Better Than Hit or Miss

Testing and training improve visual inspection of casting surfaces.

F. Peters, R. Stone, K. Watts, P. Zhong and A. Clemons, Iowa State University, Ames, Iowa

(Click here to see the story as it appears in March's Modern Casting.)

Visual inspection is a critical step in metalcasting, and many factors affect human casting inspectors. While cognitive ability can be identified through a simple test and rastering training improves the percentage of a casting surface that is inspected, discrimination between acceptable and unacceptable surfaces continues to be problematic. Undetected surface defects result in additional rework cycles, or worse, returns from the customer. Marking acceptable anomalies as defects results in excessive processing. Operators must be provided with appropriate training and inspection aids to help reduce measurement errors.

Known Factors Affecting Visual Inspection

Environmental factors:  The work environment can play a major role in the outcome of visual inspection tasks. Before assumptions can be made on individual or task factors, first it must be determined if the environment itself represents a confounding variable. Temperature is an obvious but often overlooked factor in visual inspection environments. Temperature and humidity together affect the ability to achieve optimal thermal exchange with the environment: too hot and the body will focus on cooling; too cold and it must focus on conserving heat. Both of these affect a person’s ability to achieve optimal cognitive performance.

It has been found that cooling sensations (e.g. from a fan) activate the brain and excite the nervous system controlling thermoregulation. This activation of the sympathetic nervous system elevates mental alertness, increasing general attention capabilities. Ideally, use work stations that have localized cooling sources to keep an environment near 70F (21C) with a relative humidity between 60% and 70%.

Operator factors:  Some of these include visual acuity, color vision and depth perception. As an eye becomes less accurate, so will its ability to accurately interpret defects. Experience on the job can be a very positive individual factor. In fact, inspectors who are experienced in performing a specific visual inspection task have been found to concentrate longer and perform with a higher degree of accuracy than those with less experience. Most of these benefits, however, were found in search tasks that have a controlled amount of variation, unlike typical casting surface inspection tasks.

General intellectual aptitude is another known factor in determining the capabilities of a visual inspector. As one would expect, higher intellectual aptitude correlates positively with higher levels of visual inspection success. This appears to be due to an increased ability to adapt strategies and utilize memory techniques that aid in the visual inspection process. The cognitive inspection style of individuals also appears to be a factor in visual inspection capability. Specifically, the Matching Familiar Figures Test (MFFT) has been shown to have potential in determining individual potential in performing visual inspection tasks.

Task factors:  The task itself can greatly affect an operator’s ability to perform a visual inspection. Depending on the kind of visual inspection being performed, posture and physical strain can impact visual inspection capabilities negatively. Several authors have shown select visual inspection jobs suffer from decline in visual inspection performance as a result of physical strain. As time on task increases, participants tend to lean forward, change postures more frequently, report more discomfort and take more time performing visual inspection.

Another task factor is the level of documentation provided for on-the-job comparisons. It is much easier to perform a relative judgment than it is to make an absolute one. It can be very beneficial to provide aids, such as comparator plates that allow visual inspectors to compare a possible flaw to a known category of flaw. Several studies have shown visual inspectors that are given an effective comparison aid consistently outperform those forced to make absolute judgments alone. Therefore, simply giving the visual inspector a visual reference aid can greatly increase his or her basic capabilities.

Effect of Cognitive Style

Several studies have been proposed to classify and select inspectors for visual inspection. The MFFT, which asks the participant to choose from six figures the one that is the same as the one on top of the screen, is one evaluation tool. Subjects can be classified into four different cognitive styles according to their accuracy and response time: fast-accurates (shorter time, fewer errors); reflectives (longer time, fewer errors); impulsives (shorter times, more errors) and slow-inaccurates (longer times, more errors).

A study of a variety of people in the metalcasting field who were given the MFFT and a computerized visual inspection task found the average values reported in Table 1. While these three groups of people are not representative of the general population, the results provide insight on their potential to do the cognitive tasks needed for visual inspection, as well as the need to make them more effective. When the cognitive style was considered with respect to a visual search task (Fig. 1), there was a statistically significant difference in the speed of the task between the fast-accurate and both the reflective and impulsive groups. Interestingly, the fast-accurate group did not have significantly different time compared with the slow-inaccurate subjects. This is attributed to the slow-inaccurates speeding up to get done with the task. This conclusion is supported by the accuracy of decisionmaking among fast-accurates, which was superior to the other three groups.

In conclusion, the MFFT was shown to be useful in determining the cognitive ability of personnel and there was a correlation between cognitive style and performance on a visual search task. This test could be used as a screening tool for new operators, or to identify the need for additional training and inspection aids among existing and future operators. It should be noted that underlying cognitive ability cannot be improved through training, but the strategies in which a person inspects a part can be improved to compensate for lower abilities.

