Visual signage commonly supports safety evacuation in road tunnel emergencies but can be of little help if dense smoke is produced by a fire. This paper covers two full-scale experimental studies (involving clicks, whistles, bells) that investigated how acoustic signage can help to provide guiding messaging in low visibility environments to help lead evacuees safely from a tunnel. Subjective assessments by the participants revealed a higher preference for the tone-based sounds (bells, whistles) over clicking.
INTRODUCTION
Past emergency experiences in road tunnels have helped to establish evacuation systems supported by visual signage. But visual aids are of little assistance when visibility is poor, such as due to smoke from a fire, obstacles or travellers with impaired vision. Behavioural research indicates that human factors also play a role in evacuation procedures.
To help support better evacuation procedures, acoustic signals have been proposed and studied for some years5–9. In particular, directional sound technology has been tested. An advantage of audio signals is the universal comprehensibility of abstract sounds rather than speech in specific languages. Yet, people are inclined to seek visual cues to find emergency exits and so further research was undertaken how they might comprehend and follow deliberate guiding sounds.
Several studies have pointed out the effectiveness of audio signals, as instructions, greatly improved evacuation procedures in many situations, including non-road tunnels11-13. But some road tunnels do not always have emergency exit doorways for sound beacons to be placed and draw evacuees, which means people must be guided out of a tunnel.
Studies of acoustic signing have included frequency modulation to give the impression of an evacuation signal coming from a specific direction. But these have been found to be only effective around specific locations. Further, such technique would need training of evacuees to work sufficiently well.
US Tariffs are shifting - will you react or anticipate?
Don’t let policy changes catch you off guard. Stay proactive with real-time data and expert analysis.
By GlobalDataMoreover, the efficacy of sound signals is dependent on the type of sound used. One study suggests that complex tones with distinctive timbre and pitch are recognisable as an evacuation signal and are easier to distinguish against background noise14. But a class of sounds15 that generate strong repulsive responses could, potentially, direct people away from danger.
While visual exit signages have been well-researched, much remains unexplored for audio signals. For the present study, different acoustical techniques were tested on participants in experiments set up in test tunnels where low visibility conditions were created.
METHOD
The first experiment was performed in the tunnel of Ladehammer wastewater and sewage treatment plant in Trondheim, and the second in the Runehamar test tunnel in Åndalsnes, both in Norway.
EXPERIMENT 1: LADEHAMMER TUNNEL
The hypothesis (H1) for Experiment 1 was to investigate if sound signals in a smoky environment can help lead people in a specified direction to safely evacuate a road tunnel.
The experimental sessions occurred at the entrance of Ladehammer tunnel, which is 75m long and 6.5m wide, leading deeper into the mountain. The road is made of concrete, the roof and walls lined with sprayed concrete and are rough and irregular. The tests took place during one week in February 2020.
Stimuli
The setting and selection of stimuli for the experimental design were as follows:
1. Audio nodes: Five sound nodes were mounted on signposts at 20m intervals, 2.7m above the road surface, and approx. 0.30m from the walls which they set to face.
2. Fan noise: Ventilation fan noise was simulated using one speaker at each end of the 80m long section of test tunnel. Real recordings from fire fans in Oslofjord tunnel were used. (During fires, fans are not running in the first minutes after detection to reduce oxygen supplies.)
3. Sound signals: Norphonic and SINTEF evaluated different signals to use for the tests – clicking (based on a footstep sound, played on single speakers with a 500ms set time delay, progressing between speakers along the tunnel in the desired evacuation direction); and, tone-based whistling sound (which used a simulated Doppler effect to give the impression of a sound source moving off along the tunnel).
Data collection: objective and subjective tests
To test the hypothesis, the success criterion was whether the voluntary participants followed the sounds in the desired direction. The participants also completed questionnaires, including multiple-choice and open questions, and more general information on the experience and preferred sound signal was sought. Three 5-point Likert-typed scale questions were used to evaluate the perceived clarity, comprehensibility, and comfort of the sound signals.
The participants were recruited via the experimenters’ own network contacts, student and school parents’ organizations, and social media. The total sample size was 30 participants (14 female, 16 male), between 18 and 63 years old (M = 36.4, SD = 12.1).
