The present study is a comparative evaluation of visual performance for the detection and recognition of trip hazards using the traditional roof bolter lighting as compared to the Saturn luminaire. The experimental design is based upon prior NIOSH research that evaluated trip object recognition given various mining machine lighting conditions.^11,12 A 2 × 6×3 (two light sources × six trip objects × three age groups) mixed-factorial within-participant design was used. Visual performance was quantified as the time to detect a trip hazard visual stimulus and the accuracy in recognising the location of the visual stimulus.

The study took place in the Human Performance Laboratory (HPL) at the NIOSH research facility in Bruceton, PA, with the laboratory simulating an underground coal mine environment. A Fletcher model HDDR roof bolting machine was placed within the HPL. The HDDR served as the test vehicle for the study. The roof bolter is painted orange and the reflectivity of the paint was measured at 25% percent.

Locations for the two phases of the study are shown in Figure 1. The interior work spaces were located in the middle walkway section of the machine and the front of the machine. Mine workers walk or crawl through the walkway section as they carry materials to the front of the machine.

A lighting survey indicated that the end of the walkway had the lowest levels of illumination. The front location is where the mine worker spends most of the time conducting the drilling and bolting tasks. Location 1 for Phase 1 is the walk-thru area. Location 2 for Phase 2 is in the area most used by the operator during bolting operations. The locations of the luminaires tested are also shown in Figure 1, with six CFLs being the existing luminaires located as they are on current bolters. A single Saturn luminaire was placed at each of the two locations to illuminate the working area. Participants remained stationary and standing at each location, at about 1.5m (5 ft.) away from the trip object apparatus (Figure 2).

The walk-thru roof bolter has a roof support mechanism that gets raised to reinforce the roof of a mine while the roof bolting operations take place. The height of the roof support affects the angle of the luminaire placed on the bolter and the illumination of the working area in Phase 2. The height was set at a 2.3 m, which is representative of the height of a mine roof, and was not varied during the study.

An electro-mechanical trip object apparatus (Figures 2 and 3) was used to randomly present the trip objects for the warmup session used to familiarise participants with the testing, and for the trip object detection and recognition testing of Phases 1 and 2. One apparatus was placed on the centre walk-through area floor and another on the floor near the roof support at the front of the roof bolter. Each apparatus was 216mm in height and contained six cylindrical trip objects that projected from the top surface of the apparatus. The trip objects were raised by an electro-mechanical solenoid at random intervals when the participant held down a computer mouse button. The trip objects were fully raised within 100 ms. Each trip object was 25.4mm long with an outer diameter of 3.3 cm and an inner diameter of 2.2 cm. The objects were painted a dark gray colour so that they would have a very low contrast and a reflectivity of about 6%, which is similar to an object (mine cable, pipe, or tool) coated with the coal dust and other material from the mine floor. A data acquisition computer was present behind the participant during all trials for manually entering data voiced by the participant. Headphones playing an audio file of sound from a roof bolter during operation were placed on a participant in order to isolate the participant from audio cues of the mechanical actuation of the trip objects and to create a more realistic environment.

The participants were federal employees. Three age categories were established: youngest 18–25 years; middle 40–50 years; oldest>50 years. The age group from 26 to 39 years was not included because there are generally minimal changes in vision for those ages.¹⁹ Thirty people participated, with 10 in each age group. The average ages were 23.3 years (St. Dev=1.77), 45.1 years (St. Dev=3.40), and 57.7 years St. Dev=3.30), for the age groups of youngest, middle, and oldest, respectively. The average age of all volunteers was 42 years. Age was an important factor to consider for mining given the median coal miner age of 44.7 years.²⁰ There were no exclusions based on sex, race, or ethnicity. Only the participants that passed vision tests for distance visual acuity, contrast sensitivity, and peripheral vision were accepted for the study. The visual acuity and peripheral vision tests were conducted using the Titmus V4 vision screener and contrast sensitivity tests were conducted using the Mars Letter Contrast Sensitivity charts. Participants were required to have: normal or corrected vision with an acuity of 20/40 or better; contrast sensitivity values of 1.60 to 1.92 for participants ≤60 years old and 1.52 to 1.76 for participants >60 years old; peripheral vision of at least 80 degrees for each eye. Participants that had self-reported radial keratotomy, monocular vision, glaucoma, or macular degeneration were excluded. Miners were not recruited because of potential expectancy biases that could confound empirical data. Miners could immediately determine that some of the lighting conditions were very different; thus, a negative bias could exist because the lighting is not what they are accustomed to, or a positive bias could exist if the person perceives something new as better.

The protocol was approved by the NIOSH Institutional Review Board. Informed consent was obtained from all participants in the study. Participants could withdraw from the study at any time. All participants completed all parts of the study.

