Various fire incidents recently occurred underground, and each fire had the potential to result in a major incident. No lives were lost during any of these fires, but some necessitated the activation of emergency and escape procedures.
During investigation it was found that none of these fire detection monitors registered conditions that would trigger fire alarms in the control rooms at the mines. All incidents of conveyor belt related fires were detected by the early observations of employees working in relative proximity, or passing by the locations where the fires originated.
As a result, a decision was made to investigate opportunities for improvement in the effectiveness of the fire detection systems currently deployed in the underground mining facilities of some collieries. A case study involving some underground collieries was done to determine the effectiveness of their fire detection systems and identify the aspects that could be improved.
These collieries mostly use telemetric continuous monitoring fire detection systems with carbon monoxide (CO) sensor technology, aimed at reacting to the concentration of CO released during a condition of combustion. It is known that when smoke containing high CO concentrations is generated, the material is already heated beyond the designed operating temperature and is approaching its flash point, leaving an extremely limited margin for reaction time before the open fire occurs.
In this case study, 29 fire incidents were considered and their grouping is tabled in Figure 1. Furthermore, conveyor belt operations contributed to 83% (24 of the 29) of all these incidents.


The Case for Change
From this investigation, it was clear that drastic improvement in the effectiveness of fire detection systems was required. Various aspects of the fire detection process were assessed for effectiveness. The objective of this initiative was to investigate the responses of existing toxic gas detection devices and/or new technology of fire and toxic gas detection devices that would trigger emergency actions at their soonest to prevent fires from reaching life threatening conditions. The focus was on conveyor belt trajectories and conveyor belt material, along with their payload.
The most critical aspects are briefly addressed in this report.
Constraint 1:
Single Detection Gas Dependency
The use of CO sensors did not only provide early detection, as per our expectation, but we also found that the material needs to be in an advanced state of combustion for alarm conditions in order for CO to be liberated. The need was to identify at least one additional toxic substance that would be released and detected while the material was being heated, before the combustion reached flame propagation. This substance would act in combination with the CO toxin as a fire indicator.
Early Warning Toxin Indicators
Fire detection trials were conducted at the Kloppersbos testing facility of the Council for Scientific and Industrial Research (CSIR) in South Africa for alternative early warning toxin indicators. The following chemicals that were detected during previous conveyor belt tests for toxicity in a laboratory, that are also listed as potentially lethal, were assessed:
HBr | Hydrogen Bromide |
HCL | Hydrogen Chloride |
HCN | Hydrogen Cyanide |
HF | Hydrogen Fluoride |
H2S | Hydrogen Sulphide |
N2O | Nitrous Oxide |
SO2 | Sulphur Dioxide |
COCI2 | Phosgene |
CO | Carbon Monoxide |
CO2 | Carbon Dioxide |
Fires and Explosions Test Facility


The CSIR-managed fires and explosions testing, training, research and development facility at Kloppersbos helps achieve safe working conditions for South Africans in industries where fires and explosions may occur. The facility provides a full-scale surface testing facility for the evaluation of underground explosion suppression systems, the flammability of conveyor belts, dust suppression systems for continuous miners, and similar investigations.
The unique location and size of the facility allows for the demonstration of large-scale coal dust/methane explosions, making it an ideal venue for basic and refresher training sessions for underground workers.
A schematic layout of the fire testing tunnel is presented in Image 1. The two air tunnels shown at the top of the layout are connected to a centrifugal fan providing thorough ventilation for the testing facility to resemble underground ventilation conditions.
Image 2 is a picture of the test facility where a 1.8 metre x 4.0 metre length of conveyor belt is set on fire. The performance and behaviour of the fire is monitored by multiple sensors placed on the other side of the fire and downstream from the observing team. The heat source underneath the conveyor belt consists of six burners fuelled by Propane to produce 3000kW energy.