Experimental setting and procedure
Upon arrival, participants were briefed, filled in pre-test questionnaires and then each undertook their sequence of test session, all without briefing about the audio signals.
To simulate poor vision, test participants were given ski goggles covered with duct tape. They began in a car parked in the tunnel, were led to the starting spot, spun around to be disoriented and then moved to different, random, locations. They were then instructed to start walking towards a safe exit, guided only by their interpretations of sounds. Helpers shadowed them for safety. When it was clear that participants had determined to follow a specific direction for their evacuation, they were guided back to the parked car for the final, subjective, evaluation.
This procedure was repeated with different scenarios. Five different scenarios were tested in each experimental session: one reference scenario without sound signal (but with low fan noise) and four scenarios with sound signals (two with high fan noise and two with low fan noise). The direction of the sound signals was randomised.
The reference scenario helped to investigate possible systematic errors, such as traces of fresh air flowing, slope of the road surface, light leakage, any surrounding faint noise from the treatment plant, etc. The low fan noise (approx. 65 dBA – weighted decibels) helped mask any sound from the treatment plants. Measurements at Oslofjord tunnel showed high fan noise reached up to 95 dBA but the test limited the simulated sound to 85 dB (decibels)
Upon completing the scenarios, the participants were asked to fill out a post-test questionnaire. Most experimental sessions lasted just under 30 min, sometimes well so.
EXPERIMENT 2: RUNEHAMAR TUNNEL
The hypothesis of Experiment 2 (H2) was whether prior briefing on following sound signals improves safe evacuation numbers.
The second experiment was carried out as a full-scale experiment in a real road tunnel – Runehamar tunnel, administrated by the Norwegian Public Roads Administration (NPRA, or Statens Vegvesen). Runehamar is about 5 km from Åndalsnes and is a two-way asphalted road tunnel, its ceiling and walls in rough blasted rock. The tunnel is approx. 1600m long, 9m wide and 6m high. It was retired more than 20 years ago when a new parallel tunnel was built for road traffic. The tube is now mostly used as test tunnel, e.g., for fire research. Experiment 2 took place during one week in August 2020.
Stimuli
The first sound node in the experimental area was located 200m into the tunnel, the entrance to which was covered with tarpaulin to keep out daylight. The setting and selection of stimuli for the experimental design were as follows:
a. Audio nodes: The same audio nodes as the first experiment were used, but this time ten nodes were used, mounted on the walls at 25m intervals and facing the rock. Since technical cabinets are located every 125m in Norwegian tunnels, it is considered that 25m will support a straightforward installation. The total length of the test section was, therefore, 225m.
b. Smoke machine: The local fire service operated the machine. Smoke density varied due to drafts and visibility ranged from 1m to 20 m but held thick enough.
c. Fan noise: With tarpaulin covering a tunnel entrance, fans could not be used. Their noise was simulated, as before. A pilot test gave fan noise of between 73 dBA and 77 dBA, which is lower than real fire fan noise. As before, for the experiment the simulation noise was limited to prevent discomfort for the participants. The fan noise masked surrounding other sounds.
d. Sound signals: clicking was used again and also a bell sound. The signals were played sequentially with a marked time delay of 1 second to mimic sound moving physically through the tunnel. The delay had to be more than in Experiment 1 considering the bell sound was 0.7 seconds long. The sound level of the bell was adjusted to be clearly audible over background noise. Subsequently, the clicking sound was also adjusted.
Data collection: objective and subjective tests
The success criterion for evaluating the efficacy of the sound signals was similar to the Ladehammer experiment.
Participants were local, recruited on a voluntary basis again, via newspaper advertisements and a local sports club. The sample size consisted of 33 participants (10 female, 23 male) aged between 23 and 78 years (M = 43, SD = 11.8). Only two reported having previously had evacuation training. Four participants (i.e., 12 % of the sample) declared having reduced hearing but none used any type of hearing aid.