The existing roof bolter luminaire uses a compact fluorescent (CFL) light source that is shrouded by an amber polycarbonate globe to protect it. The existing machine lighting was at 100% light output.

The Saturn luminaire uses an array of 12 cool-white LEDs with a secondary optic to provide a type III light distribution intended for luminaires mounted at or near the side of medium-width roadways. The light output of the Saturn luminaire was dimmed to 75% and 50% of maximum to enable the study of how the light output affects trip hazard detection. This dimming was achieved by placing neutral density filters onto the Saturn luminaire bezel. Table 1 and Figure 4 define the lighting parameters.

Figure 4 depicts the SPDs where the CFL has a predominance of medium and long-wavelengths in contrast to the Saturn LED light that has more short-wavelengths. This is an important distinction because in low light mesopic conditions, the eye rod photoreceptors dominate relative to the cone photoreceptors. The rod photoreceptors are more sensitive to the short wavelengths while the cone photoreceptors are more sensitive to the medium and long wavelengths. The relative stimulation of the rods is quantified by the S/P ratio where a higher S/P ratio indicates greater stimulation of the rods with respect to the cones, so obstacle detection in mesopic conditions improves as the S/P ratio increases.²¹

Figure 5 shows CFL and Saturn luminaires. Figure 6 shows the light distribution patterns of these luminaires. Table 2 lists the average luminance for the trip objects associated with Phases 1 and 2. The luminance was measured at the top, middle, and bottom surface of each trip object in the up position and measured with a Konica Minolta LS-100 luminance meter. The measurements were taken once the light output stabilised, which took about 30 minutes, when viewed from where the participant would stand for Phases 1 and 2.

As Table 2 shows, there was a drastic difference between Phase I and Phase II in the average luminances for the CFL luminaires. This was due to the placement of the CFL luminaires with respect to the location of the trip objects. The CFL luminaires of Phase 1 were placed much farther away and only provided indirect illumination of the trip objects.

Data were analyzed through a mixed linear model using Proc Mixed SAS version 9.4. Participant and trip object were treated as random effects, and the remaining independent variables of light, age, and object were treated as fixed effects. Trip object was treated as a random effect because object locations were varied to control for the learning effect that would occur if the objects were always presented in the same location. The researchers did not have a primary interest in the effect of a limited set of object locations on detection time.

Data screening revealed two data points which represented unreasonably fast detection times (53 ms in Phase 1 and 61 ms in Phase 2); these data points were excluded from the analysis. Examination of graphs and descriptive statistics showed that the detection time data were positively skewed; therefore, prior to running the mixed model, a log transformation was applied to better meet the assumption of normality. Subsequently, back translation was applied so that means and confidence intervals could be reported using the original metric. Separate analyses for each phase were conducted given that the overall lighting conditions differ significantly. Post hoc contrasts were also conducted using SAS. The Bonferroni correction was applied.

The study began with a warmup session that was intended to familiarise the participants with the test setup. To obtain individual reaction times, the participants completed a baseline test of trip object detection trials under well-lit laboratory illumination conditions. Participants then underwent a 20-min dark adaptation time to adapt to the reduced illumination afforded by the roof bolter lighting.

The study was conducted in two phases that were counterbalanced where 50% of the participants began Phase 1, took a short break, then began Phase 2; for the other 50% of the participants, this order was reversed.

The presentation of the objects was randomised and controlled by a micro-controller. The participants pressed and held down a mouse switch when they were ready to begin the test. Next, at a random time from 0.5 to 2 s, one trip target was presented for 1 s. Participants released the mouse switch when they saw an object and the datum recorded was the time from presentation of the object to the release of the mouse switch. Next, participants verbalised the location of the trip object (locations 1–6). The responses were recorded along with the actual location of the trip object. This datum was used for measuring the accuracy of detection. For a given trial, the six trip objects appeared 24 times and a null condition appeared six times. The presentation order was shuffled using the Fisher–Yates shuffle. There were 30 trials each for the baseline test and each of the four lighting conditions, resulting in a total of 150 trials.

Four seconds was the maximum time that the switch could be depressed. A 4-s timeout was recorded as a missed object and a value of −1 was assigned for the detection time. A value of zero was assigned for a false positive detection such as when the participant released the mouse button before a trip object appeared.

Data collection was accomplished by a combination of automated and manual methods. The automated method used a PC-based data acquisition system to acquire data and then store the data in a spreadsheet-compatible file. The automated data were for trip object detection and detection time. The manually collected data were for participant identification of the trip object and were stored on the data acquisition PC.