This picture was taken two minutes after the burners were ignited. The violent flame gives the impression that the conveyor belt is burning; however, the flames are from the burners themselves and the conveyor belt is merely in its smouldering stage. This is the key point at which to detect any fires in their infancy before the flash point is reached. This smouldering stage is where the early warning system is expected to respond, and where so many systems have failed in the past to detect a smouldering conveyor belt. At this point, it is important to note that the CO levels only reached 18ppm max. The CO level would only exceed the 18ppm much later once the flash point was reached, which occurred approximately 12 minutes into the fire.
Research Findings
The graph in Figure 2 shows the results from a conveyor belt fire test. This graph illustrates what toxins are released by the belt in its smouldering stage before the flash point is reached and during full combustion. These findings assisted in evolving current fire detection systems, mostly reliant on CO concentrations, into an early warning system to detect combustion before fires started.


The graph was rescaled to 0.4-2 volt to show the relation between the gases detected as some sensor spans would over-shadow the other smaller gas ranges if kept in the ppm scale. For example, the carbon monoxide (CO) span was 500ppm compared to the phosgene (COCL2) span of only 1ppm.
It is important to understand the events that take place within the first 12 minutes, as they show that the CO sensor does not reach critical alarm point above 20ppm, yet a fire is raging on!
In a real-life scenario, relying on the CO sensor itself would have failed. There needs to be another deciding factor to determine the difference between a conveyor belt fire and, for example, a diesel vehicle parked near the sensor(s). The solution would be in the form of a combination of sensors that can be used to determine the difference, so that false alarms can be prevented and true alarms acted upon immediately within the shortest time possible.
The graph also shows that the SO2 sensor provided 11 minutes of early warning before combustion started. The CO concentration fluctuated in the same 11 minutes and consequently no CO alarm was reached. The short-term exposure limit (STEL) of 5ppm can be used as a guideline to trigger an alarm. The alarm would only then be deactivated if the SO2 reading reached ZERO ppm (if using a hysteresis triggered type setting).


The behaviour of some toxic gases during the early stages of the combustion process is illustrated in Figure 3. The scale on the left of the graph is reflecting 0.35 to 0.65 volts. Although this test lasted 34 minutes, only the first four minutes is discussed here.


It is demonstrated that SO2 in the smouldering stage peaked at 11ppm and dropped to a low of 3.5ppm before the flash point was reached. The belt was absorbing the energy before it released “flash point”. Both CO and SO2 are present, however, CO also dropped which would never have indicated an alarm. Now if both CO and SO2 levels are present, a conveyor belt smouldering/burning is evident.
Image 3 is a picture of a conveyor belt on fire. This is a real-life scenario where only four metres of conveyor belt is on fire. The brownish colour of the smoke indicates that this belt has flame retardant properties. The smoke rollback is indicative of the heat intensity where smoke is then found to pollute airways in a mine that would normally not be considered affected by the fire.
Alarm Settings for Gas Detection Sensors
Absolute Values
In the scenarios assessed during the case study involving the six mines mentioned before, the first and second alarm levels for CO concentration were set at 30ppm and 100ppm absolute value respectively. These alarm levels were selected to eliminate the influence that exhaust fumes from diesel powered vehicles would have when passing by or working in the vicinity of fire detection sensors. The findings from the fire tests, however, illustrated that fires can already be burning with life threatening results and not yet be flagged to control room operators for emergency response. Lowering the CO alarm levels could result in numerous alarms from vehicle exhaust fumes and that would lead
to complacency.
Adding the early warning capability of the SO2 as a fire trigger would vastly enhance the fire detection systems without having to lower the CO alarm levels.
Rate of Change
A further enhancement in the integrity of fire detection systems would be to consider the use of Rate of Change as an alarm condition. The “Rate of Change” (ROC) method for additional alarm conditions in this case study is based on a running window of 15-minute and 1-hour accumulated gas sample readings from the gas sensor. The accumulated value is an indication of the last 15 minutes and/or 1 hour that the total gas volume has passed the sensor.
The method is to accumulate the gas reading at a constant sample rate allowing us to average the reading to a 1-minute sample average.