Field Study Results

Field studies were conducted at steel and iron foundries producing shell, investment, chemically bonded sand and green sand molds for castings from 1 to 10,000 lbs. (0.5 to 4,500 kg). An eye tracker was developed so studies of the search strategies being utilized during the inspection process could be determined (Fig. 2). This tool allows analysis of the search pattern as well as determining the percent of the casting surface that is being inspected. It is mounted on a face shield and goggles to provide eye protection. This is often used in the shop floor environment. The eye tracker has three cameras, two recording the world view in front of the operator and the other records the location of the pupil within the eye. Two cameras on the world view are necessary to collect data on operators that are inspecting larger castings. The video from the two world view cameras is stitched together, then the world view and the pupil view are compiled into a video of the world view with crosshairs to show where the user was looking at any given time. This compiled video can be analyzed, frame by frame, to determine the operator’s search patterns.

To understand the results of the eye tracker, the useful field of view (UFOV) needs to be defined. UFOV is the area where humans can see the most detail in their vision due to the greater concentrations of rods at that particular location. For the purpose of this study a reasonable value of 6 degrees is assumed. Assuming that the parts were 22 in. (56 cm) from the operator’s eye, the useful field was calculated to be 2.25 in. (5.7 cm) wide. Figure 3 shows an example of a casting with the search path line and the useful field of view overlaid on the part. For data collected on castings up to approximately 15 in. (38 cm), the eye track revealed that on average, 59% of the casting surface was not being viewed sufficiently enough for the inspector to make meaningful decisions. In addition, 71% of the time, the search pattern was not consistent with the one used for the immediate preceding casting of the same geometry.

Figure 4 shows the results from another field study at a metalcasting facility producing large castings (greater than 1,000 lbs. [450 kg]). The initial results from the eye tracker showed a random search pattern used to inspect the castings, which led to only 84% coverage of the casting surface. After some basic rastering training, where the operators were instructed to use a consistent left to right search pattern, the eye tracking results were again collected. This time, 96% of the casting surface was adequately covered.

Determining Methods to Improve Visual Inspection

The field studies showed operators were not using a consistent search pattern nor was there adequate coverage of the casting. To help alleviate these problems, laboratory studies were conducted on methods to improve the search pattern as well as to study the ability to discriminate between surfaces.

In the first experiment, three conditions were tested related to the training and inspection aids provided to the subjects. The first group was only given instructions on what surface indications for which they were to look, but was not instructed on how to inspect the castings (basic training group). The second group received the basic training as well as using a left to right search strategy called rastering (raster group), with instructions to raster the entire time. The third group was provided the same training as the raster group but was also provided green static lines projected onto the part with a projector (overlay group).

Once training had been completed, the participants were asked to inspect 90 parts from five differing casting types ranging from 2 to 55 lbs. (1 to 25 kg). The laboratory experiment was set up to closely resemble a casting visual inspection process. They would inspect the part for surface anomalies. Each anomaly that met the criteria was to be circled using a piece of chalk. Then, the part was placed back onto the conveyance system and the process repeated for each part.

To determine which training condition was significant from one another, a Tukey-Kramer HSD test was done. This showed that the raster training group was superior to the basic training group, and there was no statistically significant difference between the overlay group and either other group.

For false alarm rates, there was a significant difference between the training approaches. The basic training group was significantly different from both the overlay and the raster training group. The lack of search pattern training resulted in more false alarms.  But the analysis of variance showed that there was not a significant difference in hit rates among the different training groups. The raster training group was superior to the basic training group with regards to overall signal detection, but despite the overlay group’s raster training, they were not superior to the basic overlay group. The likely reason for the overlay group not performing superior to the others is that the overlay lines were distracting. They are salient and caused the inspectors’ eyes to be drawn to those areas and not to the surface areas to be inspected. This is a common effect. Another factor to consider was that the participants would have to utilize more mental effort to confirm line existence, therefore slowing the process. This time loss had to be compensated by reducing peripheral inspection activities. These conclusions were confirmed with the results from the eye tracker used on these subjects.

The raster training group viewed the highest percentage of the part and made the best decisions about what an indication was. The overlay group had the second highest values, and the basic training group had the lowest percentage of the part viewed. The participants would view only a certain percentage of the part before making a decision, which would generally be a bad one. However, it was found that both groups with rastering training did adopt a more optimal (rastering) search pattern as opposed to the basic training group, whose search patterns were more random. The systematic search pattern likely would have helped with both rastering groups, if not for the distracting effect of the overlay.

There are many considerations for improving the visual inspection process. The MFFT is effective in identifying the cognitive ability of the inspector and cognitive ability is correlated to performance on a visual inspection task. It can be used to help identify new inspectors and also to recognize the training needs for those of lower cognitive ability. Field studies showed operators are not seeing enough of the casting surface to make a useful judgment. Experiments that provide an overlay on the casting to encourage the operators to inspect the entire part proved to be distracting and therefore ineffective. Rastering training did prove to be effective and can be easily implemented to help train operators in a systematic and thorough search pattern. A person’s ability to discriminate between acceptable and unacceptable remains problematic and requires diligence to ensure that inspectors continue to make appropriate decisions.    

This article is based on a paper (13-1546) presented at the 2013 AFS Metalcasting Congress in St. Louis.