Pair-wise comparisons of all lighting condition were conducted for Phase 1 and Phase 2. The Bonferroni correction was used to control for inflated Type I error rate. Because there were six pairwise comparisons for each phase, the significance level for each comparison was set at 0.008 (0.05/6). In both phases all comparisons were statistically significant at the corrected level except for the comparison between the Saturn 75% and the Saturn 50% conditions. Differences among the Saturn luminaires were much smaller than differences between the CFL and the Saturn luminaires. The largest difference among the Saturn luminaires for both Phase 1 and Phase 2 was between the Saturn 50% and 100%, with the Saturn 100% having the faster average detection time.

Post hoc contrasts of luminaires were also conducted within each age group. Because there were 18 contrasts for each phase (six contrasts by three age groups), the significance level for each comparison was set at 0.0028 (0.05/18). In Phase 1, differences between the CFL luminaires and each of the three Saturn luminaires were significant (at the corrected level) in every age group. The results indicate that the detection times when using any of the Saturn luminaires were significantly faster compared to the CFL luminaire in all age groups. The difference between the Saturn 100% and the Saturn 50% was significant for both the young and middle age groups, and the difference between the Saturn 100% and the Saturn 75% was significant for the young group.

In Phase 2 differences between the CFL luminaire and each of the three Saturn luminaires were significant for each age group with one exception. In the young group, the difference between the CFL luminaires and the Saturn 50% condition was not significant at the corrected level. While none of the differences among the three Saturn luminaires were significant at the corrected level in the middle and old groups, in the young group the Saturn 50% was significantly different from both the Saturn 75% and the Saturn 100% conditions, with the slowest detection time being for the Saturn 50% condition.

Signal detection theory (SDT) can indicate decision quality under conditions of uncertainty.⁸ SDT defines four categories: a ‘hit’ is correctly identifying that a trip obstacle was present; a ‘miss’ is a failure to identify a trip object when it is present; a ‘false positive’ is identifying a trip object when none was present (null condition); a ‘correct rejection’ is identifying the null condition. The sensitivity index (d’) statistic, shown in equation (1), is commonly used in SDT to quantify the detectability of a signal (trip object) that is present or not present.²² Detectability increases as d’ increases, while a d’ near zero indicates chance detection (no detectability). (1)d′=Z(hit rate)−Z(false positive rate) Where Z is the Z-transform.

Table 3 depicts the miss rates, false positive rates, and d’ values for the age groups and lighting conditions. There were no significant differences (maximum difference of 0.4%) in miss rates among the Saturn lighting conditions, so only the 100% Saturn luminaire values are listed. All values of d’ were considerably greater than zero, indicating that the results are not by chance.

For Phases 1 and 2, all light output levels of the Saturn luminaire resulted in more rapid detection time, regardless of age group as compared to the CFL luminaire that is the current standard for roof bolting machines. We infer from the results that the Saturn luminaires enabled better detection times because of the higher S/P ratio, higher CCT, and the greater object luminance produced by the Saturn luminaire. In general, there is an improvement in detection when luminance increases and when the S/Ps of the light source increases.^14,21,23 The lowest Saturn luminaire S/P ratio was 1.67 compared to 0.61 for the CFL. Median object luminance was only 0.019 cd/m² for the CFL luminaire compared to 0.216 cd/m² for the Saturn luminaire at 50% light output of Phase 1. The Phase 2 median object luminance afforded by the CFL and Saturn luminaires was similar; for instance, there was 0.627 cd/m² and 0.597 cd/m² for the CFL luminaire and the Saturn luminaire at 50% light output, respectively. The Saturn luminaire at 50% light output resulted in an 8% improvement in average detection time compared to the CFL luminaire even though the CFL luminaire provided slightly higher mean luminance. The 8% improvement with the Saturn luminaire at 50% light output is likely due to the much higher S/P ratio.

Post hoc tests of the Saturn luminaires indicated that the Saturn luminaire at 100% light output provided the best mean detection time which was likely due to it affording the highest median object luminance and a much higher S/P ratio than the CFL luminaire. However, the differences were very small between the Saturn luminaire at 100% and 50% light output – a 30 ms reduction for Phase 1 and a 26 ms reduction for Phase 2. It does not appear there is a significant practical advantage for using the Saturn luminaire at 75% or 100% light output based on mean detection time.

Overall the results are somewhat similar to a prior study of mining machine lighting that compared the use of LED lighting as an auxiliary source for the existing incandescent and fluorescent machine lighting. The addition of an LED light source resulted in mean detection times that were significantly faster by approximately one second than mean detection times achieved without the LEDs.¹⁰ The mean object luminance was much greater with the auxiliary LED sources (0.18cd/m² versus 0.052 cd/m²) thus accounting for the faster mean detection times. It is difficult to make direct comparisons to this prior study because the mean object luminance was very different and because the S/P values were not provided in the prior study. However, both studies indicate that increasing object luminance can improve object detection.