Figure 4 illustrates how the ROC can vastly improve the early warning alarm conditions when configured in conjunction with the absolute value alarm settings, as discussed above. The dataset collected from the Scada system after a recent fire in an underground coal mine was used to illustrate this concept.
This calculation is based on a 15-minute interval, meaning once the first 15-minute accumulated result is determined, it will only be compared in 15-minute time intervals, therefore looking at the ROC for a 15-minute period instead of every single minute. When the sixteenth minute result is logged, the first minute result will be excluded from the value set.
Under the prevailing conditions of this fire incident, a >2% ROC would be recorded two hours before the fire reaches the flashpoint using a 15-minute interval, and to compliment the 15-minute setting, a 1-hour ROC set at >1% would have recorded about two hours and 30 minutes before the belt burst into flames. The real alarm was triggered at 30ppm, about two minutes after the flashpoint, and in the presence of 3.5m/s velocity in the belt road, the fire was out of control within five minutes.
Smoke or Particle Concentration Sensors
The use of particle sensors as smoke detectors was also tested. The effectiveness of this method for fire detection is illustrated in Figure 5.


This technology detects the presence of airborne particles and can work very well in the absence of dust, fumes and stone dust applications.
The ROC concept can also be programmed as an early warning alarm indicator. The technology was tried and tested through the years and no further research was done in this regard.
Linear and Thermal Fire Detection Systems
Both these methods of fire detection were proven through the years to be very successful at the point of application. They are not discussed in this report because their effectiveness is known, and no further research was done in this regard.
Constraint 2:
Sensor Location and Placement
Sensor location relative to the ventilation flow direction is paramount to the reaction of the sensor to the gas. Sensors must always be installed in locations where the general body of air would fully flow over the sensor.
Sensor placement is also critical to this effect. Placement refers to the actual position in terms of height, accessibility for maintenance and performance testing, and optimal coverage of the ventilation current. During fire detection testing, as discussed before, it was found that as little as a 1-metre sub-optimal placement can influence the gas reading dramatically and alarm conditions can be missed. Extreme high velocities and harsh environmental conditions must also be avoided because these conditions will affect the sensor lifespan.
Constraint 3:
Telemetric Technician Skills, and Equipment
Incident investigations after fires have occurred, normally focus mainly on the prevailing conditions that could have caused the fire, identifying what preventive controls have failed, and why the telemetric gas detection devices did not warn of an impending danger in progress. This section of the report will briefly discuss the technical matters of telemetric gas detection practices that are paramount to the effectiveness of fire detection systems.


The following matters must, as a minimum, be addressed through training modules to ensure that telemetric technicians are suitably skilled for the task of effectively operating fire detection systems:
- Understanding the design, assembly and operation of gas detectors and sensors.
- Be familiar with the various types of sensors and how to identify different materials to be monitored.
- Common application, limitations, interferences and poisoning of electrochemical sensors.
- General overview of fixed installation apparatus, operating instructions, operational checks, cabling, communication network and maintenance requirements.
- Workshop calibration kits and equipment, sensor testing and record keeping.
- Fixed gas detectors’ receiving, handling, storage and pre-installation checks.
- Performance verification through bump testing, e.g. in-situ T90 performance tests.
- Calibration gas characteristics composition, gas and flow regulator selection criteria, cannister handling, storage and expiry date management.
General Observations
The following is a summary of general observations from the conveyor belt fire testing:
- Conveyor belts do not just burst into flames at the slightest heating incident. It requires a fair duration (between 11 and 21 minutes) of exposure to heat in excess of 500°C for a stationary conveyor belt to start burning.
- Damage to conveyor belts, idlers and pulleys may result in the coal underneath the belts to burn. Coal burns easier than the conveyor belt material.
- Certain toxins are released during the heating activity before the belt material actually participates in the combustion process.
- All belts can burn. Flame-retardant belts, however, burn less material after the heat source is removed. Non flame-retardant belts can incinerate completely.
- The more intense the heat generated from the fire, the further the smoke would flow against the airstream in airways with lower prevailing velocities.
- Phosgene and sulphur dioxide release was severely aggravated when water was dispersed onto the belt.
- Location of detectors is critical to the effectiveness of the detection system.
Conclusion and Recommendations
This case study delivered extremely insightful information for improvement opportunities in the fire detection systems currently deployed in underground coal mines in South Africa.


The extent of this study was appreciated by all stakeholders that participated in this process.
Recommendations to improve the fire detection system effectiveness:
- Retain the current CO sensors. Introduce SO2 sensors together with the current CO sensors to enhance the early detection of heating incidences.
- Configure all CO and SO2 sensors to alarm with a rate of change according to site specific specifications.