

After
By introducing the integrated test and analysis solution we add a near-instantaneous feedback loop and the process becomes "in control." With this feedback loop, the system can review the test data to look for any repetitive fault discovered in the testing process and calculate to see if it is statistically relevant. This relevancy acts as a filter to ensure that the manufacturing process doesn’t get shut down because of spurious data. In our example, the linking of test and statistical analysis would show that two consecutive boards that both failed with identical problems would be a statistical anomaly and the system would raise an alarm on the manufacturing line. Because the alarm was raised as early as possible, the number of boards incorrectly manufactured would be at a minimum and hence the cost of the repairs, far lower. For example, the problem items might be limited to the two boards that originally failed the tests, plus potentially up to 10 counting others that were in the actual manufacturing process. Following the calculations from above, the repair costs would be in the area of only $7,000, with potential savings over $58,000.

While this is a very simplistic example, it illustrates the very real costs associated with the repair of complex products that can be eradicated when intelligent systems are introduced to the process.

Conclusion

Integrating National Instruments TestStand and Northwest Analytical Quality Analyst, companies can react to issues on the production line much faster, reducing wasted materials and time spend on repairs. With test management and statistical analysis integrated into an automated system, organizations can target problem components and processes in less time with more accuracy, reducing the cost of manufacturing quality products.

Reader Comments | Submit a comment »