Experimental setting and procedure
Similar to Experiment 1, the participants were received in a welcoming protocol and briefed, signed consent forms and filled in a pre-test questionnaire to gather demographic data. They were taken to a car parked inside the tunnel and then driven 250m with their eyes covered to avoid spatial recognition of tunnel features. The experiments were undertaken during the Covid-19 pandemic and, therefore, taped ski goggles were not used to avoid participants sharing items. Once the participants reached their start positions, the hypothetical scenario was presented verbally by the test administrator, inside the car, as follows:
“You have reached about the middle of a long tunnel when the traffic suddenly stops. You do not know why the traffic stops, and smoke rapidly surrounds the car. You are not sure how far it is to the nearest exit or emergency exit. The following evacuation message is played on the DAB network.”
The test administrator then played an evacuation message on a portable speaker in the car. Three scenarios were evaluated by the participants: one reference scenario without signal followed by one scenario for each signal (in randomised order).
The evacuation messages were different for the reference scenario and the sound signal scenarios and were played twice before each scenario.
- Reference scenario: “Fire in the tunnel. Evacuate immediately.”
- Sound signal scenarios: “Fire in the tunnel. Evacuate immediately. Follow the sound.”
After hearing the evacuation message, participants were asked to step outside the car, close the door, assess the situation. The direction of the sound (and thereby hypothetical tunnel exit) was randomised but balanced in number so that suggested movement inwards or outwards in the tunnel had about the same distribution. After assessing the sounds, the participants went back inside the car and explained their choice of direction to take. The procedure was then repeated for the different scenarios.
There was no time limit for the participants to decide which direction to take in each scenario, but they often decided within 1-2 mins. Each experimental session, including transport in and out of the tunnel, lasted less than 30mins.
Upon finishing, participants fill out a post-test questionnaire.
Statistical analysis
The statistical analysis was performed using the Statistics and Machine Learning Toolbox in MATLAB16.
It was assumed that the reference case, with no sound signal, had a random behaviour with 50% chance of leading to evacuation in either direction in the tunnel. This was supported by the results from Experiment 1 where 17 went in one direction and 13 in the other. In Experiment 2, 26 went outwards, towards the entrance direction, and seven went farther into the tunnel. This difference was significant (binomial test using z-test approximation, z = 3.31, p < 0.001) and must be considered when performing the statistical analysis on the sound signals. In Experiment 1 a binomial test using z-test approximation was used to compare the sound signal results with a 50% random choice, while in Experiment 2 a chi-squared test was used to compare the sound signal results with the observed reference.
A Fisher’s exact test was used to compare the different sound signal cases, both in Experiment 1 and 2. A logistic regression was done for other variables (e.g., age, sex, hearing impairment, and direction of the sound).
RESULTS
Experiment 1
Fan noise level did not have any significant impact on the outcome of the test. Distribution was identical for the clicking sound, while for the whistling sound a slightly larger proportion chose the correct direction, toward evacuation, when the fan noise was high, though this difference was not statistically significant. Eighteen of the participants stated that they could not hear the sound signal quite as well on each test, but this is not reflected in the results.
Since the fan noise level did not have any impact on the results, these tests were combined and the analysis found that 42 of 60 experiments went in the correct direction. This is significantly different from a random 50% distribution (z = 2.582, p = 0.005) and supports the first hypothesis, H1 – that sound signals in a smoky environment do lead people in a specified direction and so help in tunnel evacuation.
In the post-test questionnaire, participants were asked if the sound helped. Among those who answered positively, the success rate was 75% (compared to 67% overall). The difference between the sound signals was not statistically significant (Fischer’s exact test, p = 0.81).
A logistic regression analysis showed that none of the other variables had any significant effect on the results.
Experiment 2
Most (91%) of participants chose the correct way out with the bell sound. The comparable share for the clicking sound was 85%. A chi-squared test confirms that this is statistically significant (p < 0.001) for both sounds. These results also support H1.
In addition, statistical analyses were performed to test the second hypothesis, H2, which consider that providing a message to follow sound signals does increase the number of participants going the right way. Forty out of 60 participants went in the right direction in Experiment 1, and 28 out of 33 went right in Experiment 2. Using a one-sided Fisher’s exact test, this is a significant difference (p = 0.047), suggesting that prior information is important.
Furthermore, the subjective responses from the questionnaires revealed a slightly higher preference for the bell sound in terms of clarity and comprehensibility. Six participants found the clicking too low/difficult to hear while three said so of the bell sound. Twenty-one participants (64%) preferred the bell sound as opposed to only four (12%) preferring the clicking sound. Eight participants (24%) offered no preference. The difference is statistically significant.
