How much inspection data should you save?

One difference between inspection and conventional test is the nature of the data the two processes generate. Test results can be stored as text files, spreadsheets, and the like. Inspection systems can offer similar data, but they can also measure component dimensions and solder volumes. They can even provide images of failed components or other devices so a manufacturer can trace production problems out to the field.

Images represent a new dynamic in data storage, because they take up far more space than raw data files. Pass/fail decisions can be made by pattern-matching algorithms that compare two images or they can be made by a system comparing measurements on an image to a predetermined ideal.

These images reflect worst-case results from a series of production runs. Courtesy of Agilent Technologies.

Jeff Bishop, automated optical inspection (AOI) product manager at Agilent Technologies in Raleigh, NC, put it this way: “From our perspective, it doesn’t take much longer to create a report with 70 variables than 7, but so much data often becomes overwhelming and therefore goes unused. The challenge at inspection is to provide the appropriate data at the most opportune time to help users make quick and effective decisions.” So, how do you do that?

Keeping every data point is like saving old magazines in case you want to read some of the articles someday. As long as the collection is small, you can probably find what you want. In a room full of magazines, however, anything that might interest you is lost in the “noise” of the surrounding clutter. If you save only individual articles, the volume of your stash increases much more slowly. And purging the pieces that you no longer want preserves your ability to locate a specific item from what remains.

Most engineers would find pruning a magazine collection far less traumatic than parting with “critical” data. And the definition of “critical” can vary dramatically depending on the application and the customer.

Product liability concerns, for example, require that a manufacturer maintain verifiable documentation that a board contained no defects when it left the factory. And legal considerations require that makers of medical and military equipment be able to trace every component in a system or subsystem back to its lot number and its original source.

This level of traceability can actually provide an economic benefit in some cases. Bishop offered an example. “Proper data tracking could reduce the cost of automobile recalls. Say an automaker discovers a bad batch of ICs in an engine-control module. Right now, they [automakers] recall every car containing that module for replacement. But if they could trace the bad batch to particular board serial numbers, and if they knew the VIN numbers of the vehicles that received those boards, they could recall only the affected vehicles, drastically reducing the recall load and the associated costs.”

Manufacturers of handheld devices and other commodity products may not need such tight controls. They are often more concerned about analyzing trends in the results so they can adjust their processes to increase yields and reduce rework costs.

Conventional test stations generally get boards from manufacturing delivered in batches. A production run from Tuesday may not reach in-circuit test until Wednesday. The lag between assembly and test reduces a manufacturer’s ability to respond quickly to modify the production process and lower the defect rates.

Inspection, on the other hand, often occurs in-line. Bishop said that performing automated optical inspection (AOI) of solder paste after the printer step “gives you volumetric or area data in real time. You can compare the height of the deposit, its x-y position, and potentially its slope or any detected deformity against a known-good ideal. Some of that data may be useful, and some may not. Analysis could reveal that at a specific time, the paste step deposited insufficient solder volume on some (or all) solder pads. Perhaps the stencil needed cleaning. Appropriate corrections could reduce the future defect rate. Once you make the changes and achieve the corresponding improvement, there is generally no technical reason to retain all of the raw data. You track only trends, means, and extremes.”

Looking at trends can provide more useful information than looking at the raw data. Courtesy of Agilent Technologies.

Bishop continued, “If you manufacture products in high volumes, processes change frequently in response to inspection and test data and to changes in the product or customer requirements. Except for legal considerations, the raw data loses its value after a week or 10 days. Results from pre-reflow inspection, for example, become much less valuable after the board has gone through the reflow oven.”

For contract manufacturers, Bishop commented that the increased incorporation of inspection has led to different rules for what data to store and for how long. One recommendation is to adjust data life based on the cycle of delivering products to customers. “If the cycle to the customer is 45 days, for example, you may want to keep the data for 60. Data life may also depend on the types of analysis tools that you use. Data trend images and Pareto charts may need raw data from 90 or 120 days. Of course, high-reliability and high-liability situations may still require keeping at least some data forever.”

Data from solder-joint inspection can help predict not only whether a board is good or bad the way test does, but whether a good board is likely to fail a few years down the road. A marginal solder joint may pass electrical test every time, but its shape may indicate a long-term problem in coping with heat or vibration or just normal wear. Process changes that result from analyzing the data using stricter criteria than mere pass/fail can help minimize the number of field failures, reducing warranty-support costs and increasing customer satisfaction.

The vast amount of data that an operation can generate during post-reflow inspection does not always need to be saved. Reducing the data “clutter” by setting priorities and determining the relevance of the data allows you to retain only trends, extremes, specific measurements, and sample runs. This approach will help you save the information you need for making effective and efficient process and product decisions.