Common Problems
Although visual inspection is cost-effective, it does have several common problems. An inspector’s ability to replicate his or her inspection results is often poor. At best, an inspector can only find about 85% of all defects within a manufacturing lot. This is similar to the newspaper cartoon where one has to find ten things that are different between two pictures. Despite one’s best efforts, it is sometimes just not possible to see what is in front of our faces.

Multiple inspectors often disagree with each other’s inspection results. Interpretation of what is and what is not a defect is a common difference. Overzealous inspectors will tend to reject more product, while less experienced inspectors may overlook defects. This can cause inspector preferences to formulate within a manufacturing group. Production supervisors will prefer inspectors who give consistent results day to day to reduce criticism from upper management.

The misnomer of 100% visual inspection is pervasive, since this is really 100% rework. With the advent of zero defects, total quality management, and Six Sigma, it is common practice to espouse 100% visual inspection of the product, when really this inspection is designed to cull out rejects for the customers. Inspection should be used to measure the process so that it can be corrected when problems present themselves. 100% inspection is a method used by organizations to filter out excess process variability and subsequently product variability. Invariably, due to the efficiency of inspection, a certain percentage of defective products will leak through the system and get to customers.

With a poor inspection program in place, inspectors will tend to report the same level of defects. From the inspector’s perspective, processes do not vary so therefore defects neither do defects and there will be a propensity to report similar numbers of rejects. Supervisors will also perceive a change in the number of reported defects as inspector and not process driven, and they will subtly encourage biasing results to a consistent baseline. How many times has an increase in reject levels been addressed by the supervisor’s “retraining” the inspector? Conversely, for process changes that are highly visible to operators such as new equipment, automation, or a new supplier, they may report high defect levels just because they are unsure of the change and decide to look all the harder for rejects.

Develop Workmanship Standards
The first step in any operation using visual inspection is to establish workmanship standards. The establishment of clear workmanship standards improves the ability of different operators to get the same results. The best standards are real defective product or representations of real product. Photographs can be substituted, if they clearly represent the defect. Sources of standards can be a reject bin, customer complaints, or current production. Written descriptions of workmanship standards are usually inadequate.

Standards should be ranked from high to low for each defect. At a minimum, the standards should be fail, barely fail, barely pass, and pass. Common problems that occur for organizations implementing standards are the different perceived quality levels of different people. One person’s pass is another person’s marginally fail. Likewise, the best standards have only one defect associated with it.

Standards can only be useful if they are made available to workers. They must be located in an area that is easily accessible to the inspectors. Ideally, they would be next to him or her while performing the inspection so that they can directly compare their product to them. It is not appropriate for them to be in an engineer's cubicle across the plant. It is important to note that color standards will fade after exposure to light and will therefore need to be replaced occasionally.

Establish Dedicated Inspection Areas
Establishing dedicated inspection stations also improves the quality of a visual inspection method. Lighting is often critical to color determinations and other slight defects. A single station can have the lighting controlled so that if a light that simulates natural sunlight is required, it can be inspected under those conditions. Likewise, poorly lit areas will reduce the ability of the operator to perceive defects.

Dedicated areas will also force inspectors to focus on the inspection task. In modern process manufacturing, the same person who makes the product often will inspect it. This leads to the problem of inspecting on the fly or trying to operate the equipment while performing a critical visual inspection. It is much better for these operations for the manufacturer to leave his or her machine, go to an inspection area, and formally inspect the products.

A single location also gives the potential for computerized data collection and a natural home for workmanship standards. Much visual inspection is poorly documented if at all. By collecting accurate count data on a personal computer, inspectors can more easily see changes to the process in a real-time environment. For a multiple shift operation, these locations can become the central point of communication of quality and production issues on the floor. For management, computerized data collection can give everyone the ability to not only see real-time data, but easily access historical data for the investigation of customer complaints, troubleshooting, or benchmarking.

A defined location also permits the inspector to maintain his or her equipment in an environmentally controlled environment. Microscopes, lenses, magnifying glasses, or pin gauges are susceptible to damage, temperature effects, humidity changes, or dust. Inspectors can also more easily keep the equipment calibrated and maintained if they are not routinely moved about the shop floor.

Qualify Inspection Methods
Finally, inspection methods should be rigorously qualified. Qualification is an independent randomized study to determine if an inspection is truly accurate and repeatable. Bad product is randomly seeded into good product and provided to inspectors to determine if they can find the defective product. If they can all find the defective product, then the method has been validated or qualified.

To achieve maximum confidence in the study, random sets of samples should be assembled to give a minimum 10^-6 probability of failure. For example, 3 operators may be asked to identify the single defective product in a lot of 10 twice (10^-1 x3x2). Other combinations are:

Inspectors	Trials	Defect Levels	Probability
Inspector #1	1	1 in 1000	10-6
Inspector #1	1	1 in 100	10-6
Inspector #2	2	1 in 100	10-8
Inspector #2	2	1 in 10	10-6
Inspector #3	3	1 in 5	10-7

 By forcing the same operators to check the same samples twice or even three times, it ensures that the inspection method is robust.

It is of critical importance during a qualification to randomize all of samples so that the identity is never known. A random number chart or computerized spreadsheet can generate lists of random numbers that should be used to label all product. The reason the statistics work out with such a high level of confidence is based on the assumption that all inspections are completely independent of one another. If an inspector can identify the products, then subsequent trials are meaningless. Even worse, the inspector may be able to identify the defective product based on the production date if the identity was not obscured completely invalidating the study.

The best qualification studies maximize variation wherever possible. Different people, days, machines, shifts, inspection equipment, etc. can all be used to maximize variability. A robust inspection method will not be influenced by any of these factors. A weak method will show problems. The off-shifts usually have less experienced people, older machines tend to generate poorer quality product, and different inspection stations may have slightly different lighting arrangements.

Visual inspection is a potent method for detecting defective products. With the proper workmanship standards, dedicated inspection areas, and a good qualification, there is no reason why the data from a visual inspection cannot be used to release product or troubleshoot processes.