DISCUSSION
In Experiment 1, the clicking sound was a recording of a footstep while the whistling sound simulated a doppler effect, making the acoustic effects completely different. Several of the participants preferred the latter as it was perceived to be ‘alarm-like’ and clearly heard.
A challenge with the doppler sound was that it required accurate synchronisation of the loudspeakers, and the psychoacoustic effect could vanish if some nodes did not play its sound. The doppler sound was also initially constructed to work for segments of 50m-70m of the tunnel test zone. Since this method did not outperform the clicking sound objectively, it was rejected for Experiment 2. Instead, a more audible sound (bell) was employed.
Although the test set-ups differed, the statistical comparison between Experiment 1 and 2 indicates that providing prior specific instruction about following the sound does significantly increase the success rate. This supports the hypothesis from Experiment 2.
In Experiment 2 the participants were not blindfolded, they did not know how long the tunnel was, and for the reference case they were only told to exit the tunnel. Because of the smoke filling the tunnel and the tarpaulin covering the entrance, the participants could not sense in what could be the best way to exit. A possible explanation of why most went outwards (back the way they came in) is that it feels safer to go back the way you have entered. If there had been a car crash or a fire, it seems reasonable to think it happened in part of the tunnel not yet travelled through. This is supported from other studies3.
The subjective evaluations from participants in Experiment 2 helped in results comparison, including a comment that the clicking sound could be confused with other sounds in the tunnel. A more distinct sound would be wise and optimised with adjusted sound level, frequency content or time delays between sound nodes. Since about 90% of the participants went the right way, it is still not certain that the potential for improvement is particularly great.
With fan noise level used during both experiments less than actual noise from those in fire ventilation mode, no participant considered the sounds to be too high, suggesting additional research could explore increasing the level.
Other sounds may influence the understanding of evacuation sound messaging in case of a real fire. Evacuees from the Gudvanga tunnel fire, in 2013, explained that they could hear explosions (probably from the tires of the burning truck), cars crashing into other cars and walls, and people panicking17.
A limitation in the study was that the sound levels from the nodes were not measured, only subjectively assessed as to be ‘clearly audible’ and comparable in loudness. There may have been differences in how well the sounds were perceived. Some participants said the click-based sound was more difficult to hear. However, this limitation does not diminish the relevance of the findings exposed.
Another challenge, not considered in this study, was the effect of traffic jams and sound shielding or blocking, and echoes, during evacuation. Especially large trucks could give shadowing effects and thus lead to confusion. This effect should be examined in future studies.
An earlier study18 indicated the importance of simple and clear instructions in the evacuation message – inform people what is happening and then instruct them what to do. The evacuation messaging played during Experiment 2 was based on DAB messages that NPRA currently uses, followed by a short instruction to follow the provided sound. While it remains uncertain whether this is the optimal message to give, considering that 90% of the participants chose the right direction it is uncertain what potential there is to improve the message.
Language choice is probably a more important factor. All participants in these experiments in Norway were locals and spoke Norwegian, and the message used was in Norwegian. Considering that people of different nationalities can use public road tunnels, the use of different languages and their effect on evacuation should be investigated further.
Other factors that can also come into play are the use of female/male voice, the tone of voice and speed of speech in the audio messages.
Moreover, when as many as nine out of ten people go the right way, there is a great chance that you can get a ‘sheep flock/herd behaviour’ effect in a busy road tunnel11. In an early phase of the fire scenario, when the evacuees are not surrounded by smoke, the few people who go the wrong way should see others go the opposite way and might change direction.
CONCLUSIONS
The results of both experiments suggest that guidance sound signals can be used to lead evacuees in a desired direction for safe evacuation of a road tunnel when visibility is poor. Messaging to follow sound signals means more people go the right way.
The difference between the sound types was not statistically significant. However, indicated a clear preference for the bell and whistling sounds over the clicking sound signal. Validation studies are required along with further research under real conditions.
ACKNOWLEDGEMENT
The project was initiated by Trafsys AS and funded by NPRA (Statens Vegvesen) through an innovation programme from Innovation Norway. Working along with SINTEF on the studies, Norphonic was the technology designer and equipment vendor to Trafsys AS within the research project.