Miss rates are of prime importance for safety because workers cannot avoid a trip hazard if they do not see it. The results for miss rates (Table 3) provide compelling evidence for the benefits of all the Saturn lighting conditions of Phase 1, where the miss rates were <0.5% for all age groups using Saturn luminaires in contrast to the CFL luminaires, which ranged between 32.5% for the youngest group and 50.4% for the oldest group. There were no statistically significant differences (maximum difference of 0.4%) in miss rates among the Saturn lighting conditions, so miss rates do not appear to be a primary consideration for selecting the Saturn light output. The Phase 1 miss rate results are likely due to the increased object luminance provided by the Saturn luminaires, which had an average object luminance up to more than six times that provided by the CFL luminaires. The Phase 2 miss rate results were less dramatic and seem to support the view that the luminances provided by either the Saturn or CFL luminaires was sufficient given the maximum miss rates of 0.4% and 1.5%, respectively.

Overall, age does not appear to be a major factor given it was not statistically significant for Phase 1 (F(2,2509)=2.59, p=0.076 and it had the smallest effect (F(2,2779)=4.90), p<.0.01) for Phase 2. However, the Phase 1 two-way interaction of age and light is of interest given this interaction was statistically significant. The results for only the Saturn luminaires follow a trend of increasing average detection time with increasing age and decreasing light output, which would be expected because generally detection time improves with increased luminance^14,21,23 and decreasing age.^10–12 The amount of change in average detection time for Phase 1 is very small (about a 6% difference) among the Saturn luminaires and three age groups. In general the same average detection time trend for the CFL luminaire is apparent with the exception that the mean detection time for the 40–50-year age group was higher than that of the >50 age group where mean detection time was 888 ms compared to the oldest age group mean detection time of 833 ms. The eye test results for these age groups were compared and no significant differences were apparent, thus an explanation eludes us at this point. The significant Phase 2 two-way interaction of age and light indicates that, over all lighting conditions, the oldest age group had the worst average detection time and the youngest age group had the overall best average detection. The amount of change in average detection time for Phase 2 is very small (about a 5% difference) among the Saturn luminaires at different light outputs and three age groups which is similar to the Phase 1 results. Therefore, there does not appear to be a compelling justification to use the increased light output of the Saturn 75% and 100% given the interaction with age.

A luminance of 0.21 cd/m² is required for the areas about the perimeter of the roof bolter machine. This luminance also appears to be applicable for the interior working areas of the machine. The CFL luminaire median luminance for Phase 1 was only 0.019 cd/m² compared to 0.216 cd/m² afforded by the Saturn luminaire at 50% light output, which enabled a 45% improvement in average detection time and a miss rate of 0.4% compared to the 50.4% miss rate of the CFL luminaire for the oldest age group. We note that the 0.216 cd/m² enabled by the Saturn 50% is slightly higher than the Federal requirement but we believe this 2.8% luminance increase is likely insignificant. Median luminance values that exceeded 0.21 cd/m² resulted in modest improvements in average detection time and miss rates for the Saturn luminaire at 100% and 75% light output; thus it appears that exceeding the Federal luminance requirement would yield few benefits.

The testing was conducted in a controlled laboratory environment that closely emulated an underground coal mine, but it did not include factors such as airborne dust and other environmental factors that might affect visual performance. Also, the participants were stationary and the targets popped up; hence the movement provided a visual cue that would not exist for the detection of static trip objects. Therefore, it is unknown how well the results can be generalised to the detection of static trip objects. Further, the cognitive demands on the test participants were relatively low compared to those of a miner working with the roof bolting machine in a confined, noisy environment. Thus, the significance of the working cognitive demand is unknown, but it is expected that the mean detection times and object miss rates would be higher.

Second, the research focused on the interior working spaces given one of the research questions to be addressed concerned the adequacy of the Federal requirement of a luminance of 0.06 fL (0.21 cd/m²) with respect to interior working areas of a roof bolter. The research was very limited concerning luminances lower than the Federal requirement given that only the CFL luminaire for Phase 1 had a median luminance below the Federal requirement. Additionally, lighting for the exterior working spaces was not addressed because the Saturn luminaire was specifically designed for the interior work areas. The Saturn luminaire cannot replace the CFL luminaires used on the sides of the roof bolter because it will not provide the required luminance on the mine walls.

Next, positioning of the luminaires is likely a factor that affected the mean detection time and miss rates. Different positions would provide different luminances, contrasts and shadow patterns all of which are important in detecting trip objects. Varying the position of the luminaires was outside of the scope of this paper.

Lastly, a detailed evaluation of glare for each lighting condition was outside the scope of this paper. Glare is an important factor that needs to be taken into consideration and will be addressed in a future paper.